TL;DR: This is a working library of 30 ChatGPT prompts for founders building in public — covering investor updates, customer research, hiring, product specs, growth experiments, and weekly operations. Each one is CRAFT-formatted, copy-paste ready, and built to be filled with your own variables. The prompts do the structure work. You do the judgment work.
What are the best ChatGPT prompts for founders building in public?
The best ChatGPT prompts for founders are structured, reusable templates that turn recurring high-stakes work — investor updates, customer interview synthesis, hiring rubrics, product specs, and growth experiments — into fast, consistent first drafts. The strongest ones specify context, role, action, format, and tone, include 2-3 examples of your own voice, and treat every output as a draft you sharpen rather than a finished artifact you ship.
That direct answer is the whole philosophy of this post in three sentences. The rest is the library, plus the reasoning behind why these particular prompts earn a permanent place in a founder's toolkit while most "100 ChatGPT prompts" lists do not.
If you are building in public — shipping fast, posting your numbers, and running a company with a team of one to ten — your real constraint is not ideas. It is throughput on judgment-heavy work. You have to write the investor update, synthesize ten customer calls, draft the hiring rubric, spec the feature, and design the growth experiment, often in the same week. ChatGPT is the only tool on your stack that can compress all five. The catch is that it only compresses them well when you prompt it like a founder, not like a search engine.
Why do founders get more value from ChatGPT than almost anyone?
Because founders sit at the exact intersection of high task variety and low headcount. A founder does the work of a marketer, a researcher, a recruiter, a product manager, and a copywriter — frequently before lunch. That variety is precisely what general-purpose language models are good at absorbing.
The adoption numbers back this up. A 2024 Techstars survey found that 74% of entrepreneurs treat AI as a component or enabler of their startups, and among startups, ChatGPT dominates with roughly 65% adoption, followed by Anthropic's Claude at 24%. ChatGPT itself has reached an estimated 800–900 million weekly active users, with a DAU/MAU ratio of 36% — nearly double Gemini's, according to Andreessen Horowitz's State of Consumer AI 2025 report. That kind of stickiness only happens when a tool earns a place in daily work, and for founders the daily work is exactly the kind of structured drafting these prompts are built for.
But adoption is not the same as impact. The same body of research notes a recurring failure mode: many teams try 5–10 AI tools in a month but stick with only one or two, a sign of tool fatigue that comes from chasing features instead of fit. The founders who get durable leverage are not the ones with the most tools. They are the ones with a small set of prompts they run so often the prompts become muscle memory. That is what this post is for.
The "AI-assisted founder" vs the "AI-replaced founder"
There is a line worth drawing before we get into the prompts. ChatGPT is excellent at producing the structure of a thing — the skeleton of an investor update, the scaffold of a rubric, the frame of a PRD. It is unreliable at producing the signal — the honest read on whether traction is real, the specific ask that will actually unblock you, the judgment call on a borderline hire.
The AI-assisted founder uses the model for structure and supplies the signal personally. The AI-replaced founder lets the model supply both, ships generic output, and slowly trains their investors and customers to skim past everything they send. Every prompt below is designed to keep you on the right side of that line.
How should you use these founder prompts? (The CRAFT framework)
Each prompt in this library follows CRAFT — Context, Role, Action, Format, Tone. It is a deliberately simple framework because the goal is not cleverness; it is removing ambiguity, which is the single largest cause of bad model output.
| Element | What it does | Founder example |
|---|---|---|
| Context | Grounds the model in your real situation | "We're a B2B SaaS at $14K MRR, 6 weeks post-launch" |
| Role | Sets the perspective the model writes from | "Founder writing to a tier-2 seed investor who skims 30 updates a week" |
| Action | The single concrete task | "Draft a 200-word weekly update" |
| Format | The exact shape of the output | "Headline (8 words), 3 bullets, 1 ask, 1 risk" |
| Tone | Voice and register, ideally with examples | "Confident, specific, no buzzwords. Match my voice: [paste 2 past updates]" |
Three rules make these prompts work far harder than copy-pasting them blindly:
- The variables are the value. Replace every bracketed placeholder with your specifics. A prompt with empty brackets produces a generic answer, because that is literally what you asked for.
- Add 2-3 examples to anything voice-sensitive. Few-shot prompting — giving the model a small number of input/output examples — can deliver 15–40% accuracy improvements on many tasks, with the biggest jump between one and two examples and diminishing returns past four or five. For investor updates, two of your past updates is the right dose.
- Save repeated prompts as templates. Weekly investor updates, monthly hiring loops, and retros are the same prompt with new variables every time. Store them once with
{{placeholders}}so you never pay the blank-page tax twice. (For more on the underlying technique, see our breakdown of prompt engineering frameworks that actually work.)
Now the library. Thirty prompts, organized by the five jobs that eat a founder's week.
What ChatGPT prompts help with fundraising and investor updates?
Investor communication is where a structured prompt pays off fastest, because investors want the same small set of metrics reported the same way every month. Consistency is the whole game — it lets investors spot trends instead of re-learning your definitions every cycle. ChatGPT is exceptional at enforcing that consistency.
1. Weekly investor update
Context: We're a [B2B SaaS / consumer / DTC] at [stage / MRR / users].
Last week's progress: [3 bullets].
Asks/risks: [list].
Role: Founder writing to a tier-2 seed investor who skims 30 updates/week.
Action: Draft a 200-word weekly update.
Format: Headline (8 words), 3 progress bullets (1 line each), 1 ask, 1 risk.
Tone: Confident, specific, no buzzwords. Match my voice: [paste 1-2 past updates].
The headline-first format matters. An investor managing a portfolio reads the subject line and the first sentence; everything after that is for the few who lean in. Front-load the one number that moved.
2. Cold investor email
Recipient: [VC partner name] at [firm], focuses on [thesis].
Sender: [Your role + product + 1-line traction].
Goal: 15-min intro call.
Write 3 cold email variants. Each:
- ≤ 90 words
- Opens with recipient-specific observation about their thesis
- Single CTA (calendar link)
- PS that adds value (relevant data point or intro offer)
The recipient-specific opener is non-negotiable. A cold email that could have been sent to any VC gets the response any generic email deserves.
3. Pitch deck slide outline
Product: [product]. Stage: [stage]. Round size: [$amount].
Outline 12-slide seed deck with: hook, problem, solution, market,
traction, business model, GTM, team, ask, vision, demo, contact.
Each slide: 1 headline (8 words) + 2-line caption + visual concept.
Use this to get past the blank canvas, then ruthlessly cut. The best decks have fewer words on each slide than the model wants to give you.
4. Investor update Q&A prep
Topic: [topic / quarterly review].
Generate 15 likely questions our investors will ask.
Tag each: friendly, neutral, challenging.
For each, write a 100-word answer in my voice.
The tagging is what makes this useful. You want to rehearse the challenging questions out loud before the meeting, not discover them live.
5. Founder narrative refinement
Paste current 1-min pitch: [paste].
Identify: weakest sentence, most impressive sentence, what's missing.
Suggest 3 rewrites tightening narrative without losing authenticity.
This is editorial, not generative — and that is exactly where ChatGPT is strongest. Asking it to critique your existing pitch beats asking it to write a new one from scratch.
A note on metrics. Whatever ChatGPT drafts, make sure your update carries the numbers investors actually weigh: MRR/ARR growth rate, net revenue retention, CAC payback, and burn multiple, plus a short narrative on the drivers. The model can format the bridge; you have to own the numbers.
What are the best ChatGPT prompts for customer research?
This is the category most founders under-use and where the leverage is highest. You probably run customer interviews. You probably do not extract patterns from them systematically. A structured synthesis prompt turns a folder of messy transcripts into a roadmap input in minutes.
6. Customer interview synthesizer
Interview transcript: [paste].
Extract:
- Top 3 pain points (with quote evidence, ≤ 30 words each)
- Top 3 desired outcomes
- Jargon-free language they use to describe problems
- Competitive products mentioned (with sentiment)
- 3 follow-up questions worth asking next interview
Output as structured table.
The "jargon-free language they use" line is secretly the most valuable. Those verbatim phrases become your landing page copy later (see prompt 10).
7. Multi-interview pattern synthesis
5 interviews: [paste each in turn or summary].
Identify cross-interview patterns:
- Pain points mentioned by ≥3 interviews (with frequency count)
- Language patterns (specific phrases used by multiple users)
- Contradictions between users (where they disagree)
- Latent needs no interview directly named but multiple hinted at
The "latent needs" output is where the model occasionally earns its keep entirely. Founders are too close to their own product to see the unspoken theme across five conversations; a pattern-matcher with no ego sometimes catches it.
8. ICP definition refiner
Current ICP: [paste].
Customer interview data: [paste 3 transcripts or summaries].
Refine ICP into: firmographic (company size/industry), psychographic
(values/priorities), trigger event (what makes them search now).
Flag uncertainty where data is thin.
The "flag uncertainty where data is thin" instruction is doing real work. It is your guardrail against the model confidently inventing an ICP attribute from a sample of three.
9. Survey design
Goal: [research question].
Audience: [audience].
Design 10-question survey:
- 1 NPS-style
- 3 multiple-choice
- 3 Likert (1-5 agreement)
- 3 open-ended
For each: question text, answer options, what insight it produces.
The "what insight it produces" column forces every question to justify its existence. If a question doesn't map to a decision you'll make, cut it.
10. Voice-of-customer copy extraction
Paste 10 customer reviews / support tickets: [paste].
Extract verbatim phrases customers actually use.
Group by theme. Mark which phrases would translate well to:
landing page hero, ad copy, feature names, FAQ headers.
This is the prompt that closes the loop between research and marketing. The words your customers already use to describe their pain convert better than anything you'll invent — and this surfaces them at scale. For more on turning research into copy, our guide to AI prompts for landing page copy goes deeper.
Customer research at a glance
| Prompt | Input you need | Output it produces | Decision it informs |
|---|---|---|---|
| Interview synthesizer | 1 transcript | Pains, outcomes, language, competitors | Next interview questions |
| Multi-interview synthesis | 5 transcripts | Cross-cut patterns, latent needs | Roadmap priorities |
| ICP refiner | Current ICP + 3 transcripts | Tightened firmographic/psychographic profile | Targeting and positioning |
| Survey design | Research question | 10-question instrument | Quantitative validation |
| VoC copy extraction | 10 reviews/tickets | Verbatim phrases by channel | Landing page and ad copy |
What ChatGPT prompts make hiring faster and fairer?
Early hires are existential, and most founders make them on vibes plus a rubric they invented an hour before the first call. ChatGPT can't make the decision for you, but it can force you to define the bar before you meet the candidate — which is where most hiring goes wrong.
11. Role rubric generator
Role: [Senior Backend Engineer at startup, $10K MRR].
Level: [Senior, IC4 equivalent].
Generate hiring rubric covering 5 dimensions:
technical depth, scope of impact, communication, cultural add, growth trajectory.
For each dimension: 5-point scale with concrete behavioral indicators.
"Cultural add," not "culture fit." The wording matters: you are hiring for what someone brings that you don't already have, not for how comfortably they blend in.
12. Interview question calibration
Role rubric: [paste from prompt 11].
Generate 12 calibrated interview questions covering all 5 dimensions:
- 4 technical depth questions (behavioral, not whiteboard)
- 2 scope/impact questions
- 2 communication
- 2 cultural add
- 2 growth trajectory
For each: question + what answer reveals + red flags + green flags.
The red-flag / green-flag pairing is the upgrade. It turns a list of questions into a scoring instrument, so two different interviewers can compare notes on the same scale.
13. Take-home eval rubric
Take-home assignment: [paste].
Generate evaluation rubric: 4 dimensions, weighted, scored 1-5 each.
Include 'flag' criteria (instant disqualifiers) and 'wow' criteria
(automatic advancement).
Defining "wow" and "flag" criteria before you read any submission protects you from the halo effect — the tendency to rate a charismatic candidate's mediocre work generously.
14. Post-interview synthesizer
Interview notes: [paste raw].
Extract: candidate strengths (with evidence), candidate concerns
(with evidence), open questions (what we still need to assess),
overall recommendation (advance / next round / pass), confidence
level (high/medium/low).
The "with evidence" requirement is the key constraint. It stops you from writing "great communicator" when what you mean is "I liked them." Evidence-backed notes are also what you'll want when you debrief with a co-founder.
15. Reference call questions
Role: [role]. Candidate strength to validate: [strength].
Candidate concern to probe: [concern].
Generate 8 reference call questions:
4 open-ended, 2 specific situational, 2 calibration ("on a scale of...").
For each, what we'd hope to hear vs what would be a yellow flag.
Reference calls are where most founders coast. Going in with two specific concerns to probe — and knowing what a yellow-flag answer sounds like — turns a formality into a real signal.
A caution on all five: a rubric encodes a bar, and the model only knows the bar you describe. Calibrate every generated rubric against people you've already hired and would hire again. If your best engineer would score a 3 on the model's scale, the scale is wrong, not the engineer.
How can founders use ChatGPT for product specs and feature decisions?
The product category is about converting raw input — customer pain, a feature backlog, a naming brief — into structured artifacts your team can act on. The model is a fast, tireless first-drafter here, and a genuinely useful prioritization partner.
16. Product spec from user pain
User pain (from interviews): [paste 3-5 quotes].
Generate PRD with:
- Problem statement (1 paragraph)
- Why now (1 paragraph)
- Success criteria (3 measurable bullets)
- Scope (in)
- Out of scope (explicit)
- Open questions
- Risks
Feeding it real customer quotes — ideally the verbatim language from prompt 6 — keeps the spec grounded in a problem someone actually has, not a feature you wish people wanted. "Out of scope (explicit)" is the line that prevents scope creep before a single line of code is written.
17. Feature priority scorer
Features under consideration: [list of 5-10].
Score each on: user impact (1-5), effort (1-5), strategic fit (1-5),
revenue impact (1-5).
Output as table sorted by (impact + strategy + revenue) / effort.
Flag any feature where my prior ranking conflicts with the score.
The conflict-flagging is the actual product here. You don't want the model to make the call; you want it to surface where its math disagrees with your gut, so you can interrogate which one is wrong.
18. Naming brainstorm
What it does: [1-line description].
Audience: [audience].
Constraints: [≤ 2 syllables, available .com, no negative connotations
in EN/ES/FR/DE/JA, no trademark conflict].
Generate 30 name candidates.
For each: meaning, why it could fit, why it could miss.
The "why it could miss" column is what separates this from a random word generator. It pre-empts your own objections so you can move through 30 candidates in one pass.
19. Onboarding flow draft
Product: [product]. New-user goal: [first value moment].
Design 4-step onboarding flow.
Each step: title, 1-line copy, primary CTA, drop-off risk, mitigation.
Include 'aha moment' indicator (what behavior signals success).
Naming the "aha moment" as a measurable behavior — not a feeling — gives you the activation metric to instrument later. If you can't name the behavior, you can't measure activation.
20. Pricing page rewrite
Current pricing: [paste tiers + prices].
ICP: [audience].
Rewrite for: clarity (which tier for whom), conversion (reduce friction),
expansion (clear upgrade path).
Output: 3 tier titles, 5-7 feature bullets per tier, ideal-for line,
6 common pricing FAQs with answers.
The "ideal-for line" per tier does more conversion work than any feature bullet. People don't buy a tier; they buy the version of themselves that the tier is for.
What ChatGPT prompts help with growth and marketing experiments?
Growth is where building in public and AI compound. Building in public is itself a powerful distribution strategy — one analysis of indie launches found Indie Hackers converting at 23.1% versus Product Hunt's 3.1%, driven by sustained journey-sharing rather than one-day spikes. These prompts help you produce the volume of experiments and content that "building in public" demands without burning out.
21. Growth experiment design
Hypothesis: [paste hypothesis].
Design experiment:
- Metric (single primary)
- Variants (control + 2 treatments)
- Sample size (assume effect size 5%, alpha 0.05, power 0.8)
- Duration estimate
- Risks + mitigations
- Decision criteria (what we'll do at each outcome)
The single-primary-metric constraint is the discipline most founders skip. An experiment that "improves engagement" without one metric to move is a vibe, not a test. Pre-committing to decision criteria before you see results is how you avoid p-hacking your own roadmap.
22. Landing page hero rewrite
Current headline: [paste].
Subhead: [paste].
ICP: [audience].
Generate 10 variants. Mix: benefit-focused, problem-agitate,
curiosity-gap. ≤ 12 words each. Rank top 3 by predicted CTR.
For top 3: matching subhead.
Treat the predicted ranking as a hypothesis to test, not a verdict. The model is guessing at your audience; your A/B test is the only judge that counts.
23. Email subject lines for indie launch
Email purpose: [launch / re-engagement / update].
Audience: [list segment].
Generate 30 subject line variants.
Categories: curiosity (10), benefit (10), founder-personal (10).
Constraint: ≤ 50 chars, no clickbait, no spam triggers.
Rank top 5 by predicted open rate.
The "founder-personal" category tends to win for building-in-public audiences. People subscribed to you, not to a brand — subject lines that sound like a message from a friend outperform polished marketing.
24. Cold partnership outreach
Target: [company]. Their angle: [their thesis / focus].
Our offer: [partnership type].
Goal: 30-min intro.
Write 3 outreach variants. Each ≤ 120 words.
Open with their-relevant insight, single CTA, no pitch deck attached.
"No pitch deck attached" is the rule that gets the first reply. The goal of a cold partnership email is a conversation, not a close.
25. Content calendar (next 4 weeks)
Niche: [niche]. Audience: [audience]. Pillars: [3-5 content pillars].
Generate 12-piece content calendar (3/week × 4 weeks).
Format as table: week, pillar, format (post / thread / video / longform),
working title, target keyword, predicted intent.
This is the prompt that makes building in public sustainable instead of sporadic. A calendar turns "I should post more" into a system you can actually run. Pair it with our walkthrough on building a reusable prompt library so next month's calendar is one click away.
What ChatGPT prompts streamline founder operations and personal focus?
The last category is unglamorous and high-frequency: the operational and personal work that quietly consumes a founder's week. Here the model acts as a synthesizer and a thinking partner, compressing scattered inputs into decisions.
26. Vendor decision matrix
Decision: [pick vendor for X].
Options: [list 3-5].
Criteria: [list 5-7 weighted criteria].
Score each option 1-5 per criterion. Compute weighted total.
Flag tie-breakers and dealbreakers.
Recommend choice with rationale.
The weighted total externalizes a decision you'd otherwise make on the last demo you happened to see. Pair this with chain-of-thought reasoning — ask the model to show its scoring logic — so you can audit where you disagree.
27. Async update synthesizer
Slack messages from past 24h: [paste].
Synthesize into a 100-word async update for absent teammates:
- Decisions made
- Open questions
- Action items (with owner)
- Things needing my input
For distributed teams, this is a daily time-saver. It turns a noisy channel into a scannable digest, and the "action items with owner" line prevents the most common async failure: everyone reads, no one acts.
28. Email triage
Inbox snapshot: [paste subject lines + 1-line summaries of 20 emails].
Categorize: respond now (≤ 5), respond later (≤ 10), archive (≤ 5).
For 'respond now', draft 1-line response to each.
Skip anything that doesn't need me specifically.
The "skip anything that doesn't need me specifically" instruction is the whole point. Founder time is the scarcest resource in the company; this prompt protects it by forcing a triage you'd otherwise avoid.
29. Weekly retrospective
This week's wins: [paste].
This week's stuck points: [paste].
Synthesize:
- Top pattern across stuck points
- Single highest-leverage change for next week
- One thing I should stop doing
- One thing I should keep doing more of
This is the prompt most worth making a weekly ritual. The "top pattern across stuck points" output catches the recurring failure you're too in-the-weeds to name yourself — the same role chain-of-thought plays for hard reasoning tasks.
30. Personal OKR draft
Quarter: [Q3 2026].
Role: founder of [stage] startup.
Top constraint: [biggest blocker].
Draft 3 personal OKRs. Each: objective (qualitative), 3 measurable
key results, 1 leading indicator. Tied to company-level priorities.
Tying personal OKRs to company-level priorities is the check against busywork. A founder can be productive all quarter and move nothing that matters; this forces the alignment.
How do you make these prompts stick? (Beyond copy-paste)
A library only compounds if you actually reuse it. Here's how to convert thirty prompts from a one-time read into a permanent workflow.
- Save your top 5 as templates. Identify the prompts you'll run weekly — almost always the investor update, the retrospective, the content calendar, plus two role-specific ones. Store them with
{{placeholders}}in a prompt manager so they're one keystroke away. We built Prompt Architects for exactly this, and it works across ChatGPT, Claude, and Gemini so you build the asset once. - Add 1-2 examples to anything you run repeatedly. Few-shot examples consistently outperform zero-shot prompts; the research is clear that two to three well-chosen examples capture most of the gain. For voice-matching, your own past writing is the best example there is.
- Iterate by tightening one variable per attempt. When an output is 80% right, don't rewrite the prompt from scratch. Change one thing — the tone line, the word count, the format — and re-run. Single-variable iteration is faster and teaches you what each lever does.
- Pair with chain-of-thought for hard reasoning. For the decision matrices (26), feature scoring (17), and retros (29), ask the model to think step by step and show its work. Combining few-shot examples with chain-of-thought reasoning dramatically improves performance on complex tasks.
Where AI helps vs where you stay in the loop
| Founder task | What ChatGPT does well | What only you can do |
|---|---|---|
| Investor update | Structure, consistency, voice-matching | The honest read on traction and the real ask |
| Customer synthesis | Pattern extraction, verbatim capture | Deciding which pattern becomes a roadmap |
| Hiring | Rubric scaffolds, question calibration | The yes/no call on a borderline candidate |
| Product spec | First-draft PRD, scope framing | Whether to build the thing at all |
| Growth experiment | Experiment design, content volume | Reading results and changing direction |
The pattern is consistent across all five jobs: ChatGPT is a structure layer, not a content layer, for anything where your credibility or your strategy is on the line. The output of every prompt in this post is a draft, not a finished artifact. The prompt does the structure work. You do the judgment work. That split is the entire difference between an AI-assisted founder and an AI-replaced one.
How to stop ChatGPT from sounding generic in your founder writing
This deserves its own section because it's the single most common complaint, and it's fixable. Investor updates, launch emails, and public build-in-public posts all live or die on sounding like you. Three concrete fixes, in order of impact:
- Paste 2-3 examples of your past writing. This is the highest-leverage move. The model imitates what you show it; show it your actual cadence, your actual sentence length, the way you actually open an email. Without examples, it defaults to a corporate-neutral register nobody asked for.
- Specify your voice as 5-7 attributes. "Confident, specific, slightly playful, never uses 'we believe,' short sentences, no jargon, one number per paragraph." Concrete attributes beat vague ones — "professional but human" tells the model nothing.
- Edit out the giveaways manually. Even a well-prompted draft has tells: stacked em dashes, "in summary" and "in conclusion," "it's important to note," and the both-sides hedging that takes no position. A 60-second editing pass to strip these is the cheapest credibility insurance you'll ever buy.
The deeper point: AI-detection is not your real risk. Sounding interchangeable with every other founder using the same tool is. Your authentic voice plus AI-assisted structure beats either one alone — and these three fixes are how you get there.
Frequently asked questions
What's the highest-leverage ChatGPT prompt for solo founders? Customer interview synthesis. Most founders run interviews but never extract patterns systematically. A structured-prompt extraction across 10 interviews — pulling top 3 pain points with quote evidence, top 3 desired outcomes, jargon-free language, and competitive products mentioned — produces insight that would take a week of manual review. It turns scattered conversations into a defensible roadmap.
Should I use ChatGPT for fundraising and investor emails? For drafting first cuts and generating subject-line variants, yes. For final send, edit the AI tells out. AI-detection isn't the real risk — sounding generic is. Investors skim dozens of updates a week, so use ChatGPT to structure the update while you supply the signal: real progress, specific asks, and honest risks.
How do I stop ChatGPT from sounding "AI-ish" in founder updates? Three fixes. Paste 2-3 examples of your past writing into the prompt so the model can mirror your cadence. Specify your voice as 5-7 concrete attributes ("confident, specific, slightly playful, never uses we believe"). Then edit out the giveaways: stacked em dashes, "in summary," and balanced both-sides hedging that takes no position.
Are AI-generated investor updates risky? Only if you don't review them. Investors care about signal: progress, asks, and risks reported the same way every month. Use AI to structure the update; you write the signal. Treat ChatGPT as a structure layer, never a content layer, for any high-stakes communication where your credibility is on the line.
What ChatGPT prompts actually help with hiring? Three categories work well: rubric generation (per role and level), interview question calibration (12 questions mapped to the rubric), and post-interview synthesis (extracting fit signals from your raw notes). Treat every output as a draft and calibrate it against your real hires, because a rubric is only as good as the bar it encodes.
Do prompt frameworks like CRAFT really improve results? Yes. Structured prompts that specify context, role, action, format, and tone reduce ambiguity, which is the main cause of low-quality output. Few-shot examples alone can lift task accuracy by 15-40% on many tasks, and adding 2-3 examples of your own writing gives the model a target to imitate instead of guessing.
How many examples should I include in a founder prompt? Two to three is the sweet spot. Research on few-shot prompting shows the biggest accuracy jump happens between one and two examples, with diminishing returns past four or five. For voice-matching tasks like investor updates, paste two of your best past examples; for structured tasks like rubrics, one strong example is often enough.
Should I save my best prompts as reusable templates? Absolutely. Founders run the same prompts weekly — investor updates, retros, content calendars. Saving them as templates with placeholders removes the blank-page tax and keeps your outputs consistent. A prompt library that works across ChatGPT, Claude, and Gemini means you build the asset once and reuse it everywhere.
By Nafiul Hasan — Founder of Prompt Architects, builder of a prompt-enhancement tool used across ChatGPT, Claude, and Gemini, writing from the daily reality of running a company on AI-assisted workflows. Last updated: June 10, 2026.