100+ ChatGPT Prompt Templates for Every Use Case (Free Library, 2026)

TL;DR: This is a free library of 100+ tested ChatGPT prompt templates spanning marketing, code, research, creative writing, hiring, and operations. Every one is CRAFT-formatted, copy-paste ready, and built around [bracketed variables] you fill in. Bookmark this page, grab what you need, and ship.

What are the best ChatGPT prompt templates for every use case?

The best ChatGPT prompt templates are structured prompts that specify role, context, task, output format, and tone instead of leaving the model to guess. The 100+ templates below cover marketing, code, research, creative, hiring, and operations. Each one is copy-paste ready, uses fill-in [variables], and transfers to Claude and Gemini with minor edits. Pick a category, fill the brackets, and run.

That single idea — structure over improvisation — is why a template library beats typing whatever comes to mind. ChatGPT now reaches 900 million weekly active users as of February 2026, more than double the 400 million reported a year earlier, and the overwhelming majority of those people type free-form requests and accept whatever comes back. A small, deliberate library puts you in the top slice that gets consistent, usable output on the first try.

This guide is long on purpose. Skim the table of contents, jump to your category, and treat the rest as reference. If you only do one thing, save your ten most-used templates somewhere you can reach them in two seconds.

Why do structured templates beat free-form prompts?

Because large language models are extraordinarily sensitive to how you phrase and format a request. The research here is unusually consistent: structured, context-rich prompts produce higher productivity and lower misinterpretation than vague ones. A 2025 study on prompt engineering and LLM productivity found that users who write clearer, more structured, and context-specific prompts get better results and fewer misreads from the model.

Formatting alone moves the needle. Industry analysis of prompt-engineering research reports meaningful accuracy swings purely from how a prompt is structured, with structured layouts such as labeled sections and tags outperforming plain text. The model is not reading your mind; it is pattern-matching on the shape of your request. Give it a clear shape and it returns a clear answer.

There is a nuance worth knowing. High-capability models like GPT-5 tolerate vague prompts better than older models did, while smaller and cheaper models benefit more from extra reasoning scaffolding. OpenAI's own guidance stresses that prompts should be clear, specific, and provide enough context without ambiguity. Templates bake that discipline in so you do not have to remember it every time.

Here is the difference at a glance.

How do you use this prompt library?

Each template uses CRAFT (Context, Role, Action, Format, Tone) or RTF (Role, Task, Format). You replace [bracketed variables] with your specifics and run. That is the whole workflow.

For repeated use, save the templates into a prompt manager so you are not hunting through a doc every time. A dedicated tool lets you store a prompt once, drop in placeholders, and reuse it across every platform you work in. If you want the templates as one-click presets with shared variables, that is exactly what Prompt Architects was built for — but a Notion page or a snippet expander works fine to start.

A few ground rules that apply to every template below:

• Fill variables tightly. [Senior copywriter, 10y B2B SaaS] beats [copywriter]. Specific inputs produce specific outputs.
• Run, read, refine. Treat the first output as a draft. Tighten one variable, run again. Do not rewrite from scratch.
• Chain, don't cram. Feed the output of one template into the next rather than stacking three tasks in one prompt.
• Edit everything. Templates do the structure work. Accuracy, voice, and final judgment are yours.

If you are new to frameworks, our prompt engineering frameworks guide walks through CRAFT, RTF, and when to reach for each.

Marketing and copy prompt templates (20)

Marketing is where templates pay off fastest because you run the same shapes — headlines, emails, ads — over and over. Fill the variables and generate ten variants in the time it took to write one by hand.

1. Headline generator

Role: Senior copywriter, 10y B2B SaaS experience.
Task: Generate 10 headline variants for [product] targeting [audience].
Constraints: <= 8 words each. Mix benefit (3), curiosity (3), problem-agitate (4).
Format: Numbered list with 1-line rationale per headline. Rank top 3 by predicted CTR.
Tone: Confident, specific, no buzzwords.

2. Subject line A/B

Email purpose: [purpose]. Audience: [segment].
Generate 30 subject lines: 10 curiosity, 10 benefit, 10 question.
Constraint: <= 50 chars, no spam triggers.
Rank top 5 by predicted open rate.

3. Cold outreach

Recipient: [name + company + role]. Sender: [your role + offer].
Goal: 15-min call. Write 3 cold email variants <= 90 words each.
Open with recipient-specific observation, single CTA, value-add PS.

4. Landing page hero

Product: [product]. ICP: [audience]. Offer: [offer].
Generate 10 hero headline variants <= 12 words.
Mix: benefit (4), problem-agitate (3), curiosity (3).
Rank top 3 by predicted conversion. For top 3 add matching subhead.

5. Ad copy (Meta)

Product: [product]. Audience: [audience]. Funnel: [TOFU/MOFU/BOFU].
Write 3 Meta ad variants: primary text <= 125 chars,
headline <= 27 chars, description <= 27 chars.
Plus 3-line image brief per variant.

6. Ad copy (Google RSA)

Keyword: [keyword]. Product: [product]. USP: [USP].
Generate 5 RSA groups: 3 headlines x 30 chars + 2 descriptions x 90 chars.
Variants: pain, benefit, social proof, price, FOMO.

7. LinkedIn post (operator voice)

Topic: [topic]. Audience: [B2B founders / CMOs].
Write 200-word LinkedIn post: provocative claim, 3 supporting points,
1 contrarian observation, soft conclusion.
Tone: opinion-driven, not corporate.

8. Twitter thread

Topic: [topic]. Hook angle: [contrarian/data/personal].
Write 10-tweet thread: tweet 1 hook <= 200 chars,
tweets 2-9 one insight each, tweet 10 synthesis + soft CTA.
Voice: confident, specific, no buzzwords.

9. TikTok hooks

Topic: [topic]. Generate 10 first-3-second hooks.
<= 12 words. Tag each: question, contrarian, list, problem-agitate, curiosity.

10. Instagram caption

Image description: [describe]. Brand voice: [voice].
Generate 5 captions: 1 short (<=50 words), 2 medium (100), 2 longer (150).
Include 5 niche + 5 medium + 5 broad hashtag suggestions.

11. Press release

News: [news]. Company: [company].
Write 400-word press release: dateline, headline <= 8 words,
subhead, lede, 2 supporting paragraphs, exec quote 40-60 words, boilerplate.

12. Case study

Customer: [name + size + industry]. Outcome: [metric].
Write 1500-word case study: hook (50w), problem (200), approach (300),
implementation (400), results (300), takeaway (100), pull-quote.

13. About page

Founder: [name + bg]. Company: [name + product + year].
Mission: [mission].
Write 400-word About: founding story, problem, insight, solution, vision.
Tone: human, specific, no MBA-speak.

14. FAQ generator

Product: [product]. Audience: [audience].
Generate 12 FAQs grouped: pricing (3), features (3), setup (3), trust (3).
Each Q natural-language, A 40-80 words covering objection.

15. Sales objection handler

Objection: [objection]. Product: [product]. Audience: [audience].
Write 3 response variants. Each: acknowledge, reframe, evidence, soft close.
80 words max.

16. Welcome email sequence

Product: [product]. ICP pain: [pain].
Write 5 emails: D0 confirmation, D1 quick win, D3 deeper feature,
D5 social proof, D7 conversion.
Each: subject + 80-150w body + 1 CTA.
Tone: helpful, specific, not pushy.

17. Re-engagement email

Audience: subscribers inactive 90+ days.
Write 3-email sequence: E1 we miss you, E2 here's what changed,
E3 last chance. Each: subject + body + CTA.

18. Newsletter section

Updates: [paste 3 updates]. Audience: existing customers.
80-word intro, 3 update blocks (40w each, 1 link), CTA.
Tone: informational, not pushy.

19. Brand voice document

Brand: [brand]. Founder samples: [paste 3 examples].
Distill: 5-7 voice attributes, 3 phrases we always say,
5 phrases we never say, tone matrix per channel (web/email/social/support).

20. Content calendar

Niche: [niche]. Pillars: [3-5 pillars].
Generate 12-piece calendar (3/week x 4 weeks).
Table: week, pillar, format, working title, target keyword, intent.

Tip: for any marketing prompt you run weekly, store the static parts as a Global Variable — your brand voice, ICP, and product description — so you only ever change the task line.

Code and engineering prompt templates (20)

Engineering prompts reward two patterns above all: explicit reasoning and rigid output schemas. When you ask the model to walk through its logic step by step, accuracy climbs — the original chain-of-thought research showed that prompting a model to reason before answering dramatically improved performance on math and reasoning benchmarks like GSM8K, and follow-up work confirms CoT helps most on math and symbolic reasoning tasks. Templates 23–27 and 38–40 lean on that explicitly.

21. Function from spec

Language: [TS]. Task: [description]. Constraints: [pure / no deps].
Inputs: [type sig]. Outputs: [type sig]. Edge cases: [list].
Walk through: algorithm -> edge cases -> complexity -> impl -> 5 unit tests.

22. CRUD endpoint

Stack: [Next + Drizzle + Postgres]. Resource: [name + 5 fields].
Generate full CRUD: GET list, GET [id], POST, PATCH, DELETE.
Include: Zod schema, error handling, auth check, pagination.

23. SQL query optimizer

Query: [paste]. Schema: [paste]. Sample: [N rows].
Step by step: what does it do -> where's the cost ->
what indexes help -> can it be rewritten -> final + reasoning.

24. Migration writer

Schema change: [description]. DB: [Postgres].
Write idempotent migration: DDL + data backfill + rollback + verification.
Walk through: large table size, locks, concurrent writes.

25. Stack trace parser

Stack trace: [paste]. Code: [paste].
Walk through: which line throws -> state that caused it ->
immediate cause vs symptom -> top 3 hypotheses -> fix + test.

26. Chain-of-Thought debug

Code produces [bug]: [paste].
Walk execution step by step. Where does behavior diverge?
Root cause? Minimal fix? Output corrected code with comment at change site.

27. Refactor for testability

Code: [paste].
Refactor without changing behavior: extract pure functions,
inject dependencies, reduce arity. Walk reasoning per change.
Output refactored code + 5 unit tests.

28. Code review with criteria

Diff: [paste]. Senior reviewer.
4 dimensions: correctness, performance, security, maintainability.
For each: comments under H3. Severity: blocker/suggestion/nit.

29. Test coverage gap

Code: [paste]. Existing tests: [paste].
Identify untested branches/edge cases. For each gap:
test name, input, expected output, why it matters.
Suggest 5 highest-leverage tests sorted by impact.

30. Security review

Code: [paste].
Trust boundary? User-controllable input? Auth check correct?
Common vulns: SQLi/XSS/SSRF/CSRF/IDOR -- applicable?
Output severity-tiered findings.

31. JSON entity extractor

{
  "task": "extract_entities",
  "input": "[text]",
  "output_schema": {
    "name": "string",
    "company": "string",
    "topic": "string",
    "urgency": "low | normal | high"
  }
}
Respond as JSON matching schema. No prose, no code fences.

32. PR description from diff

{
  "task": "summarize_pr",
  "input": "[diff]",
  "output_schema": {
    "title": "string <= 70 chars conventional commit",
    "summary": "string 3-5 sentences why over what",
    "test_plan": ["string"],
    "breaking_changes": "boolean",
    "migration_steps": "string | null"
  }
}

33. Issue triage

Issue body: [paste].
Output JSON: type (bug/feature/docs/question), severity (p0-p3),
reproducible (bool), missing_info (array), suggested_label (array),
first_response (<=100w in maintainer voice).

34. README writer

Project: [name]. Does: [1-line]. Stack: [list].
Generate README: hero (badges + 1-line), quick start (3-step),
usage (3 cases with code), config (env vars table), contributing, license.

35. API doc from code

Code: [handler/function].
Generate: endpoint, method, auth, request schema (with examples),
response schema (with examples), error codes, rate limits, idempotency notes.
Format: markdown H3 sections.

36. Migration guide v1->v2

Breaking changes: [list].
Generate migration guide: summary table (changed/why/severity),
per-change before/after code, mechanical steps, validation, rollback, FAQ.

37. Performance profiler interpreter

Profile output: [paste].
Identify top 3 hot spots ranked by self-time.
For each: what's hot, why, suggested optimization, expected lift.

38. Convert callback to async

Code: [callback chain].
Convert to async/await preserving behavior. Map each callback to
awaited promise. Handle errors. Output refactored + comments at non-trivial.

39. Reduce N+1 query

Code: [N+1 pattern].
Identify N+1 site. Refactor to single query or batched.
Tradeoffs: eager vs join vs DataLoader. Output refactored + benchmark expectation.

40. Flaky test diagnostician

Test: [paste]. Failure pattern: [intermittent/env/time].
What does test assert? What state makes it pass sometimes fail others?
Top 5 flake categories: timing, ordering, fixtures, network, env.
Most likely + fix that eliminates flake source.

Two notes for engineers. First, JSON-schema prompts (31–33) are the most reliable way to get parseable output you can pipe into a script — define the schema, demand JSON only, and the model stays on rails. Second, paste real code, not a paraphrase. The model reasons over what you give it; a vague description produces a vague fix.

Research and analysis prompt templates (15)

Research prompts turn raw input — transcripts, reviews, analytics, SERPs — into structured insight. The trick is to demand a specific output shape (a table, a ranked list) and to ask the model to flag where the data is thin rather than confidently inventing.

41. Customer interview synthesizer

Transcript: [paste].
Extract: top 3 pains (with quote), top 3 outcomes,
jargon-free language, competitive products mentioned, 3 follow-up Qs.
Output as table.

42. Multi-interview pattern

5 interviews: [paste].
Cross-interview: pains by >=3 (with frequency), language patterns,
contradictions, latent needs no one named directly.

43. Survey design

Goal: [research Q]. Audience: [audience].
Design 10-question survey: 1 NPS, 3 multi-choice, 3 Likert, 3 open.
For each: question + answer options + insight produced.

44. Competitive teardown

Competitor: [URL]. Our positioning: [paste].
Audit: positioning headline, top 3 features, pricing, social proof, CTA.
Output 5 differentiation gaps + 3 opportunities.

45. ICP refiner

Current ICP: [paste]. Interview data: [3 transcripts].
Refine: firmographic, psychographic, trigger event.
Flag uncertainty where data thin.

46. Voice-of-customer extraction

10 reviews/tickets: [paste].
Extract verbatim phrases customers actually use.
Group by theme. Mark which work for: hero, ad, feature names, FAQ.

47. Analytics insight extractor

Data: [table].
Generate 5 insights non-obvious to skim.
Each: insight + supporting numbers + recommended action.
Flag insights needing more data to confirm.

48. Keyword cannibalization audit

URL list: [paste].
Identify pages targeting overlapping keywords.
Table: keyword, conflicting URLs, recommended action (consolidate/differentiate/canonical).

49. Article research synthesizer

5 article URLs/abstracts: [paste].
Cross-article: shared findings, contradictions, methodology differences,
gaps for further research.

50. Topic cluster generator

Niche: [niche]. Pillar keyword: [keyword].
Generate 12 cluster topics. Table: topic, intent, target keyword,
volume estimate, why it ranks.

51. Content gap analysis

Competitors: [URLs]. Our domain: [URL].
10 keywords competitors rank for that we don't.
Table: keyword, intent, why target, priority (high/med/low).

52. Keyword intent classifier

50 keywords: [paste].
Classify each: informational, navigational, commercial, transactional.
Output as table.

53. People Also Ask expansion

Topic: [topic].
Generate 15 PAA-style questions. Group by intent:
definitional, comparative, troubleshooting, how-to, why.

54. SERP feature targeting

Keyword: [keyword].
List dominant SERP features (snippet, PAA, video, image, etc.).
Per feature: recommended content format that targets it.

55. Survey response synthesizer

50 responses: [paste].
Synthesize: top patterns, contradictions, surprising findings,
3 recommended actions based on data.

For anything analytical, add the instruction "cite the specific number behind each claim" to the prompt. It forces the model to stay grounded in your data instead of drifting into plausible-sounding generalities. If your research feeds long-form content, our GEO and AEO content guide covers how to structure those findings so AI search engines actually cite you.

Hiring and people prompt templates (10)

Hiring prompts are where consistency matters most, because inconsistent evaluation is how bias creeps in. A shared rubric, applied the same way to every candidate, is the antidote — and these templates build that rubric for you.

56. Role rubric

Role: [Senior X at startup]. Level: [IC4 equiv].
Rubric: 5 dimensions (technical depth, scope, communication, cultural add, growth).
For each: 5-point scale with concrete behavioral indicators.

57. Interview questions

Rubric: [paste].
12 calibrated questions covering all 5 dimensions:
4 technical depth, 2 scope, 2 communication, 2 cultural add, 2 growth.
Each: question + what answer reveals + red flags + green flags.

58. Take-home eval rubric

Assignment: [paste]. Generate rubric: 4 weighted dimensions, 1-5 each.
Include flag (instant DQ) and wow (auto-advance) criteria.

59. Post-interview synthesizer

Notes: [raw paste].
Extract: strengths (with evidence), concerns (with evidence),
open questions, recommendation (advance/next/pass), confidence (H/M/L).

60. Reference call questions

Role: [role]. Strength to validate: [strength]. Concern to probe: [concern].
8 reference Qs: 4 open, 2 situational, 2 calibration.
Each: hope-to-hear vs yellow flag.

61. Job description writer

Role: [role]. Company: [company]. Location: [remote/city].
Generate JD: 1-line hook, what you'll do (5 bullets),
who you are (5 bullets), nice-to-haves (3), benefits, application instructions.

62. Performance review draft

Direct report: [role]. Period: [Q].
Wins: [list]. Stuck points: [list]. Career goal: [goal].
Draft 350-word review: strengths, growth areas, calibrated rating, dev plan.
Tone: candid + supportive.

63. 1:1 agenda

Direct report: [role]. Tenure: [N months].
Recent: [context]. 1:1 agenda: 3 their-topics prompts,
2 calibration questions, 1 stretch goal check, 5 min wrap.

64. Team retrospective

Sprint goal: [goal]. Outcome: [outcome].
Generate retro: what went well, what didn't, what to change,
3 action items with owners + due dates.

65. Onboarding plan

Role: [role]. New hire: [name + level].
30-60-90 day plan: each milestone has 3 outcomes + 5 tasks.
Pair with named people for handoffs.

A caution specific to people work: never paste anything you would not be comfortable having processed by a third party, and always apply your own judgment to people decisions. The model drafts the rubric; a human owns the call.

Creative and writing prompt templates (15)

Creative prompts need the opposite of analytical ones: more room to roam, but anchored by constraints. Specify form, length, and mood, then let the model generate variants you can react to. Generating three and picking one beats agonizing over a blank page.

66. Story synopsis

Genre: [genre]. Length: [N pages].
Generate 1-page synopsis: protagonist + want + obstacle,
inciting incident, midpoint reversal, climax, resolution. Tone: agent-pitch ready.

67. Character bio

Story: [premise]. Character role: [protagonist/antagonist/etc.].
Generate bio: name, age, background (3-line), want, internal conflict,
external conflict, voice attributes (5), distinguishing detail.

68. Scene draft

Setting: [setting]. Characters: [list]. Stakes: [stakes].
Action goal: [what happens].
Draft 800-word scene with dialogue, beats, sensory detail. Show don't tell.

69. Plot hole finder

Manuscript section: [paste].
Identify: contradictions with established facts, unmotivated character actions,
unresolved setups, timeline issues. Output as numbered list.

70. Critique partner read

Manuscript: [paste].
Read as senior editor. Output: what works, what doesn't,
biggest line-edit opportunity, structural concerns.
Severity tiered. No empty praise.

71. Pitch letter

Manuscript: [premise]. Genre: [genre]. Comp titles: [titles].
Write 200-word pitch: hook line, comparison, premise, stakes, sample.
Tone: agent query ready.

72. Naming brainstorm

What it is: [1-line]. Audience: [audience].
Constraints: [<=2 syllables / .com avail / no negative connotations EN/ES/FR].
Generate 30 candidates. Each: meaning, why it could fit, why miss.

73. Slogan generator

Brand: [brand]. Promise: [promise]. Audience: [audience].
Generate 20 slogans. Mix: benefit (5), aspirational (5),
contrarian (5), playful (5). Rank top 5.

74. Speech writer

Speaker: [role + audience]. Occasion: [event]. Length: [N min].
Goal: [persuade/inform/inspire].
Draft full speech: hook, 3 main points (each with story + data),
emotional close.

75. Wedding toast

Speaker: [relationship to couple]. Couple: [names + 1-line story].
Tone: warm + funny + sincere.
Draft 2-min toast: opening anecdote, theme that connects,
specific moment, blessing.

76. Eulogy

Person: [relationship + biographical sketch].
Tone: reverent + warm + personal.
Draft 4-min eulogy: opening memory, 3 character traits with stories,
how they shaped lives, closing reflection.

77. Children's story

Audience: [age]. Length: [N words]. Theme: [theme].
Generate story: 3-act structure, simple vocabulary, vivid imagery,
satisfying ending, repeating phrase for child engagement.

78. Poem

Form: [haiku / sonnet / free verse]. Subject: [subject]. Mood: [mood].
Constraint: [length].
Generate 3 variants. Each: tighter than the last.

79. Joke writer

Topic: [topic]. Type: [setup-punchline / observational / one-liner].
Generate 10. Mix clean and edgy. Mark which work for: stand-up, twitter, dinner.

80. Tagline rewriter

Current tagline: [paste]. Goals: [emotional response].
Generate 15 rewrites. Variants: shorter (5), more specific (5), more emotional (5).
Rank by predicted impact.

For personal pieces like toasts and eulogies, the most important variable is the specific detail. Feed the model two or three real moments and it will weave them in; give it generic input and you will get a greeting-card draft. The specificity you bring is what makes the output sound like you.

Operations and personal prompt templates (20)

Operations prompts turn the messy raw material of work — Slack threads, calendars, inboxes, meeting notes — into decisions and next steps. These are the templates that quietly reclaim an hour a day once they become habit.

81. Vendor decision matrix

Decision: [pick vendor for X]. Options: [3-5].
Criteria: [5-7 weighted criteria].
Score each 1-5 per criterion. Compute weighted total.
Flag tie-breakers + dealbreakers. Recommend choice.

82. Async update

Slack last 24h: [paste].
Synthesize 100-word update: decisions made, open questions,
action items (with owner), things needing my input.

83. Email triage

20 inbox subject + 1-line summaries: [paste].
Categorize: respond now (<=5), later (<=10), archive (<=5).
For 'now', draft 1-line response.

84. Weekly retrospective

Wins: [list]. Stuck: [list].
Synthesize: top stuck pattern, single highest-leverage change,
one thing to stop, one to keep doing more.

85. Personal OKR draft

Quarter: [Q]. Role: [role]. Top constraint: [blocker].
Draft 3 OKRs: objective (qualitative), 3 measurable KRs, 1 leading indicator.

86. Meeting agenda

Meeting purpose: [goal]. Attendees: [N]. Duration: [min].
Generate agenda: 3 outcomes, time-boxed sections,
prep required per person, decision points, next-step capture.

87. Meeting summary

Notes: [paste].
Summarize: decisions made, action items (owner + due date),
unresolved questions, follow-up agenda items.

88. Calendar audit

Last week's calendar: [paste blocks].
Identify: meeting types, time spent per category,
energy mismatch (high-effort meetings in low-energy slots),
suggestion for next week's blocks.

89. Daily plan from goals

This week's 3 goals: [list]. Today's energy: [high/med/low].
Allocate today's hours: top-3 deep work blocks,
2 meeting blocks, 1 admin block. Justify each allocation.

90. Decision journal

Decision: [decision]. Options: [list]. Picked: [choice]. Reason: [reason].
Record: assumptions made, evidence supporting, evidence against,
review date (+90 days).

91. Apology email

Mistake: [what happened]. Recipient: [audience].
Write 150w apology: acknowledge clearly, take ownership,
explain (don't excuse), commit to fix, no defensiveness.

92. Tough feedback

Issue: [issue]. Recipient: [role + relationship].
Write feedback message <= 200w: specific behavior, impact,
change request, support offer. Tone: candid + caring.

93. Resignation letter

Role: [role]. Tenure: [N years]. Reason: [moving to / family / etc.].
Write 150w resignation: gratitude, transition support, last day, no burning bridges.

94. Negotiation prep

Negotiation: [salary/contract/etc.]. My ask: [ask]. Their position: [position].
Walk through: BATNA, ZOPA, anchoring strategy, top 3 questions to ask,
concession ladder, walk-away criteria.

95. Travel itinerary

Destination: [city]. Days: [N]. Vibe: [foodie/culture/relaxed].
Travelers: [N]. Generate day-by-day itinerary:
3 anchors per day, walking distances realistic, restaurant per day, contingency.

96. Learning roadmap

Goal: [skill]. Current level: [level]. Time available: [hr/week].
6-week roadmap: weekly milestone, resources (book/course/practice),
weekly project to ship.

97. Book summary

Book: [title].
Generate 200-word summary: thesis, 3 key arguments,
strongest evidence, what's wrong with it, who should read.

98. Recipe scaler

Recipe: [paste]. Original servings: [N]. Target: [M].
Scale ingredients (handle 'to taste' notes).
Adjust cook time + temp where scaling affects.
Flag any 'don't scale' items (yeast, leavening).

99. Workout plan

Goal: [strength/endurance/etc.]. Days: [N/week]. Equipment: [list].
Experience: [level].
Generate 4-week plan: per-day exercises, sets x reps,
progression scheme, deload week.

100. Habit tracker

Habit: [habit]. Why it matters: [reason].
Design tracking system: daily question (yes/no),
weekly metric, monthly review prompt, failure-recovery rule.

101. Birthday message

Recipient: [name + relationship + 2-3 facts about them].
Write warm 80-word message. Specific to them, not generic.

102. Anniversary note

Years together: [N]. Specific moments: [list 3].
Write 150-word note. Specific, not greeting card. Show, don't tell.

A reminder on the personal and health-adjacent templates (95–99): treat output as a starting point, not advice. A workout or travel plan from a model is a draft to sanity-check, not a prescription to follow blindly.

Which prompt framework should you use — CRAFT, RTF, or Chain-of-Thought?

It depends on the task. The three frameworks in this library each fit a different job, and knowing which to reach for is half the battle.

The rule of thumb: use CRAFT when the output is for humans and tone matters, RTF when you just need a fast repeatable answer, Chain-of-Thought when the problem has steps the model could get wrong, and a JSON schema when a script will consume the output. For a deeper comparison, see our ChatGPT vs Claude vs Gemini guide on how each model responds to these frameworks.

How do you make any template better? Five power moves

Templates get you 80% of the way. These five habits get the rest.

1. Save your top 10 as reusable presets. Any prompt manager handles this. Prompt Architects ships them as one-click presets across eight platforms, so the same template works in ChatGPT, Claude, and Gemini without re-pasting.
2. Add one to three examples to repeated patterns. This is few-shot prompting, and it is one of the highest-leverage moves you can make. Showing the model a worked example of your preferred format lifts consistency — a recent study found CoT-style few-shot examples improved accuracy across every model tested. Examples halve rework on the tasks you run most.
3. Specify variables tightly. [Senior copywriter, 10y B2B SaaS] beats [copywriter] every single time. The model fills ambiguity with averages; specificity steers it.
4. Chain prompts instead of cramming. Feed the output of one template into the next. Three tasks in one prompt produces a muddled answer to all three.
5. Iterate by tightening one variable per attempt. Do not rewrite from scratch when the output misses. Change the single variable that caused the miss and run again. OpenAI's own guidance frames prompt engineering as exactly this kind of iterative refinement loop.

What changed in 2025–2026 for ChatGPT prompting?

The ground shifted in a few important ways, and the templates above already account for them.

• Frontier models got more forgiving. GPT-5 and Claude Opus 4 handle vague prompts better than their predecessors. Frameworks still win for production-quality and repeatable work, but the gap narrowed for casual, one-off use. The catch: forgiving does not mean predictable. Structure still buys you consistency.
• JSON prompting went mainstream. Defining an explicit output schema is now the default way to get parseable, reliable output. Templates 31–33 use this pattern; if a script will read the answer, give the model a schema.
• Few-shot still wins on consistency. Adding one to three examples remains the most reliable way to lock in a format. The research backs this: structured demonstrations consistently lifted accuracy across models.
• Reasoning models changed the calculus. OpenAI now notes that reasoning models and standard GPT models can need different prompting — reasoning models often need less hand-holding on the "think step by step" instruction because they reason internally by default.
• Prompt engineering matured into a discipline. Surveys now catalog dozens of distinct, named prompting techniques, a sign the field moved from ad-hoc experimentation to repeatable methodology. A curated library like this one is the practical output of that shift.

One thing did not change: the model produces drafts, not finished work. Every template above hands you a strong starting point. The accuracy check, the voice pass, and the final judgment are still yours.

How do you save these templates so you actually use them?

The library is only valuable if you can reach it in two seconds. Three approaches, roughly in order of long-term payoff:

• Prompt manager (best for daily users). Store each template once with placeholders, and reuse it everywhere. Tools like Prompt Architects add Global Variables — define your brand voice or ICP once and it auto-fills across every prompt — plus a one-click Chrome extension that drops templates straight into ChatGPT, Claude, and Gemini.
• Snippet expander (best for keyboard-driven workflows). TextExpander, Alfred, or Raycast let you type a short trigger that expands into a full template. Fast, but no shared variables and no cross-device sync unless you pay for it.
• Notion or Obsidian doc (best for getting started today). Paste this library into a doc, organize by category, and copy-paste as needed. Zero setup, but you will spend more time hunting and re-filling.

Whatever you choose, the principle is the same: store once, reuse everywhere. The friction of finding a good prompt is what stops most people from using one. Remove the friction and the templates become a habit.

Frequently asked questions

Are these ChatGPT prompts free? Yes — every template above is free to copy, use, and adapt. No signup, no paywall. The variables in brackets are what you fill in. The frameworks (CRAFT, Chain-of-Thought, RTF) are public; only the curation is ours.

Do these ChatGPT prompt templates work on Claude and Gemini too? Most do. CRAFT-formatted prompts transfer directly because they rely on role, context, and format rather than vendor-specific syntax. A few that lean on ChatGPT features like custom GPTs or code interpreter need small adjustments. Test each on your target model — changes are usually minor.

How do I save these templates for daily use? Three options. Save into a prompt manager such as Prompt Architects, which supports placeholders and Global Variables. Save as text snippets in TextExpander, Alfred, or Raycast. Or keep them in a Notion or Obsidian doc and copy-paste. Cross-tool sync wins long term — store once, reuse everywhere.

Why do ChatGPT outputs vary even with the same template? Three reasons. Default sampling introduces randomness, so two runs of the same prompt differ. Model versions update, so the same template can behave differently after an upgrade. And your variable filling matters: vague inputs produce vague output, specific inputs produce specific output.

Should I trust ChatGPT-generated content blindly? No. Templates produce drafts, not finished work. Always edit for accuracy because models still hallucinate, for voice because default phrasing reads generic, and for AI tells like "in summary" openings and over-hedged both-sides answers. The template handles structure; you handle judgment.

What is the CRAFT prompt framework? CRAFT stands for Context, Role, Action, Format, and Tone. You give the model background, tell it who to be, state the task, specify the output shape, and set the voice. Structured frameworks like CRAFT consistently outperform free-form prompts because they reduce ambiguity the model would otherwise guess at.

How many examples should I add to a ChatGPT prompt? For most repeated tasks, one to three examples (few-shot prompting) is the sweet spot. Worked examples show the model your preferred format and lift consistency on tasks you run repeatedly. Beyond a handful you hit diminishing returns and waste context.

Are longer ChatGPT prompts always better? No. Clear beats long. A tight, structured prompt with the right context outperforms a sprawling one. High-capability models often do better with simpler prompts, while cheaper or smaller models benefit more from extra reasoning scaffolding. Add detail only where it removes ambiguity.

By Nafiul Hasan — Founder of Prompt Architects, where we've analyzed tens of thousands of real-world prompts to build tools that make AI output reliable. Last updated: June 10, 2026.