Free ChatGPT Templates: 200 Ready to Copy (By Use Case)

You are a working writer, not a content marketer. Plain sentences, no throat-clearing.

Here are my raw notes on [topic]. They are messy and out of order.
[paste notes]

Write a first draft of [a blog post / a memo / an update], [target length] words,
for [who reads it and what they already know].

Rules
- Use only what is in my notes. If something is missing, mark it [GAP: what's needed].
- Keep my phrasing wherever it already works. Do not upgrade my words.
- No introduction that explains what the piece will cover. Start with the point.

Cut this to [target word count] words. That is roughly a [X]% reduction.
[paste draft]

Cut in this order, and stop as soon as you hit the target:
1. Sentences that restate the previous sentence
2. Qualifiers and hedges that change nothing
3. Examples beyond the first one
4. Adverbs
5. Whole paragraphs, if steps 1 to 4 were not enough

Do not rewrite for style and do not add anything.
Return the cut version, then a one-line note saying which step got you to target.

Below are two samples of my writing, then a paragraph written by someone else.
Rewrite the third paragraph so it reads as though I wrote it.

SAMPLE A (mine)
[paste 100 to 150 words you wrote]

SAMPLE B (mine)
[paste another 100 to 150 words you wrote]

REWRITE THIS
[paste the paragraph]

Match sentence length, punctuation habits, and how much I hedge.
Do not match the topic. Then list the three habits you copied, so I can tell you
whether you read me correctly.

Do not write any prose. Produce an outline only.

Piece: [what it is] for [audience]. Target length [words].
The one thing a reader must leave with: [the single takeaway]

Give me:
- A working title and two alternatives
- Five to eight section headings in order
- Under each, one line stating what that section proves and what evidence it needs
- A list of anything I will have to go and find before I can write it

If the takeaway cannot support the target length, say so and tell me what is missing.

Line-edit the passage below. Return a table, not a rewritten passage.

Columns: Original sentence | Suggested edit | Reason (five words max)

Only include sentences you would actually change. Skip the ones that are fine.
Reasons must be concrete: unclear referent, buried verb, two ideas in one sentence.
Do not use the reason "flow" or "readability".

[paste passage]

You are a sceptical [role most likely to push back: reviewer, regulator, rival, editor]
reading this for the first time. You are not hostile, but you are not generous either.

[paste argument]

Give me, in order:
1. The weakest claim, quoted, and why it is the weakest
2. The strongest objection someone in your position would raise
3. Any place I have assumed something the reader will not grant me
4. The one piece of evidence that would make you change your mind

Do not suggest fixes. I want the diagnosis, not the prescription.

Write three different opening paragraphs for a piece about [topic],
aimed at [audience].

1. Blunt: states the conclusion in the first sentence
2. Concrete: opens on a specific scene, person or number
3. Contrarian: opens by disagreeing with what the reader probably believes

Each 60 to 90 words. No question as the first sentence in any of them.
Then say which you would choose for this audience and why, in one sentence.

Rewrite this so a [reading level or audience: a competent 14-year-old, a busy
executive with no domain background, a non-native English speaker] can follow it
on one read.

[paste text]

Rules
- Keep every technical term that carries real meaning. Define it inline on first use.
- Delete every technical term used only for tone.
- Maximum [20] words per sentence.
- Do not add a summary at the start or the end.

Afterwards, list any term you kept and the six-word definition you used.

This draft was partly AI-generated and reads like it. Fix that.
[paste draft]

Remove or rewrite:
- Openings of the form "In the world of" or "When it comes to"
- Tricolons used for rhythm rather than meaning
- Balanced both-sides sentences that commit to nothing
- Sentences that begin by naming what the sentence is about to do
- Closing paragraphs that summarise what I just read

Keep the argument identical. Return the edited draft, then list what you removed.

Give me eight headline-and-subhead pairs for [piece], read by [audience]
in [where they see it: a search result, an inbox, a feed].

Constraints: headline [under 60] characters, subhead [under 120].
Two of the eight must be plain and descriptive with no persuasion at all.

For each pair, add the angle in two words. Then rank all eight for this audience
and say what the top one risks.

I want to write about [topic] but I have not thought it through.

Do not write anything yet. Ask me one question at a time, up to [eight] questions,
and wait for each answer before the next. Aim your questions at what I know that
nobody else does, not at background you could look up.

When you have enough, say so, and give me back a one-paragraph thesis in my words.
Only then, if I ask, write the draft.

Source material:
[paste the finished piece]

Convert it into each of the following, using only what is in the source:
- A [150] word summary for [where]
- Five short social posts, each making one point, no hashtags
- An email to [audience] with a single call to action
- A slide outline: [six] slides, title plus three bullets each
- A two-sentence version for someone who will read nothing else

Where the source does not support a format, say so instead of padding it.

Read this draft as a fact-checker, not an editor.
[paste draft]

Return a numbered list of every factual claim, and for each one:
- The claim, quoted
- Type: verifiable / opinion / prediction / unclear
- If verifiable: what kind of source would settle it
- Risk if wrong: low / medium / high

Do not verify anything yourself and do not guess whether a claim is true.
List claims only. I will do the checking.

Final read of this before it goes to [recipient or platform].
[paste final draft]

Check only these, in order, and report only what fails:
1. Does the first sentence work if it is the only one read?
2. Is there a promise made early that is not kept later?
3. Any sentence I would be embarrassed to have quoted back at me?
4. Any name, number, date or link that must be checked before sending?
5. Does the ending give the reader something to do or decide?

If all five pass, say "clear to send" and nothing else.

Below is an email thread I have been added to late.
[paste thread]

Do two things, clearly separated.

PART 1: What I need to know
- What is being decided
- What each named person wants
- What is actually being asked of me
- Anything already agreed that I should not reopen

PART 2: A draft reply from me
[my role and position on this]. Tone: [direct and warm].
Under [150] words. Do not restate the thread back to people who lived it.

Write an email declining [what I am declining] from [who asked].

Context they do not know: [the real reason, whether or not I want to share it]
What I am willing to say: [the version that goes in the email]
Relationship I want afterwards: [what I need from this person in future]

Rules
- Decline in the first two sentences. No lengthy warm-up.
- One reason, stated once, not defended.
- Offer exactly one alternative if a genuine one exists. If not, offer none.
- No apologising more than once.

[Client] owes [amount], due [date], now [number] days late.
Relationship so far: [good / strained / new]. I [do / do not] want more work from them.

Write three emails I can send in sequence:
1. Assumes it was an oversight. Friendly. Under 80 words.
2. Assumes it was not. Specific about dates and amounts. Under 120 words.
3. States the next step I will take and when. Under 100 words. Not a threat, a fact.

No guilt, no passive aggression, and no exclamation marks in any of them.

I want [person A] to introduce me to [person B] at [company].
Why B should care: [the specific reason, in my words]
What I am asking B for: [one concrete thing]

Write two blocks.

BLOCK 1: A short note to A, giving them an easy way to say no.
BLOCK 2: A forwardable paragraph A can paste to B without editing, written so it
reads as A's summary of me, not as my pitch about myself. Under 90 words.

I got this wrong: [what happened, plainly]
Consequence for them: [the actual impact]
What I have already done: [remedy, if any]
What I will do: [commitment, with a date]

Write the email.

Rules
- Name the mistake in the first sentence, without softening words.
- No explanation of why it happened unless it changes what they should expect next.
- No "if anyone was affected" constructions.
- The commitment must be checkable, with a date.
- Under [120] words.

Summarise this thread for [person] who is joining it now and has [two] minutes.
[paste thread]

Format
- Situation: three sentences
- Decisions already made: bullets, each with who made it
- Still open: bullets, each with who is blocking it
- What we need from [person]: one sentence

Anything ambiguous in the thread goes under "Unclear from the thread", not into
the summary as though it were settled.

Here are my notes from a meeting.
[paste notes]

Write the follow-up email to [attendees].

Include only: what was decided, what each person owns, and the dates.
Exclude: discussion, alternatives considered, and anything said that was not agreed.
If an owner or a date is missing from my notes, write TBC and list it at the end
under "Please confirm" rather than inventing it.

Under [200] words.

Deadline: [original date]. What I need: [new date]. Reason: [the real one].
Who this affects: [downstream people or work]
What I can deliver on the original date: [partial thing, or nothing]

Write the request.
- Ask in the first two sentences.
- Give the reason once, without self-criticism.
- Lead with what they get on the original date, if anything.
- Do not ask them how they feel about it. Ask them to confirm.

Give feedback to [person, and our relationship] about [the behaviour or work].

Use this structure and label the parts:
Situation: when and where, specifically
Behaviour: what was observable, with no interpretation
Impact: what followed, on the work or on others
Ask: what I want to be different, once, concretely

Rules
- No compliment sandwich.
- No mind-reading. Do not describe their intentions.
- End by inviting their side of it, and mean it.

Change: [what is changing]. Effective: [date]. Audience: [who].
Why it is happening: [the real reason]
What I can say publicly: [the sharable version]
What is not changing: [the reassurance, if true]

Write the announcement.
- State the change and the date in the first two sentences.
- Then the three objections this audience will raise, answered in one line each.
- No enthusiasm the audience will not share.
- End with where to ask questions and by when.

I emailed [who] about [what] on [date]. No reply.
What is genuinely in it for them: [the specific value]

Write three follow-ups, each shorter than the last, each adding something new
rather than repeating the ask:
Touch 1 (day [3]): one useful thing, no ask beyond a yes or no
Touch 2 (day [8]): a different angle on why now
Touch 3 (day [15]): close the loop and give them a graceful exit

None may contain the words "just checking in" or "bumping this".

Write a short reusable reply for [type of incoming request I get often]
that I can send [several] times a week with only [one] variable changed.

Requirements
- Under 60 words
- Clear no in the first sentence
- One sentence explaining my standing policy, so it does not read as personal
- One line pointing them somewhere genuinely useful, if such a place exists
- Signable without editing anything except [the name]

Below are [number] email subject lines and first lines.
[paste list]

Return a table only, one row per email:
Sender | What they want (five words) | Category | Suggested action | Minutes

Categories: needs me / needs someone else / needs a decision / read only / bin
Suggested action: reply now / reply this week / delegate / archive / unsubscribe

Then list the three I should do first and why, in one line each.
Do not draft any replies.

Here is the finished email body. Write the subject line.
[paste body]

Give me six options:
- Two that state the ask literally
- Two that name the outcome for the recipient
- Two that would work if this landed among [number] unread messages

Under [50] characters each. No colons used to fake structure.
Then pick one and give the matching preview text, under [90] characters.

Facts only, no adjectives from me:
- What it does: [literal function]
- Who buys it: [role, company type, trigger event]
- What they used before: [status quo, including doing nothing]
- What they measure: [the metric they are judged on]
- What we can prove: [evidence, or "nothing yet"]

Write three positioning statements, each in this form:
For [who], who [situation], [product] is a [category] that [benefit].
Unlike [alternative], it [difference].

Then say which claim in each one we cannot currently back up.

We sell [product] to [buyer]. We believe they buy because [assumption].

Write [15] interview questions to test that belief with real customers.

Rules
- Every question asks about past behaviour, not future intention.
- No question mentions our product before question 10.
- No leading questions. If a question implies a preferred answer, rewrite it.
- Include three questions designed to surface why they would NOT buy.

Group them: context, current workaround, trigger, evaluation, objections.

Product: [what it does]. Visitor: [who, arriving from where, expecting what].
Single action we want: [the one conversion]
Proof available: [list, or "none"]

Plan the page section by section. For each section give:
- Its job in one sentence
- The headline
- What content goes in it (described, not written)
- What would make a visitor leave at this point

Six to nine sections. Do not write body copy. If we lack proof for a section
that needs it, mark the section [BLOCKED: needs proof].

Platform: [platform]. Format: [format].
Hard limits: headline [n] characters, body [n] characters, description [n].
Product: [one sentence]. Audience: [who and their state of mind]
Claim we can substantiate: [claim] Source: [where it comes from]

Write [10] variants inside the limits, counting characters and showing the count.
Split them: 4 problem-first, 3 outcome-first, 3 plain and factual.
No variant may contain a claim not in the substantiated claim above.

Source:
[paste the article, transcript or report]

Extract [12] social posts. Each must make one point that stands alone and each
must come from a different part of the source.

For each: the post, the exact sentence in the source it came from, and the format
(observation / number / contrarian take / how-to / quote).
No post may summarise the whole piece. No hashtags. No emoji unless I ask.
If the source only supports [n] genuinely distinct points, stop at [n] and say so.

Themes for the quarter: [theme 1], [theme 2], [theme 3]
Channels: [channels]. Cadence: [posts per week]. Team capacity: [hours per week]

Build a [12] week calendar as a table:
Week | Theme | Format | Working title | Effort (S/M/L) | Depends on

Rules
- No more than one L per week.
- Every theme appears in at least three different formats.
- Flag any week where planned effort exceeds capacity.
- Leave [two] slots per month empty and label them REACTIVE.

Target query: [query]. Searcher intent as I read it: [what they actually want]
Our angle that others do not have: [the differentiator, or "none yet"]

Write the brief, not the article:
- The one-sentence answer the page must give in its first paragraph
- Six to ten section headings, at least half phrased as questions
- Under each, the specific question it answers
- What evidence or original material each section needs
- Three things every competing page does that we should not

If the angle is "none yet", say the page is not worth writing and explain why.

List: [who they are and why they subscribed]. Issue [n], sending [date].
Raw material I have: [links, notes, half-ideas]

Plan one issue:
- The single idea the issue is about
- The opening 40 words
- Three sections, each with a job
- One thing to cut if it runs long
- The call to action, and what makes it earned rather than tacked on

If my raw material does not add up to one idea, tell me which two ideas it
actually contains and make me choose.

Here is a competitor's [homepage / ad / email].
[paste text]

Analyse it as a message, not as a product:
- Who they are talking to, in their words
- The problem they claim to solve
- What they claim without proving
- What they prove
- The objection they are working hardest to defuse
- What they are conspicuously silent about

Then: one sentence on what this tells us about their buyer, and one on where
our message differs honestly. Do not evaluate their product quality.

Customer: [name or type]. What they achieved: [outcome, if known]
What we want the case study to prove: [the claim]

Write the interview questions in four blocks:
Before: what life looked like, what it cost them, what they tried
Decision: what triggered the search, who else they looked at, what nearly stopped them
After: what changed, what did not, what surprised them
Numbers: the specific figures to ask for, and the polite way to ask for each

Add three questions whose answers would make us drop this case study.

Launching: [what]. Date: [date]. Who it is for: [audience].
The one thing it lets them do that they could not before: [capability]
What it does not do yet: [honest limits]

Write the same announcement at three lengths:
- 25 words, for a chat message
- 100 words, for an email
- 300 words, for a blog post

All three must contain the limits. The 25-word version must still be true if
read alone. No version may use the word "excited".

Video topic: [topic]. Platform: [platform]. Viewer state: [scrolling, half-watching]
The payoff the video actually delivers: [payoff]

Write [15] opening hooks, first three seconds only.
Split: 5 that state a result, 5 that name a mistake, 5 that open mid-story.

Every hook must be one the video can pay off. Mark any that overpromise with
[OVERPROMISE] and rewrite it beneath. Under 12 words each.

Here are [three] things we have published that sound right.
[paste sample 1]
[paste sample 2]
[paste sample 3]

Infer our voice and write a charter a new writer could follow:
- Five rules stated as do-this-not-that pairs, each with a line quoted from the samples
- Words we clearly use
- Words we clearly avoid
- Sentence rhythm, described concretely
- Two cases where the voice should be broken, and why

Do not describe the voice with adjectives like friendly or professional.

Compare these two versions of the same message for [audience].
VERSION A: [paste]
VERSION B: [paste]

Score each 1 to 5 on: clarity of the ask, specificity, credibility of claims,
relevance to this audience, and how it reads on a phone.
Return a table with the scores and a five-word reason per cell.

Then name the single strongest sentence across both, and write a version C
that keeps it and fixes the lowest-scoring dimension.

Prospect: [company]. Person: [name and role]. Meeting is [when], about [topic].
What I know already: [paste notes, site copy, job ads, anything]

Turn this into a brief:
- What this company appears to be trying to do this year, with the evidence line
- What this person is likely measured on
- Three things that changed for them recently, if the material shows any
- Two questions only someone who read this material could ask
- What I do not know and should ask directly

Mark anything inferred as [INFERRED]. Do not add facts from outside my material.

Recipient: [name, role, company]
The one specific, verifiable thing I know about them: [detail]
What I do: [one sentence]
The single outcome I am offering: [outcome]

If the specific detail above is generic or made up, refuse to write the email and
tell me to go and find a real one.

Otherwise: under 90 words, the detail in the first sentence, one sentence on the
outcome, one question that is easy to answer with a yes or no. No pleasantries.

Deal context: [what they came in for]. My product: [one sentence].
What I must learn to qualify: [budget, timing, decision process, or your criteria]

Build a [30] minute discovery plan:
- Opening two sentences
- Eight questions in the order I should ask them, each with the reason it is placed there
- Two follow-up probes per question, for when the answer is thin
- The three answers that mean I should disqualify and end early
- The last question to ask before booking a next step

Play the buyer. You are [role] at [company type] and your objection is:
[the objection, in their words]

Rules for you
- Stay in character. Do not coach me.
- Give ground only if my answer genuinely addresses the objection.
- If I use a technique instead of an answer, say so as the buyer would.
- After [six] exchanges, break character and tell me which of my responses
  actually moved you and which one you would have hung up on.

Start with the objection. I will reply.

Demo with [who] on [date]. What they reacted to: [the moments they leaned in]
What they questioned: [their doubts]. Agreed next step: [what and when]

Write the follow-up.
- Reference one specific thing they said, quoted.
- Answer the doubt they raised, in two sentences, including anything we cannot do.
- Restate the next step with the date and who owns it.
- Attachments or links: [list], each with one line saying why it is there.
Under [150] words.

My champion is [name, role]. The new stakeholder is [name, role],
who cares about [their concern] and has probably heard nothing about us.

Write an email to the new stakeholder that:
- Says how I got to them, honestly
- Frames the project in terms of their concern, not my champion's
- Contains one sentence my champion would be happy to see forwarded
- Asks for [the smallest useful next step]

Do not imply more consensus than exists. Under [120] words.

Our price: [price and structure]. Their reaction: [what they said]
What I can flex on: [levers]. What I cannot: [hard floors]

Write a script, not an email:
- My opening response to their reaction, word for word
- Two questions to ask before naming any concession
- The order in which to release the levers, cheapest first
- The exact sentence for holding the floor without hostility
- The sentence that ends the conversation if we are below the floor

Deal: [account]. Lost to: [competitor / no decision / budget]. Cycle length: [time]
Everything I have: [paste notes, emails, call summaries]

Analyse it:
- The earliest point in the timeline where this was already lost, and the evidence
- What I believed that turned out not to be true
- The question I should have asked and did not
- Which of the stated reasons was likely the real one, and why
- One thing to change in the next deal, specific enough to check

No consolation. If it was unwinnable from the start, say that instead.

Account: [name]. Renewal: [date]. Contract: [value].
Signals I have: [usage, contacts, tickets, quiet periods, champion changes]

Score each factor 1 to 5, where 5 is high risk, in a table:
Factor | Score | Evidence | What would lower it

Factors: champion status, breadth of use, value demonstrated, competitor presence,
budget cycle, unresolved issues.
Then give a total, the single highest-leverage action, and the date to take it by.
Say if my signals are too thin to score honestly.

Us: [product, one sentence]. Them: [competitor].
What I know about them: [paste only what I have verified]

Build a battlecard:
- Where they genuinely win, named plainly
- Where we genuinely win
- The deal shapes where they are the right choice, and we should walk away
- Three questions that surface the difference without mentioning them
- What to say if the buyer raises them first

Do not state anything about their pricing, ratings or roadmap that is not in my
material. Where I have not supplied it, write "not verified".

Person: [name, role]. Why I want to connect: [the honest reason]
Shared context: [group, event, mutual contact, or "none"]

Write a connection note under [300] characters.
- If shared context is "none", say why I am reaching out in the first clause.
- No compliments about their profile.
- No pitch. The goal is acceptance, not a meeting.
- One line that gives them a reason to reply, if they want to.

Here are my raw call notes.
[paste notes]

Return only these fields, in this order, one per line, with nothing else:
Company | Contact and role | Stage | Pain (their words) | Current solution |
Decision process | Budget signal | Timeline | Next step | Next step date |
Risk | Confidence (high/medium/low)

For anything the notes do not contain, write UNKNOWN. Do not infer.
Quote their words where the field asks for their words.

Customer: [name]. What they have got out of it: [specific result, in their words]
Who I want to meet: [description of the ideal introduction]

Write the ask.
- Open with their result, quoted, not with my request.
- Ask for one specific type of person, not "anyone who might benefit".
- Make it trivially easy to say no.
- Include a forwardable two-sentence blurb they can paste.
Under [110] words.

Explain this code to someone competent in [language] but new to this codebase.
[paste code]

In this order:
1. What it does, in two sentences
2. Its inputs, outputs and side effects
3. A line-by-line walk-through of only the non-obvious parts
4. The assumptions it makes about its caller
5. What would break it

Do not suggest improvements. If something is genuinely ambiguous without more
of the codebase, say which file you would need to see.

Review this diff for correctness. Ignore style, naming and formatting entirely.
[paste diff]

Report only defects, each as:
- Location
- What goes wrong, and the input that triggers it
- Severity: breaks / silently wrong / edge case
- The smallest change that fixes it

Look specifically for: off-by-one, unhandled null or empty, error paths that
swallow failures, changed behaviour for existing callers, and concurrency.
If you find nothing, say "no correctness defects found" rather than inventing one.

Error:
[paste stack trace]

Relevant code:
[paste code]

What I have already tried: [list]

Give me five hypotheses, ranked by likelihood, and for each:
- Why the trace supports it
- The single cheapest check that would confirm or eliminate it
- What I would see if it were true

Do not write a fix yet. Start with the cheapest check across all five.

Function under test:
[paste function]

Before writing any test code, produce a table of cases:
Case | Input | Expected | Why this case matters

Cover: the happy path, boundaries, empty and null, wrong types, the largest
plausible input, and any case where the current behaviour is arguably wrong.
Mark that last group [BEHAVIOUR QUESTION] rather than assuming an answer.

Then, and only then, write the tests in [framework].

Refactor this so that [the goal: it is testable, the nesting is flat, the
duplication is gone], while keeping observable behaviour identical.
[paste code]

Return three things:
1. The refactored code
2. A list of every behaviour you deliberately preserved, including the odd ones
3. Anything you were tempted to fix but left alone, and why

If you cannot preserve a behaviour, stop and tell me which one and what it costs.

Translate this from [source language] to [target language].
[paste code]

Rules
- Idiomatic in the target, not a transliteration.
- Preserve behaviour including error cases.
- Where the target has no equivalent of a source feature, say so in a comment
  and show what you did instead.
- List the standard-library or dependency differences I will need to handle.

Then list every place where the two versions could diverge under load or failure.

I need a regular expression for [flavour: PCRE, JavaScript, Python] that matches
[what it should match] and does not match [what it must not match].

Give me:
1. The expression
2. A plain-English reading of it, part by part
3. Ten test strings: five that must match, five that must not, with expected results
4. Two inputs that would make it behave badly, including any catastrophic backtracking

If the thing I described is not reliably matchable with a regex, say so and name
what to use instead.

Schema:
[paste CREATE TABLE statements or column lists]

Question to answer: [what I want to know]
Dialect: [Postgres / MySQL / SQLite / BigQuery]. Rough row counts: [numbers]

Write the query, then:
- Explain the join logic in two sentences
- Name the assumption you made about duplicates and nulls
- Say which index would matter at these row counts
- Give me one sanity-check query that would tell me the result is wrong

Current error message: [paste it]
When it fires: [the condition]
Who sees it: [end user / developer / operator]
What they can actually do about it: [the real remedy, or "nothing"]

Rewrite it. The new message must say what happened, what it means for them, and
what to do next, in that order. Under [140] characters if it is user-facing.

If the honest answer to "what can they do" is nothing, say so in the message and
tell me what to log instead.

Write reference documentation for this, for a developer integrating against it.
[paste code or signatures]

For each public function: purpose, parameters with types and constraints, return
value, errors thrown and when, and one realistic example call with output.

Rules
- Document only what the code does, not what it should do.
- Where behaviour is unclear from the code, write [UNDOCUMENTED: needs an answer]
  rather than guessing.
- No introductory paragraph about what the module is for.

Upgrading [package] from [version] to [version] in a [language, framework] project.
How we use it: [paste the imports or call sites]

Produce a migration plan:
- Breaking changes that affect our usage specifically, each with our call site
- Changes that affect nobody in our code, listed separately and briefly
- Order of operations, smallest reversible step first
- What to test after each step
- How to roll back at each step

Flag anything you are unsure about rather than assuming the changelog is complete.

Symptom: [what is slow, measured how]
Numbers I have: [timings, percentiles, volumes]
Code or query involved: [paste]

Do not optimise anything yet. Instead:
1. List the plausible causes, ordered by how often they turn out to be the cause
2. For each, the measurement that would confirm it and the tool to take it
3. What my existing numbers already rule in or out
4. The one measurement to take first

Then say what you would expect the profile to look like if cause 1 is correct.

Review this for security issues only.
[paste code]
Context: [where it runs, what it trusts, who can reach it]

Check specifically: input handling, injection into any interpreter or query,
authentication and authorisation boundaries, secrets in code or logs, unsafe
deserialisation, and anything that trusts a client-supplied value.

For each finding: the line, the attack, the precondition, and the fix.
Rank by exploitability given the context above, not by generic severity.
State plainly if the context I gave is too thin to judge.

Diff:
[paste diff]
Why I made this change: [the reason, in my words]
Related issue: [id or "none"]

Produce:
1. A commit subject under 72 characters, imperative mood
2. A commit body: what changed and why, no bullet lists of file names
3. A PR description with: the problem, the approach, what a reviewer should look
   at hardest, what is deliberately out of scope, and how to test it

Do not describe the diff line by line. The reviewer can read it.

I have a bug I cannot pin down. Do not propose causes yet.

Ask me questions one at a time, waiting for each answer, until you can state a
single testable hypothesis. Prefer questions that split the possibility space in
half over questions that confirm a favourite theory.

Start by asking what changed most recently and what the last known good state was.
When you have the hypothesis, state it, plus the one experiment that would falsify it.

Below is a column of [what it is] with inconsistent formatting.
[paste up to 50 rows]

Return a table: Original | Cleaned | Rule applied | Confidence

Rules
- Do not guess at ambiguous values. Mark them AMBIGUOUS and leave Cleaned empty.
- Where a value could be two things (a date in two formats, a name in two orders),
  list both and mark it AMBIGUOUS.
- At the end, list the cleaning rules you applied, so I can run them on the full set.

Spreadsheet: [Excel / Google Sheets]. My data: [describe columns and their letters]
What I want to calculate: [in plain language]

Give me:
1. The formula
2. A plain reading of what each part does
3. What it returns when the data is empty, has text where numbers should be,
   or contains duplicates
4. A smaller test case with three rows and the expected result

Do not calculate anything from my real data. I will run the formula myself.

Here is a table of results.
[paste table]
What it measures: [definition of each metric]
What I expected to see: [my prior]

Tell me:
- The three things that stand out, with the numbers that support each
- Which apparent patterns could be explained by how the data was collected
- What is missing from this table that I would need before acting
- Which of my expectations this does and does not support

Do not recompute my numbers. Do not speculate about causes I cannot test.

We want to measure [the thing]. Current rough definition: [what people say now]

Write a proper definition covering:
- The exact numerator and denominator
- The population included, and who is excluded and why
- The time window and how late-arriving data is handled
- What it does when the denominator is zero
- Three ways this metric could be gamed
- One companion metric that would catch that gaming

Then give the one-sentence version to put on the dashboard.

Here are my survey questions.
[paste questions]
Who is answering: [population]. What I want to learn: [the real question]

For each question: the bias it introduces, if any (leading, double-barrelled,
loaded scale, recall failure, social desirability), and a rewritten version.

Then answer two things: which questions do not serve my stated goal at all, and
what I am not asking that I will wish I had.

I want to show [the relationship or comparison] to [audience] so they [decide what].
My data: [rows, columns, types, ranges]

Recommend a chart type and say why, then:
- What it will show clearly
- What it will hide
- The second-best option and when it would be better
- Three ways this chart type is commonly made misleading, and how to avoid each
- The axis, sort order and baseline decisions I need to make deliberately

Product: [what it is]. Event that defines the cohort: [signup, first purchase, etc.]
Outcome I care about: [retention, revenue, usage]
Data I have: [tables and fields]

Plan the analysis before any query is written:
- How cohorts are bucketed and why that interval
- The outcome measured at which offsets
- How partial cohorts at the edges are handled
- The confounder most likely to fool me here
- What result would be genuinely surprising, defined before I look

Result: [paste the finding, with its numbers]
Method used: [what was done]
Audience: [who needs to understand it and what they will do with it]

Explain it in under [200] words, covering what was found, how confident we should
be, and what it does not tell us.

Rules
- No statistical vocabulary without a plain-language gloss in the same sentence.
- State the uncertainty as a range in real units, not only as a p-value.
- End with the one decision this result should and should not influence.

Here is my analysis and conclusion.
[paste both]

Act as a reviewer whose job is to stop bad analysis shipping:
- Which alternative explanation have I not ruled out
- What selection effect could produce this pattern with no real effect
- What would this look like if the opposite were true
- Which number is doing the most work, and how fragile is it
- What I would need to collect to settle it

Do not tell me the analysis is good. Assume there is at least one real problem.

Here are the column names from [table or export], with three sample rows.
[paste]

Produce a data dictionary: Column | Likely meaning | Type | Unit or format |
Confidence | Question to ask the owner

Rules
- Confidence must be honest. Most of these are guesses from a name.
- Any column whose meaning you cannot infer gets Confidence: none and a question.
- Flag columns whose names suggest they contain personal data.

Experiment: [what was tested]. Hypothesis before the test: [stated in advance?]
Results: [paste numbers, sample sizes, duration]
Decision to be made: [what happens next]

Write the readout:
- What we tested and why, in three sentences
- What happened, in real units with uncertainty
- Whether it answers the question we asked
- The recommendation, with the confidence level attached
- What we would need to be more confident, and whether it is worth the cost

If the sample or duration cannot support a decision, lead with that.

My forecast: [the numbers and the period]
How I built it: [describe the method, however rough]

Do not produce a forecast. Instead, list every assumption baked into mine,
grouped as: stated, implied, and probably unnoticed.

For each: what it assumes, how sensitive the result is to it, and the cheapest
way to check it. Then name the single assumption that would do the most damage
if it is wrong, and what the forecast becomes if it is.

Anomaly: [metric] moved from [x] to [y] on [date].
What I know: [context, releases, campaigns, seasonality, known incidents]

Give me a triage checklist in the order I should work through it, cheapest first:
instrumentation change, definition change, data pipeline delay, composition shift,
seasonality, external event, real change.

For each: the specific check, the evidence that would confirm it, and roughly how
long the check takes. Stop-rule: which finding means I can stop looking.

My research question: [the question as I currently have it]
Why I am asking: [the decision it feeds]

Do not answer it. Decompose it:
- The sub-questions that must be answered first, in dependency order
- Which are empirical, which are definitional, and which are value judgements
- For each empirical one, what kind of evidence would settle it
- Which sub-question, if answered, would make the others unnecessary

Then rewrite my original question as the sharpest version of itself.

Here are the sources I have found.
[paste titles, publishers, dates, URLs]

For each, in a table: Source | Type | Independence | Recency risk | What it can and
cannot support

Type: primary / secondary / aggregator / promotional / unknown
Independence: is the publisher interested in the answer being a particular way

Then rank them, and name the one claim I should not make from any of these.
Do not tell me what any source says. You have only seen the titles.

Claim under examination: [the claim]

Source A: [paste relevant passage]
Source B: [paste relevant passage]
Source C: [paste relevant passage]

Return a table: Source | What it actually says, quoted | Scope and date |
Agrees / disagrees / addresses something else

Then: where they genuinely conflict, where they only appear to, and what the
strongest defensible sentence is that all three support. Quote only from the
text above. If a source does not address the claim, say so.

Paper or report:
[paste abstract, method and results, or the whole thing]

Produce a fixed-format note:
- Question asked
- What was done, in three sentences
- Sample or data, with its limits
- Headline finding, with the actual numbers
- What the authors themselves say it does not show
- How it could be wrong
- One sentence on how it relates to [my topic]

If a section is not present in what I pasted, write NOT IN SOURCE.

Source text below. Extract every claim relevant to [my question].
[paste text]

For each: the claim in my words, the exact sentence it came from in quotation
marks, and a location anchor (section, paragraph or page if the text shows one).

Rules
- No claim without a quoted sentence.
- If the text implies but does not state something, mark it [IMPLIED] and quote
  the nearest supporting sentence.
- If my question is not addressed, return "not addressed in this source" and stop.

My position: [state it as strongly as I hold it]
Why I hold it: [my reasons]

Build the strongest honest case against me. Not a caricature: the version someone
intelligent and informed would actually argue.

Give me: their core claim, their three best supporting points, the evidence they
would cite, the weakest part of my position from their view, and what they would
concede to me. End with the one question they would ask that I would struggle with.

Below are documents, notes and messages about [the subject], in no order.
[paste material]

Build a timeline: Date | What happened | Source | Certainty

Rules
- Only include events the material states. No connective inference.
- Where a date is approximate or contradicted, say so in the Certainty column.
- After the table, list the gaps: periods where something clearly happened but the
  material does not say what.

I am new to [field]. Here is material I am reading.
[paste text or list the terms I keep hitting]

Build a glossary of the [20] terms that matter most, ordered so that no definition
depends on a term defined later.

Each entry: the term, a definition in one sentence, what it is often confused
with, and why it matters in this field specifically.
Mark any term where usage genuinely differs between sub-fields.

I have [30] minutes with [who they are and what they know].
What I am trying to learn: [the real goal]
What I already know: [paste my current understanding]

Write the questions.
- Two that establish how they know what they know
- Six that only they could answer, not findable in public material
- Two that test my current understanding against theirs
- One that invites them to tell me what I did not think to ask

Order them so an early answer does not spoil a later question.

Here are my notes from [number] sources on [topic].
[paste notes]

Write a synthesis memo, not a summary of each source:
- The three findings the sources agree on, with which sources support each
- Where they disagree, and the most likely reason for the disagreement
- What nobody addresses
- The confidence level of the overall picture, and what drives it
- What I should do or decide, given all of it

Cite by source name for every claim. Anything with one source gets marked as such.

Below are extracts from [number] sources on [topic].
[paste extracts, labelled]

Find every place two sources contradict each other. For each:
- The contradiction, with both quoted
- Whether it is a real disagreement or a difference of scope, definition or date
- Which is better positioned to be right, and why
- What would resolve it

Ignore differences of emphasis. Only flag things that cannot both be true.

Belief I currently hold: [the belief]
Confidence: [rough percentage]
Evidence I am relying on: [list it]

Help me make this falsifiable:
- The observation that would most reduce my confidence
- The observation that would most increase it
- Which of my current evidence is weakest, and why
- What I would expect to see if I were wrong but had not noticed yet
- A concrete check I could run in under [an hour]

Do not argue for or against the belief.

Concept: [the concept]

Explain it three times:
1. To someone who has never heard of it: no jargon, one analogy, 100 words
2. To someone in [adjacent field]: use their vocabulary, say what it is not, 150 words
3. To someone who will have to use it correctly: precise, including the edge cases
   and the common mistake, 250 words

Then state the single sentence that all three versions are trying to convey.

I am going to explain [concept] in my own words. Your job is to find where my
understanding is thin, not to praise it.

My explanation:
[paste your attempt]

Return:
- What I got right, in one line
- Every place my explanation is wrong, with the correction
- Every place it is right but for the wrong reason
- The question I could not answer given what I just wrote
- The next thing to learn, and why it is next

Material:
[paste notes, chapter or article]

Write [20] recall questions that test whether I understood it, not whether I
remember its phrasing. Mix: [8] factual, [8] applied, [4] that require combining
two parts of the material.

Do not include the answers in this response. Number them.
When I answer, mark each right, partly right or wrong, and tell me which parts
of the material I should re-read.

Goal: [what I want to be able to do, stated as a capability not a topic]
Deadline: [date]. Time available: [hours per week, realistically]
Starting point: [what I already know]. Resources: [what I have access to]

Build a four-week plan as a table: Week | Focus | Activity | Output | Check

Rules
- Every week produces an output I can be wrong about, not just reading done.
- The check is how I know the week worked.
- Front-load anything the later weeks depend on.
- If the goal does not fit the hours, say so and propose a smaller goal.

Topic: [the technique or problem type]
My level: [where I am]

Give me:
1. One fully worked example, showing every step and stating why each step was
   chosen, not just what it was
2. The same problem type with different numbers, unworked
3. The three mistakes people make on this exact problem
4. A way to check my own answer without you

Do not solve number 2. I will, and then you can mark it.

Topic: [topic]. Here is what I currently believe about it:
[paste your beliefs, however rough]

Identify the misconceptions. For each:
- What I believe
- What is actually the case
- Why this misconception is common and where it usually comes from
- The example that makes the difference obvious
- What downstream conclusion of mine is wrong because of it

If a belief is correct but commonly doubted, say that too.

Explain [concept] with an analogy to [domain I already know well].

Then, in equal detail, explain where the analogy breaks: which properties do not
carry over, which conclusions I would reach that are wrong, and the case where
relying on the analogy would actively mislead me.

Finish with a second, different analogy that breaks in a different place, and say
which of the two is safer for a beginner.

Quiz me on [topic] at [starting level].

Rules
- One question at a time. Wait for my answer.
- If I get it right, make the next question harder. If wrong, easier, and explain
  the specific gap before moving on.
- Every fourth question, return to something I got wrong earlier.
- No multiple choice unless I ask.

After [12] questions, stop and give me a map of what I know solidly, what is shaky,
and what to study next.

Topic: [topic]. Scope: [how far to go]

Produce a text concept map:
- The [5] to [8] core concepts, each in one line
- For each pair that is genuinely related, one line stating the relationship and
  its direction (causes, is a type of, is required for, contradicts)
- The one concept everything else depends on
- The two concepts most often confused with each other

No diagram. Indented text and explicit relationship statements only.

This paragraph from [source] is dense and I have read it four times.
[paste paragraph]

Do this:
1. Restate it in plain language, keeping every claim
2. Define each technical term as it is used here, not in general
3. Identify the sentence that carries the paragraph's actual argument
4. Say what the paragraph assumes I already know
5. Give the one question I should be able to answer if I have understood it

Topic: [topic]. Current level: [level]. Target: [what I need to handle]

Write [10] practice problems that step up in difficulty:
- 1 to 3: single-step, testing the definition
- 4 to 6: two-step, requiring a choice of method
- 7 to 9: multi-step, with one distractor detail each
- 10: a problem where the correct answer is that there is not enough information

Answers in a separate block at the end, each with the reasoning, not just the result.

Here are the questions I got wrong and what I answered.
[paste question, my answer, correct answer, for each]

For each: was this a knowledge gap, a misreading, a method error, or a careless
slip? Give the evidence for your classification from what I wrote.

Then aggregate: which category dominates, what that implies about how I should
prepare differently, and the three specific things to drill. Do not re-teach the
content unless the category is knowledge gap.

Book: [title and author]. Why I am reading it: [the reason]
Time available: [hours or weeks]. Background: [what I bring to it]

Give me a reading plan:
- Which chapters matter most for my reason, and which I can skim or skip
- The order to read them in, if not front to back
- One question to hold in mind for each key chapter
- What to write down after each, in one line
- The test of whether I got what I came for

Say plainly if this book is the wrong one for my stated reason.

Job description:
[paste it]

My CV:
[paste it]

Return a table: Requirement | Do I meet it? | Evidence from my CV | Gap

Rules
- Evidence must be quoted from my CV. If there is none, the answer is no.
- Do not soften a no.
- Separate the requirements that are genuinely load-bearing from the boilerplate.

Then: the three gaps worth addressing in a cover letter, and the one that means
I should probably not apply.

Here is how I currently describe a role, in duties.
[paste the bullets]

Additional facts I can supply: [numbers, scale, before-and-after, team size,
what would have happened without me]

Rewrite each as an achievement bullet: what changed, by how much, and my part in
it. Keep the verb specific.

If I have not given you a number or an outcome for a bullet, do not invent one.
Return it as [NEEDS: what number would make this land] instead.

Role: [title] at [company]. Job description: [paste key parts]
Why this company specifically: [the honest reason, even if it is money or location]
The most relevant thing I have actually done: [the story, in rough form]

Write the letter.
- Open with the relevant thing I did, not with the fact that I am applying.
- One paragraph connecting it to what the role needs.
- One sentence on the honest reason, made presentable but not fabricated.
- No sentence that would work for a different company.
Under [250] words.

Question I will be asked: [the question]
The real situation I want to use: [dump everything you remember, unstructured]

Shape it into an answer:
Situation, Task, Action, Result, and one sentence on what I would do differently.

Rules
- Use only what I gave you. Mark anything missing as [YOU NEED TO RECALL: what].
- The Action section must be the longest, and must say what I personally did.
- Under 90 seconds spoken, which is roughly 200 words.
- Then list the two follow-up questions this answer invites.

Role: [title]. Company: [what I know about it]. Stage of process: [round]
Who I am meeting: [their role]
What would make me decline an offer: [my actual deal-breakers]

Write [10] questions, ordered:
- Three that test my deal-breakers without announcing them
- Three about how the work actually gets done here
- Two that only this person's role could answer
- Two whose answers would tell me something bad if the answer is evasive

For each, add what a worrying answer would sound like.

Offer: [what was offered]. My target: [number]. My floor: [number].
My leverage: [competing offer, scarce skill, internal need, or "none"]
Non-salary things I would trade for: [list]

Write the script:
- The exact sentence to open with after they name the number
- What to say if they ask for my number first
- Two responses to "that is the top of the band"
- The sentence for trading salary against [the non-salary item]
- The words for accepting, and the words for walking away

No apologising. No justifying the number with personal expenses.

Role: [title]. Notice period: [length]. Last day: [date].
Relationship with manager: [good / poor / neutral]
What I want afterwards: [reference, rehire eligibility, quiet exit]

Write the letter. Under [120] words.
- State the resignation and the last day in the first two sentences.
- One line of genuine thanks, specific enough to be true.
- No reasons, no feedback, no future plans.
- One line offering handover help, with what I will actually do.

Review period: [dates]. My role: [title].
What I did: [dump it all, unsorted]
What went badly: [be honest here, it stays between us until I edit]
How I am measured: [the stated criteria]

Organise it against the criteria. For each: what I did, the evidence, and the
result.

Include one genuine weakness with what I did about it. Do not use a strength
disguised as a weakness. Flag anything where I have claimed a result I cannot show.

Role: [title] at [company type]. Reporting to: [who]. Team: [size and shape]
What they hired me to fix: [the stated reason]
What I suspect the real reason is: [my read]

Write a 30-60-90 plan:
- Days 1 to 30: what to learn and from whom, with the questions
- Days 31 to 60: the first thing to change, chosen for being reversible
- Days 61 to 90: the thing that proves the hire was right

Each phase needs one visible output. Name the political risk in each phase.

Person: [who and why they are interesting]
How I found them: [the honest route]
What I actually want: [advice, an introduction, a job, information]

Write a message under [100] words.
- Say what I want by the second sentence.
- One specific reason it is them, drawn from something real.
- Make the ask small enough to say yes to in five minutes.
- No flattery, no "picking your brain", no offer to buy coffee.

Current field: [field]. Target field: [field].
What I actually do day to day: [describe it concretely]
What I know about the target: [however little]

Produce three lists:
- Skills that transfer directly, with the target-field name for each
- Skills that transfer with translation, and how to describe them
- Genuine gaps, ranked by how much they block me

Then: the smallest project I could do that would evidence the top gap being closed.
Do not tell me my experience is more relevant than it is.

Role: [title] at [company]. Stage reached: [where it ended]
What they said: [paste the rejection]
What I want: [feedback / to stay in touch / to reapply later]

Write a short reply, under [80] words.
- Thank them once, briefly.
- Ask for one specific piece of feedback, phrased so it is easy to give.
- If I want to reapply, say so in one sentence with a timeframe.
- No arguing with the decision, no restating my qualifications.

Job description:
[paste it]

Tell me what it probably means:
- The actual job, stripped of aspiration
- Which requirements are real and which are wish-list, with your reasoning
- What the phrasing suggests about the team's current state
- Any phrase that commonly signals a problem, quoted, with what it usually means
- The three questions to ask at interview to test your reading

Mark every inference as an inference. You are reading tea leaves, not facts.

Meeting purpose as it was given to me: [the vague version]
Attendees: [who, and their roles]. Length: [minutes]

Turn it into an agenda:
- The decision or output this meeting must produce
- Items in order, each with an owner, minutes, and whether it is inform, discuss
  or decide
- What must be read or done beforehand
- The item to cut first if we run late

If the purpose does not need a meeting, say so and name what would replace it.

Raw meeting notes:
[paste them]

Return only two lists, nothing else.

DECISIONS: what was decided, who decided it, and any condition attached
ACTIONS: task, owner, due date

Rules
- Nothing goes in DECISIONS unless the notes show it was actually settled.
- An action with no named owner is written as OWNER UNASSIGNED, not guessed.
- Anything discussed but unresolved goes under a third heading, OPEN.

Meeting: [what it is]. Organiser: [who and our relationship]
Why I am declining: [the real reason]
Whether I want the outcome anyway: [yes / no]

Write the decline, under [60] words.
- Decline in the first sentence.
- If I want the outcome, propose the smaller thing that gets it: a written update,
  a fifteen-minute call, a delegate.
- No inventing a conflict. If I am declining because it is not a good use of my
  time, say a version of that which is polite and honest.

Everything on my plate:
[paste the unsorted list]

Fixed commitments: [meetings and deadlines]
Real working hours available: [number]
The one thing that must happen this week: [it]

Produce a week plan.
- Put the one thing first, in the largest uninterrupted block.
- Group similar tasks so I switch context less.
- Show total planned hours against available hours.
- List what does not fit, and say plainly that it will not happen.

The task as it sits on my list: [the vague version]
Why it matters: [the reason]. What "done" looks like: [if I know]

Break it down:
- The first physical action, small enough to start in two minutes
- The next four actions in order
- Which of them depend on someone else, and who
- The one that I am actually avoiding, and why it is probably that one
- What "done" means, in checkable terms

If "done" is undefined, ask me the one question that would define it.

Who: [landlord / provider / institution]
What happened: [the facts, with dates]
What I have already done: [previous contact and their responses]
What I want: [the specific remedy]
What I am prepared to do next: [the escalation, if any]

Write it.
- Facts and dates first, in a numbered list.
- The request stated once, specifically, with a date to respond by.
- The escalation mentioned once, as a fact, not a threat.
- No emotion, no history beyond what supports the request.

Decision: [what I am choosing between]
Option A: [describe]. Option B: [describe]
What matters to me, roughly in order: [criteria]
Constraints that cannot move: [money, time, location, people]

Build a weighted comparison. Assign weights to my criteria and say what you
assumed. Score each option, show the arithmetic in the table, and give the total.

Then ignore the total and tell me: which criterion is actually deciding this, and
what I would need to believe for the other option to win.

Here is what is nagging me:
[dump the list, however trivial]

Sort it into: do now (under 5 minutes), book a time, needs a phone call, needs
someone else, waiting on something, and can be deleted from my brain entirely.

For anything recurring, say what would stop it recurring.
For anything with a deadline I have not mentioned, ask.
Return it as a checklist I can work through, not as advice.

What went wrong: [the facts, in order, with dates and amounts]
Evidence I hold: [receipts, screenshots, reference numbers]
What I want: [refund, repair, apology, replacement]
What I do not want: [vouchers, goodwill gestures, an explanation]

Write the complaint.
- Reference numbers and dates in a block at the top.
- The narrative in under [150] words.
- The remedy stated as one sentence, with a response deadline.
- One line on what I will do if there is no response, stated neutrally.

Call: [who I am calling and about what]
What I need to get: [the outcome]
What I am afraid will happen: [the honest fear]

Write me a script:
- The opening sentence, word for word
- The three facts I must get across, in order
- What to say when they say [the thing I am dreading]
- The sentence for asking them to repeat or slow down
- The closing that confirms what was agreed and gets it in writing

Keep every line short enough to read aloud without stumbling.

Errands: [list them, with any that are time-bound]
Where I will be: [start point, and any fixed stops]
Time available: [window]. Transport: [walking, car, transit]

Produce a route in order, with rough timings, grouping by location.
Note anything that will be closed at the time I would arrive, if I have given you
opening hours; if I have not, list what I need to check first.
Then say what to drop if I am running [20] minutes behind.

Task I do repeatedly and keep re-figuring-out: [the task]
How I currently do it: [walk through it, messily]
Where it usually goes wrong: [the failure points]

Write a standard procedure:
- Preconditions and what I need to hand
- Numbered steps, each a single action
- The checkpoint after each risky step
- What to do when [the usual failure] happens
- How long it should take when nothing goes wrong

Keep it to one screen. Anything longer will not be used.

What someone asked for: [the requested feature or fix]
Who asked: [who and how often]
What I currently believe the problem is: [my guess]

Do not describe a solution. Write the problem statement:
- Who has the problem, specifically
- What they are trying to do when they hit it
- What they do today instead
- What it costs them, in their terms
- How we would know the problem was solved, without naming a feature

Then list what we do not know, and the cheapest way to find each out.

Feature: [name]. Problem it addresses: [the problem statement]
Users affected: [who]. Constraints: [technical, legal, timeline]

Draft the spec sections with content, not placeholders:
Problem, goal, non-goals, users and their jobs, the proposed shape, what changes
for existing users, what we are explicitly not doing, open questions, and how we
will know it worked.

Non-goals must contain at least three real things people will ask for.
Open questions must contain anything you had to assume to write this.

Feature: [describe it]
Users: [roles involved]
Edge cases I already know about: [list]

Write the user stories. For each:
- As a [role], I want [capability], so that [outcome]
- Acceptance criteria as Given / When / Then, including at least one failure path
- What is out of scope for this story

Split anything that cannot be built and tested in [one] iteration.
Add a final story for the case nobody plans for: what happens on first use with
no data at all.

Committed scope: [list the items]
Deadline: [date]. Capacity honestly available: [what the team can actually do]
What the deadline is really for: [demo, contract, event, arbitrary]

Propose the cut.
- What ships, and why this set is coherent on its own
- What is cut, each with the argument for cutting it
- What is cut but must be communicated, and to whom
- The thing that looks cuttable but is not, and why
- The version of this that is worse but certain

Do not propose working longer hours as an option.

Project: [what it is]. Timeline: [dates]. Team: [who and what else they are on]
Dependencies: [external teams, vendors, approvals]

Build a register: Risk | Likelihood | Impact | Early warning sign | Owner | Response

Rules
- The early warning sign must be something observable before the risk lands.
- Include at least two risks that are about people rather than technology.
- Include one risk that comes from the project succeeding.
- Sort by likelihood times impact, and mark the top three.

Project: [describe it]. Planned launch: [date]

It is [six months] after launch and the project failed. Not partly: it failed.

Write the story of how, in past tense and in detail. Cover what the first sign was,
what we told ourselves at the time, the decision that turned out to be the fatal
one, and who noticed first.

Then, out of character: which parts of that story are most plausible, and what we
could put in place this week that would have caught it.

Items to prioritise:
[paste the list, with any known effort or impact]

My criteria, in order: [criterion 1], [criterion 2], [criterion 3]
Constraint: [capacity for the period]

Score each item against each criterion, showing your reasoning in one line per
cell. Produce the ranked list and the cut line given my capacity.

Then tell me which two items the scoring separates by less than the noise in my
own estimates, so I know where the ranking is not real.

What shipped: [list the changes, technically]
Who uses this: [audiences]
What broke or changed for existing users: [be specific]

Write two versions.
1. For users: what they can now do, what changed in something they already used,
   and what they need to do about it. No internal terms.
2. For internal teams: the same list with the technical detail, plus what support
   should expect to hear about.

Anything with a migration or a breaking change goes at the top of both.

Project: [name]. Period: [dates]. Audience: [who and what they care about]
What happened: [everything, unsorted]
What is off track: [the honest version]

Write the update, under [250] words:
- Status in one word, then one sentence justifying it
- What moved, with evidence
- What did not, and what we are doing about it
- Decisions needed from this audience, with dates
- What I would like them to stop worrying about

No hedging language on the status. Pick a word and defend it.

Team: [size and context]. Period: [what we are looking at]
What obviously went wrong: [the elephant, if there is one]
Team dynamic: [safe to speak / guarded / new to each other]

Design the session:
- The opening question, chosen to fit the dynamic above
- The activity, with timings, for [60] minutes
- How to surface the elephant without putting one person on the spot
- The question that gets at systems rather than individuals
- How the session ends with at most three changes, each with an owner

Say what you would do differently if the dynamic were the other one.

Requests received:
[paste them, with who asked and when]

Return a table: Request | The underlying need | How many asked | Already possible? |
Category | Response

Categories: build, already possible, workaround exists, wrong product, not now
For "already possible" and "workaround exists", write the one-line reply to send.
Group requests that are the same need in different words, and say so explicitly.

Planned work, in order: [list the items and rough dates]
Where the product is going: [the thesis, in my words]
Audience for this narrative: [team, execs, customers]

Write the narrative, not the list:
- The one sentence that explains why this order and not another
- Three phases, each with what becomes true when it is done
- What we are deliberately not doing this period, and why that is a choice
- The assumption the whole plan rests on

Then say which item in my list does not fit the narrative.

What I am handing over: [area or project]
To: [who, and what they already know]
My last day on it: [date]

Write the handover:
- What this is and why it exists
- The current state, honestly, including the parts that are held together with tape
- Who to talk to for what, by name and topic
- The recurring things, with their cadence
- What will break first if nobody watches it
- The decisions I would make next, and why, so they can disagree knowingly

What they wrote:
[paste it]
What actually happened on our side: [the facts]
What I can offer: [the remedy I am authorised to give]
What I cannot: [the hard limits]

Write the reply.
- Acknowledge the specific problem in the first sentence, not their feelings in
  the abstract.
- State what happened plainly. No passive voice about our own actions.
- Give the remedy and the date.
- Name the limit once, without apologising for it repeatedly.
Under [150] words. No exclamation marks.

Issue that comes in constantly: [describe it]
The variants of it: [list how it differs case to case]
The correct resolution: [what we actually do]

Write [five] macros covering the variants, each under [80] words, each with the
variable parts marked in square brackets.

Then write the decision rule: given an incoming message, which macro to use, as a
short if-then list an agent can follow without thinking. Include the case where
none of them apply and it must be escalated.

Customer report:
[paste the ticket thread]
What I have already tried: [steps and results]
Impact: [how many users, how badly, since when]

Write the escalation for engineering:
- One-line summary
- Steps to reproduce, numbered, exactly
- Expected versus actual
- Environment and version details present in the thread
- What is already ruled out, with evidence
- What the customer has been told so far

Anything not in the thread is listed as MISSING, not filled in.

Customer said: [paste their words, however unclear]

Do two things.

1. The clarifying questions to send them, at most [four], ordered so the first
   answer might make the rest unnecessary. Written in plain language, no jargon.
2. A draft bug report with everything currently known, and explicit UNKNOWN
   markers where their answers will go.

Do not guess what they meant. If two readings of their message are both plausible,
list both.

Request: [what they asked for]
Policy: [the relevant rule, quoted]
Decision: [approved / partial / declined]
Why: [the actual reasoning]
Discretion available to me: [what I could do if I chose]

Write the reply.
- Decision in the first sentence.
- The reason in one sentence, referencing the policy without quoting it at length.
- If declined, state the one thing that would change the outcome, if anything would.
- No "unfortunately". No blaming the policy as though it were weather.

Resolved ticket:
[paste the thread including the resolution]

Write the help article:
- Title as the question the customer would search
- One-sentence answer at the top
- Steps, numbered, with what they should see after each
- The two things that commonly go wrong, and what to do
- When this article does not apply

Strip everything specific to this customer. If the resolution depended on
something an agent did that the user cannot do, say so and say who to contact.

Draft reply:
[paste it]
Customer situation: [what they are dealing with]
Relationship: [new / long-standing / previously unhappy]

Rewrite it to sound like a person, keeping every factual statement identical.

Rules
- Do not add sympathy that is not warranted by the situation.
- Remove corporate hedges: at this time, we are unable to, please be advised.
- Keep it the same length or shorter.
Then list every factual claim in your version, so I can check nothing drifted.

Incident: [what is broken]. Who is affected: [scope]. Started: [time]
What we know: [facts only]. What we do not: [be explicit]
Next update: [when]

Write three messages:
1. First notice, within the first [15] minutes: what is affected, what we are
   doing, when the next update comes
2. Mid-incident update: what has changed since, no speculation on cause
3. Resolution notice: what happened, what we did, what we are changing

No cause is stated in messages 1 or 2 unless it is confirmed.

Customer: [tenure, plan, usage]
Reason for leaving: [what they said]
What I know that they may not: [relevant facts, if any]
What I can offer: [the actual levers]

Write the conversation, not a template email:
- The question to ask before offering anything
- What to say if the reason is price, and it is genuinely too expensive for them
- What to say if the reason is a missing capability we do not have
- The point at which the right answer is to help them leave cleanly

Include that last part. Some customers should go.

What good support looks like here: [describe your standard]
Channel: [email, chat, phone]. Typical issue types: [list]

Build a scoring rubric with [five] dimensions, each scored 1 to 4, with a written
descriptor for every level, not just the top and bottom.

At least one dimension must measure whether the customer's actual problem was
solved, separately from how the reply sounded. Add the two things a scorer should
never penalise. Then show it applied to one example reply.

Open issues:
[paste list with status, age, and how many customers affected]

Write the digest for [audience: the support team / customers / leadership]:
- Grouped by what the customer experiences, not by internal component
- For each: what customers see, the workaround, the status, and what we are saying
- Sorted by number affected, not by internal priority
- One line at the top naming the single biggest source of contacts this week

Anything with no workaround is marked NO WORKAROUND in the group heading.

Topic: [topic]. Learners: [who, how many, what they already know]
Time: [minutes]. What they must be able to do afterwards: [the capability]
Materials available: [what you have]

Plan the session:
- Opening hook, [3] minutes, connected to something they already care about
- The core input, with the one idea it must land
- The practice activity, where they do it and can get it wrong safely
- The check: how you know, before they leave, whether it worked
- What to cut if you lose [10] minutes

Practice must take more time than input.

Topic I have been asked to teach: [topic]
Learners: [who]. Available time: [hours]
Assessment already fixed: [if any]

Write [four] to [six] learning objectives, each in the form: by the end, learners
will be able to [observable verb] [object] [under what conditions].

Rules
- No objective uses understand, appreciate, or be familiar with.
- Each must be checkable in the time available.
- Order them by dependency.
- Then list what the topic contains that these objectives deliberately exclude.

Task being assessed: [describe it]
What distinguishes excellent from adequate work here: [in your words]
Level of the learners: [level]

Build a rubric: [four] criteria, [four] levels each, with a descriptor per cell.

Rules
- Descriptors say what is present, not what is missing.
- No criterion rewards effort or length.
- One criterion must address the thing novices most often get wrong.
- Add two examples of work that would be hard to score, and how to score them.

Work submitted:
[paste it]
Level: [where the learner is]. Assignment brief: [what was asked]
What they struggled with last time: [if known]

Give feedback in this order:
1. The one thing that works, named specifically enough to repeat
2. The single most important thing to fix, with why it matters
3. The concrete next step, small enough to do today
4. One question that makes them think rather than correct

No more than one thing to fix. Do not rewrite their work for them.

Core task: [describe it]. Learning objective: [the objective]
Group range: [from what to what]

Produce three versions of the same task that all meet the objective:
- Scaffolded: with structure and a partially worked start
- Standard: as designed
- Extended: same objective, more open, harder to satisfy

They must be genuinely the same task, not easier and harder topics. Then give the
one question you would ask any learner, regardless of version, to check the
objective was met.

Material: [text, case, video]
Group: [who they are, how willing they are to talk]
Time: [minutes]. Goal of the discussion: [what should change in their thinking]

Write [eight] questions:
- Two that anyone can answer from the material, to get people talking
- Three that require a position, where reasonable people would differ
- Two that connect it to their own experience
- One that nobody in the room can settle, to end on

For each, add the follow-up to use when the first answer is one word.

Role: [title]. What they need to do independently by [timeframe]: [capabilities]
Who can teach what: [people and their areas]
What we currently do: [the existing onboarding, honestly]

Design the curriculum:
- Week by week, what they learn and from whom
- The first real task they own, and when
- What they should be allowed to break
- The check at each stage, and what happens if it fails
- What we can hand them as a document instead of a meeting

Concept to be practised: [concept]. Group size: [n]. Time: [minutes]
Room and materials: [what you have]
The mistake this exercise should surface: [the misconception]

Design the exercise:
- Setup and instructions, as you would say them aloud
- What participants do, minute by minute
- The moment where the misconception shows up
- The debrief questions, in order
- How to run it with half the time

Say what could go wrong in the room, and your response to each.

The mistake: [what they got wrong]
Why it matters: [the consequence]
Their situation: [confidence level, seniority, whether it is repeated]
Setting: [in front of others / privately / in writing]

Write what to say.
- Name the mistake plainly and early.
- Separate the mistake from the person in the wording.
- Explain the consequence once.
- Give them the corrected approach concretely.
- End with what you want to see next time, not with reassurance.

If the setting is in front of others, say whether it should wait.

Objective: [the learning objective]. Level: [level]. Format: [format]

Write [12] assessment items testing this objective at increasing difficulty.

Rules
- No item tests recall of wording rather than the objective.
- Every distractor in a multiple-choice item must be a real misconception, and you
  must say which one.
- Include two items where the correct answer is that the question cannot be
  answered with the information given.
- Provide the answer key with the reasoning, in a separate block.

Learner: [name or role]. Period: [dates]
Evidence I have: [work, attendance, participation, assessments]
Audience: [parent, manager, the learner themselves]
What I want to happen next: [the outcome]

Write the note, under [200] words:
- Where they are, against what standard, with one piece of evidence
- What has changed since last time
- The single thing to work on, and what support it needs
- What the reader can do

No adjectives that could apply to anyone. Every claim tied to something observed.

Below are transactions: date, description, amount.
[paste them]

Return a table: Date | Description | Amount | Category | Confidence | Recurring?

Categories: [your list, or say "propose a set and show it first"]

Rules
- Do not total anything. I will sum the column in a spreadsheet.
- Where a description is ambiguous, mark Confidence low and say what it could be.
- Flag anything that looks like a duplicate charge.
- Flag any recurring charge that appears to have increased.

I want to change how I handle [money situation].

Do not give advice yet. Ask me questions one at a time, up to [ten], covering what
comes in, what goes out, what is fixed, what I have already tried, what I am
avoiding thinking about, and what would count as success.

When you have enough, summarise my situation back to me in one paragraph and name
the single decision that matters most. Say clearly that you are not a financial
adviser and where a professional is warranted.

Option A: [describe it, with the numbers]
Option B: [describe it, with the numbers]
My situation: [what is relevant, including timeframe and risk tolerance]
Assumptions I am making: [list them]

Compare them:
- What each costs and returns under my stated assumptions
- Which assumption each option is most sensitive to
- What is true of A regardless of assumptions, and the same for B
- The scenario where each one clearly wins
- What I have not asked about that could change the answer

Do not compute compound figures. Give me the formula and I will run it.

Invoice [number], [amount], issued [date], due [date], now [n] days overdue.
Contact: [name and role]. Prior contact: [what I have already sent]
Relationship value: [do I want future work]

Write the next message, matched to how overdue it is:
- Under 14 days: assume oversight, keep it to three sentences
- 14 to 30: state the amount, the date and the terms
- Over 30: state what happens next and when, as a fact

Attach nothing new. Do not renegotiate the amount in this message.

Work: [what I am quoting for]. Client: [who]
My rate or pricing basis: [how I price]
Scope as discussed: [what was agreed]. Unknowns: [what is unclear]

Write the quote:
- Scope, itemised, in the client's language
- What is explicitly excluded
- Assumptions the price depends on, each with what happens if it is wrong
- Payment terms and schedule
- Validity period

Do not calculate the total. Leave [TOTAL] for me to fill in after checking the maths.

Here is my subscription list: name, amount, billing period, last used.
[paste it]

Return a table: Service | Annual cost | Last used | Verdict | Action

Verdicts: keep, downgrade, cancel, needs a decision from me
Rules
- Anything unused for [90] days defaults to cancel unless I have said otherwise.
- Group services that overlap in function and say which is redundant.
- Note anything that will auto-renew within [30] days.
- Do not total the column. Flag the three largest instead.

Policy document:
[paste the relevant sections]
Audience: [who has to follow it]
Questions they keep asking: [the common confusions]

Rewrite it as a plain-language guide:
- What you can do, as a short list
- What you cannot
- The grey areas, named as grey areas, with who to ask
- The three most common mistakes and their consequences
- What to do when it is unclear

Quote the original wording wherever the exact words matter legally.

The plan: [describe it, with the numbers and the timeframe]
What has to be true for it to work: [my assumptions]

Do not tell me whether the plan is good. Instead, run it through:
- Income arrives [30] days late every month
- The largest cost is [20]% higher than budgeted
- The best case and worst case for [the variable I am least sure about]
- One thing outside the plan going wrong at the worst moment

For each: what breaks first, how much warning I would get, and the cheapest
mitigation. State any figure you are unsure of rather than estimating it.

Change: [from what to what]. Effective: [date]. Who is affected: [segments]
Why: [the real reason]. What improves for them, if anything: [honestly]
Grandfathering: [what, if anything]

Write the notice:
- The new price and date in the first two sentences
- The reason once, without a paragraph of justification
- What existing customers specifically get or keep
- The options available to someone who does not want to pay it
- Where to reply

Do not describe the increase as an investment in them.

Money in, with expected dates: [list]
Money out, with due dates: [list]
Current balance: [amount]. Buffer I want to keep: [amount]

Build a watchlist, not a forecast:
- The dates where the balance is most likely to be tight, and what causes each
- Which incoming item is doing the most work, and what happens if it slips a week
- The outgoing items that could be moved, and by how long
- The one number to check every [week], and what value should worry me

Show the arithmetic you use so I can verify it. Do not present totals as final.

Destination: [where]. Dates: [when]. Travelling with: [who, and their limits]
Budget shape: [what matters, what does not]
Non-negotiables: [the things that must happen]
Pace preferred: [packed / slow / mixed]

Build a day-by-day outline: morning, afternoon, evening, with travel time between.

Rules
- Leave one unplanned block per day.
- Do not state opening hours, prices or booking requirements. Instead, list what I
  must verify before booking, per item.
- Flag any day where the travel time exceeds [two] hours in total.

Trip: [destination, duration, season]
Activities: [list]. Accommodation: [type, and what it provides]
Luggage limit: [what I am carrying]. Laundry available: [yes / no]
Things I always forget: [list]

Produce the list grouped by: worn, carry-on, checked, and documents.

Rules
- Quantities, not just items, based on the duration and laundry answer.
- A separate short list of what I can buy there instead of carrying.
- The five items that would ruin the trip if forgotten, marked clearly.
- Do not advise on climate specifics you cannot verify. Tell me to check.

Looking for: [restaurants / activities / neighbourhoods] in [place]
Who is going: [group and their constraints]
What would make something wrong for us: [dealbreakers]
Budget per person: [amount]

Before recommending anything, state the criteria you will use to reject options,
based on my dealbreakers.

Then give [eight] candidates with, for each, why it fits and what would make it a
mistake. Mark anything you are not confident still exists or still operates, and
tell me to verify it.

Journey: [route, connections, times]
Tightest connection: [where and how long]
Consequences of missing it: [what happens]
What I am carrying: [bags, and whether they are checked]

Build the contingency plan:
- The three most likely failure points, in order
- For each: the sign that it is happening, the decision point, and the alternative
- What to have to hand before leaving, and where
- The one thing to sort out before departure that would prevent the worst case

Do not include phone numbers or reference details you cannot know.

Group: [how many, and the relevant differences between them]
Deciding: [what]. Constraints: [budget, dates, must-haves per person]
What has already caused friction: [the honest answer]

Facilitate the decision:
- The questions to put to everyone individually, before any group discussion
- How to collapse the answers into [three] options
- The trade-off each option makes, stated so nobody is the villain
- The decision rule to agree on before revealing preferences
- What to do when one person is the outlier

Do not pick for us.

Event: [what]. Date and duration: [when]. Attendees: [how many, who]
Venue: [describe it]. Team: [who is helping and what they can do]
The moment it all exists for: [the thing that must go well]

Build the run of show as a table: Time | What happens | Who runs it | What must be
ready | What to do if it slips

Work backwards from the key moment. Include setup, the first fifteen minutes,
and the pack-down. Flag every point where one person is a single point of failure.

Event: [what, when, where]. Guests: [who, how many, how well they know us]
What they need to know: [logistics]
What they will get wrong if we do not say it: [the predictable confusions]

Write the sequence:
1. The save-the-date or invitation
2. The details message, [one week] out, with everything they need to act on
3. The day-before message, under [60] words, only what changes their behaviour
4. The thank-you afterwards

Each message must be readable on a phone lock screen for its first line.

Options:
[paste what I know about each: capacity, cost, location, what is included]
Event: [what it is for]. Non-negotiables: [list]

Build the comparison table with my non-negotiables as the first rows, then cost,
then everything else.

Rules
- Only include facts I have supplied. Anything else is a row marked NEED TO ASK.
- Add a row for what each venue would make difficult, inferred from what I gave you
  and marked as an inference.
- End with the five questions to ask all three before deciding.

Trip: [passport nationality, destination, purpose, dates, route including transits]
Travelling with: [others, and their nationalities if different]

Produce a checklist of the document categories I need to confirm: entry permission,
passport validity rules, onward travel evidence, insurance, health documentation,
and anything specific to transiting.

For each, state what to check and the official source type to check it against.
Do not state any requirement as fact. Rules change and vary by nationality, so this
is a list of what to verify, with the government or airline as the authority.

Trip: [where and when]. Purpose: [work / personal]
Expenses: [paste the list with dates, amounts and currencies]
What I need to produce: [expense claim, report, notes for next time]

Do three things:
1. Group the expenses by category and flag anything missing a receipt or a date
2. List what needs converting, with the rate I should look up, without converting it
3. Write the short trip note: what worked, what I would do differently, what to
   book earlier next time

Do not calculate any total or conversion.

Premise: [describe it in a few sentences]
Genre and length: [what it is]. What I think it is about: [the theme]

Do not develop it. Test it:
- What is the actual conflict, and is it in the premise or only implied
- What has to be true about the protagonist for this to be their story
- The version of this premise the reader has already seen, and how mine differs
- Where it will sag, structurally
- The question the ending has to answer

If the premise is a situation rather than a story, say so and tell me what is missing.

Character: [everything I know so far, however thin]
Story context: [what happens to them]

Interview them. Ask one question at a time and answer in their voice, staying in
character. Start with what they want and move toward what they will not admit.

After [ten] exchanges, break character and tell me: what is now specific about this
character, what is still generic, and the contradiction in them that is worth
building on. Flag anything I invented on the spot that contradicts what I told you.

Scene:
[paste it]
Currently in: [whose point of view]
Rewrite in: [whose point of view]

Rewrite it so that:
- The events are identical
- What this character notices is different, and revealing
- Something the original point of view stated is now unknown
- One thing the reader learns that the previous version could not show

Then say in two lines what the story gains and loses in this version.

World: [the concept]. Story that happens in it: [what I am writing]
The one rule that makes it different: [the central conceit]

Do not describe the world. Derive its constraints:
- Three consequences of the central rule that follow logically
- What it makes impossible that we take for granted
- Who benefits from it and who suffers, and how they organise
- The thing everyone in this world argues about
- Where my conceit contradicts itself, if it does

Keep everything downstream of the one rule. Add nothing decorative.

Scene with dialogue:
[paste it]
Characters and what each wants in this scene: [list]

Rewrite the dialogue only. Keep the beats and the outcome.

Rules
- Nobody says exactly what they mean.
- Cut every line that only conveys information the reader already has.
- Give each character one verbal habit and use it twice, not five times.
- Remove dialogue tags that describe emotion. Let the lines do it.
Then list the three lines you would cut entirely if forced.

Subject: [what it is about]. Form: [free verse, sonnet, chorus-verse, other]
The feeling it should leave: [in one word]
The image I keep coming back to: [the image]

Do not write it. Build the scaffold:
- The structure, section by section, with what each does emotionally
- Where the turn goes and what turns
- Which lines carry the weight, marked as placeholders
- Three ways the central image could develop across the piece
- The ending's job, and the two obvious endings to avoid

I will write the lines.

Mechanic: [describe how it works]
Game: [genre, length, who plays it]
What it is meant to make players feel: [the intent]

Critique it:
- What players will actually do with it, including the boring optimal strategy
- Where it stops being interesting, and after how long
- What it teaches players that the rest of the game contradicts
- The smallest change that fixes the biggest problem
- What would be lost by removing the mechanic entirely

Assume players will break it. Tell me how.

Naming: [what it is]
What it should suggest: [the associations]
What it must not suggest: [the wrong associations]
Constraints: [length, language, pronounceability, availability]

First, write the rejection rules you will apply. Then give [20] candidates
grouped by approach: literal, compound, borrowed, invented, oblique.

For each, one line on what it suggests and one risk. Mark any that are likely to
be already taken or hard to say aloud. Do not check availability; you cannot.

Project: [what I want to make]
Time I genuinely have: [hours per week, honestly]
Skills I have: [what] and lack: [what]
Why previous attempts stalled: [the honest reason]

Plan it around the stalling reason, not around the ideal path:
- The smallest version that is still worth having
- The first session's work, specific enough to start tonight
- Where this project will stall, based on what I told you
- The checkpoint before that point, and what to do there
- What to cut when I fall behind, decided now

Recipe:
[paste it]
What I need to change: [dietary need, missing ingredient, equipment, servings]
What I have: [ingredients and equipment]
What matters most in the result: [texture, flavour, time, cost]

Give me the adapted version, and for each change:
- What you substituted and in what ratio
- What it will do to the result, honestly
- The step that needs adjusting because of it

Flag anything where the substitution genuinely will not work and say so instead
of pretending. Do not adjust anything involving food safety without saying to check it.

Recipient: [who they are, what they are into, what they already have]
Occasion: [what]. Budget: [hard limit]
What they would hate: [the wrong kind of gift]
What I gave them last time: [if applicable]

Give [12] ideas across four bands: under a quarter of the budget, half, most of
it, and one that needs a group.

For each: why it fits this person specifically, and the risk it does not land.
No experiences that require me to know their calendar. Nothing that needs sizing.

I am stuck on [the project]. Here is where it stands:
[describe the current state]
Here is what I have tried: [list]

Do not encourage me. Ask questions one at a time until you can name what is
actually blocking me, which is usually one of: I do not know what happens next, I
know and do not like it, I am avoiding a decision, or the thing is wrong further
back than where I am stuck.

When you can name it, say which one, give the evidence from my answers, and
propose the single smallest next action.