10 AI Prompts I Use as a Startup Founder
I've shipped products to real customers, raised small rounds, and sat through enough customer interviews to know this: most founder writing is done under panic at 11pm. The pitch deck. The investor update. The cold email to an enterprise prospect. The customer-interview script you scribbled on a napkin.
These are the 10 prompts I actually use. Each follows a four-part structure I'll break down at the end. Every output below is real, edited only for length.
1. The Idea Stress Test (Kill It Before You Build)
The single most valuable thing AI can do for a founder is tell you your idea is bad before you spend six months building it. This prompt is designed to find the weakest assumption, not to validate.
Role: A skeptical seed-stage investor who has passed on 200+ startups.
Context: I am considering building [PRODUCT]: a [ONE-SENTENCE
DESCRIPTION] for [TARGET CUSTOMER]. The core hypothesis is
[CORE HYPOTHESIS]. The wedge is [WHY NOW / WHY THIS WEDGE].
My plan to acquire the first 100 customers is [GTM IDEA].
Constraints: Do NOT validate this idea. Your job is to find the
single weakest link in the chain — the assumption that, if wrong,
kills the company. Identify the weakest assumption, explain why it
is fragile, and name the cheapest possible test that would falsify
it in under two weeks for under $200.
Output: Three sections. (1) The weakest assumption (one sentence).
(2) Why it is fragile (two sentences, with a concrete failure
scenario). (3) The cheapest falsification test — the exact test,
the metric that would kill the idea, and the kill threshold.
Real output (idea: a config-file generator SaaS for solo developers):
(1) Weakest assumption: Solo developers will pay for a config-file generator when free templates exist everywhere.
(2) Why it is fragile: The audience has near-zero willingness to pay for tools, and the free alternative (copy from a repo) is one Google search away. Failure scenario: you ship the product, get 1,000 free users, and convert 0 to paid because the free tier was already good enough.
(3) Cheapest falsification test: Build a one-page landing page with a Stripe "Buy Now" button at $19. Drive 200 visitors via a single Reddit post in r/webdev. Kill threshold: fewer than 3 purchases in 72 hours means the willingness-to-pay assumption is dead. Total cost: under $50.
Run this on every idea before you write a line of code. If the model can't find a fragile assumption, you haven't thought hard enough — re-run it.
2. The Investor Update the LP Can Actually Read
Investors skim. This prompt forces you to lead with the number that matters, admits the bad news before they have to dig for it, and asks for exactly one thing.
Role: A founder writing a monthly update that a busy investor
will read in under 60 seconds.
Context: Company: [NAME]. Month: [MONTH]. Stage: [STAGE].
This month's headline metric: [METRIC] went from [OLD] to [NEW]
([DIRECTION]). The three things I did this month:
- [THING 1]
- [THING 2]
- [THING 3]
The thing that did not work: [FAILURE]. The hires I need to make:
[HIRING NEEDS]. The one decision I am stuck on: [STUCK DECISION].
Constraints: The first line must be the metric. The update must
be under 200 words. You must name the failure before any investor
can find it on their own. You must end with exactly one ask.
Output: The email. Plain text. No section headers. The single ask
on the last line, bolded, in one sentence.
Real output (month: MRR grew, churn spiked, hiring need = senior eng):
MRR grew from $8.2k to $11.4k (+39%), but logo churn spiked from 4% to 9% this month — the two largest customers left for an enterprise competitor.
We closed the Series A bridge extension, shipped the team-collaboration feature (the one I flagged in May), and ran 14 customer interviews to understand the churn. The interviews surfaced the issue: our mobile experience is materially worse than the competitor's, and both churned customers named it unprompted.
The thing that did not work: the lifecycle email campaign to re-engage churned trials. 0 conversions out of 240 sends.
I need to hire one senior mobile engineer (iOS or React Native) before September — if you have anyone in your portfolio network who might be open, that is the single biggest unlock.
Can you forward this to any operator you know who has built and shipped a native mobile app at scale?
That bolded last line is the entire point. Updates without a specific ask get filed and forgotten.
3. The Customer Interview That Doesn't Lead the Witness
Most founders run customer interviews that confirm what they already believe. This prompt constructs a script that forces the interviewee to describe their actual behavior, not their opinions.
Role: A customer development researcher trained in The Mom Test.
Context: I am interviewing [CUSTOMER PERSONA] about [PROBLEM AREA].
I believe they struggle with [HYPOTHESIZED PROBLEM]. The product I
am considering would [PROPOSED SOLUTION]. I have 30 minutes with
this person. They are a warm intro from [MUTUAL CONTACT].
Constraints: Write a 10-question interview script. No question may
mention my product, my solution, or my hypothesis. Every question
must be about the interviewee's past behavior, not their future
intentions. No "would you use X?" questions allowed. Include a
specific follow-up question for any answer that mentions a feeling
("frustrating", "annoying") to convert it into a concrete story.
Output: The 10 questions, in order. After each, write the
follow-up probe in italics. Number the questions. The last question
must ask for the name of one other person with the same problem.
Real output (persona: engineering manager, problem: onboarding new hires):
- Walk me through the last time you onboarded a new engineer — from their first day to the day they shipped their first production change. Follow-up: how long was that, in calendar days?
- What did you personally spend time on during that onboarding? Follow-up: of those things, which felt like the biggest time sink?
- Tell me about a specific moment in that onboarding that felt harder than it should have. Follow-up: what did you do right after that moment?
- When you onboarded your most recent hire, what tools or documents did you actually open? Follow-up: which of those did you have to update before sharing? ... (continues to Q10)
- Is there one other engineering manager you know who has onboarded someone in the last 90 days? Follow-up: would you be willing to forward a one-line intro?
The "past behavior, not future intent" rule is the entire game. People cannot predict what they will do; they can only describe what they did.
4. The Pitch Deck Section That Survives the First Slide
Founders waste months on deck design. The design doesn't matter if the words are weak. This prompt forces you to write each slide as a single sentence that could be read aloud in 10 seconds.
Role: A founder writing a 12-slide seed deck for a 7-minute pitch.
Context: Company: [NAME]. One-liner: [ONE SENTENCE]. Stage: [STAGE].
Founders: [FOUNDER BACKGROUNDS]. The ask: [AMOUNT] for [EQUITY OR SAFE].
The wedge: [WEDGE]. The 3 proof points I have: [POINT 1, 2, 3].
Constraints: For each slide, write the headline (one sentence,
max 12 words) and the body (max 30 words). No bullet may contain
the word "leverage", "platform", "ecosystem", "synergy",
"revolutionary", "disruptive", or "AI-powered" as a standalone
claim. The headline of slide 1 must be a number. The headline of
slide 12 must be a specific ask with a specific dollar amount.
Output: The 12 slide headlines, each with its 30-word body. Then
a one-paragraph critique: which slide is weakest, and why.
Real output (slide 1, then the self-critique):
Slide 1 — $1.4M to turn 30 minutes of weekly support tickets into a shipping checklist. We take the support inbox, surface the bugs worth fixing, and route them to the right engineer — so the bugs that actually lose customers get fixed first.
(skipping slides 2-11 for length)
Slide 12 — Raising $1.4M on a $12M post-money SAFE to hire two engineers and reach $30k MRR in 9 months.
Self-critique: Slide 5 (the market-size slide) is weakest. You wrote "$8B TAM" with no bottom-up math. An investor will discount the entire deck the moment they see a TAM with no derivation. Replace it with: "we can reach $5M ARR by capturing 0.4% of the 22,000 Series A–C SaaS companies, who spend an average of $5,700/year on support tooling."
The self-critique is the most valuable part of this prompt. It catches the slide that loses the room.
5. The Cold Email That Earns a Reply
Cold emails from founders fail because they pitch. This prompt forces a trigger-first subject line, a 90-word body, and a question (not a meeting link) as the call-to-action.
Role: A founder writing a cold outbound email to [BUYER PERSONA]
at [COMPANY TYPE]. The trigger: I saw [SPECIFIC TRIGGER — a blog
post they wrote, a hire they made, a funding round, a product
launch]. The product: [ONE-SENTENCE DESCRIPTION]. The single
proof point: [ONE NUMBER OR ONE NAMED CUSTOMER]. I have no prior
relationship with this person.
Constraints: The subject line must be lowercase, under 6 words,
and must reference the trigger — not the product. The body must
be under 90 words. The first sentence must be about them, not me.
The CTA must be a question that takes one sentence to answer, not
a Calendly link. Banned words: "revolutionary", "game-changing",
"seamless", "synergy", "excited", "thrilled", "delighted".
Output: The subject line on its own line, then the email body.
Real output (buyer: VP Eng at a Series B, trigger: they blogged about onboarding pain):
subject: your onboarding post + the 23% number
Hi [NAME] — the part of your post about engineers taking 47 days to ship their first change matched a number we just measured: the teams we work with cut that to 21 days by reordering the first-week tasks.
The biggest single lever was splitting "environment setup" from "first PR" — most teams bundle them, which is where new hires stall.
Worth a quick exchange on what your onboarding doc looks like today? Not a demo, just a question.
The lowercase subject, the number in the subject, the "not a demo" disclaimer — these are the three moves that get cold emails opened. Pitch on the first email and you are dead.
6. The Job Description That Attracts the Right Person (and Repels the Wrong One)
Most job descriptions are Wish Lists. This prompt forces a JD that describes the work, names what will make the candidate quit, and bans the "culture fit" escape hatch.
Role: A founder writing a job description for [ROLE] at [STAGE]
company. The work this person will do in their first 90 days:
- [WORK 1]
- [WORK 2]
- [WORK 3]
The single most important outcome for the first year: [OUTCOME].
The team size: [N]. The reporting line: reports to [FOUNDER/EXEC].
Constraints: The first section must be "What you will do in week 1",
written as a numbered list of 5 concrete tasks. The second section
must be "What will make you quit" — name 3 things that, if true
about the role, mean the candidate should not take the job. The
third section is the requirements — max 4 bullets, each must be
testable from a resume in under 10 seconds. Banned phrases:
"rockstar", "ninja", "wear many hats", "fast-paced environment",
"work hard play hard", "culture fit", "passion for", "self-starter".
Output: The three sections, in order. Under 250 words total.
Real output (role: first product hire at a seed-stage SaaS):
What you will do in week 1
- Sit in on 5 customer calls and write a one-page synthesis of what we heard.
- Draft the spec for the feature we have already pre-sold to 3 customers (due in 6 weeks).
- Run the weekly metrics review with the founders.
- Pick one thing in our onboarding flow that is broken and ship a fix.
- Tell us which of our 14 "top priorities" should actually be cut.
What will make you quit
- You want a team of 5 PMs to delegate to. We have 0.
- You need a 6-month roadmap signed off before shipping. We change direction every 2 weeks.
- You are uncomfortable being the single point of failure for product decisions.
Requirements
- Shipped a B2B SaaS feature from spec to release as the lead PM.
- Can write a SQL query that joins 3 tables without help.
- Has sat in on at least 25 customer calls.
- Can disagree with the founder in writing, clearly, without softening.
The "what will make you quit" section filters out 80% of unqualified applicants before the screening call. It is the single highest-leverage section of any JD.
7. The Pricing Page That Justifies the Top Tier
Most pricing pages anchor on the cheap tier. This prompt forces the top tier to look like the obvious choice for the right customer — using honest framing, not dark patterns.
Role: A founder writing a 3-tier pricing page for [PRODUCT].
Tiers: [TIER 1: NAME + PRICE + WHO IT'S FOR], [TIER 2], [TIER 3].
The most expensive tier is for [CUSTOMER TYPE]. The single reason
someone picks the top tier over the middle: [REAL REASON]. The
middle tier exists to: [REASON — usually team size or a specific
feature]. The cheap tier exists to: [REASON — usually individual use].
Constraints: Do NOT use the phrases "Most Popular", "Best Value",
"Ultimate", or "Enterprise". The top tier's description must be
the only one that names a specific outcome (not a feature). The
middle tier must be described as "for teams that need X", not as
a discount. The cheap tier must explicitly say who should NOT buy it.
Output: The 3 tiers as they would appear on the page. For each,
a 2-line description (line 1: who it's for; line 2: what they get).
Then a one-line "completing the set" line at the bottom that frames
the top tier as the obvious upgrade path.
Real output (3 tiers: $9 individual, $49 team, $199 company-wide):
Individual — $9
For one person building one project. Includes the 7 launch prompts.
If you have a team or run a business, this is not the right tier — you will outgrow it in a week.Team — $49
For a team of up to 5 that ships together. Includes everything in Individual, plus the 25-prompt marketing copy pack and the 30-prompt productivity library.Complete Library — $199
For founders and operators running the whole show. Includes all 5 packs (130+ prompts) and the 10-workflow AI startup operations system — the exact system used to run product, growth, and customer ops from a single prompt library.Most teams that buy the Team pack come back within 30 days for the Complete Library — the workflows are what turn individual prompts into a system.
The "completing the set" line at the bottom is honest (most buyers do upgrade) and it frames the top tier as the destination, not as the most expensive option.
8. The Standup That the Whole Team Reads in 60 Seconds
Founder-written standups usually become essays. This prompt forces a status update that fits in a Slack message, leads with the blocker, and ends with one named person who can unblock it.
Role: A founder writing a daily team standup in Slack.
Context: Yesterday I: [THING 1], [THING 2], [THING 3]. Today I
plan to: [THING 1], [THING 2]. The single blocker is: [BLOCKER].
The person who can unblock it is: [NAME]. The decision I need the
team to make today is: [DECISION].
Constraints: The first line must be the blocker, not the wins.
No bullet may exceed 12 words. The update must fit in a single
Slack message (under 150 words). The decision must be phrased as
a binary (A or B), not an open question.
Output: The Slack message, formatted with emoji headers (🟥 blocker,
✅ done, 👉 today, ❓ decision). The last line is the decision,
phrased as a binary.
Real output:
🟥 Blocker: the Stripe webhook is failing in staging — @maria can you look at the event signature validation?
✅ Yesterday: closed the seed bridge extension. Shipped the onboarding fix (the 21-day one). Wrote the investor update.
👉 Today: customer interview at 11. Draft the Q3 OKRs. Review the mobile engineer resumes.
❓ Decision: do we hire the iOS contractor for 3 months at $X, or push the mobile launch to Q4 and hire full-time? Need a call by EOD.
Leading with the blocker is counterintuitive — most founders lead with the wins. But the team can only help with the blocker. The wins are for the investor update.
9. The Board Memo That Anticipates the Hard Question
Board meetings fail when the founder gets surprised by a question they should have seen coming. This prompt generates the three questions the board will ask, plus the hardest one — the one they will save for the end.
Role: A founder preparing a board memo for [BOARD MEETING TYPE:
quarterly / emergency / pre-round]. The company's situation:
[STAGE], [HEADLINE METRIC + DIRECTION], [BIGGEST RISK]. The three
topics on the agenda: [TOPIC 1], [TOPIC 2], [TOPIC 3]. The board
members are: [PROFILE 1 — operator], [PROFILE 2 — financier],
[PROFILE 3 — independent].
Constraints: Write the memo first — under 400 words, structured
as: the number, the trend, the risk, the ask. THEN write the
questions section. Generate 4 questions: 3 the board will ask
based on the agenda, and 1 "the hard one" they will save for
the end. For each question, write the answer in 2 sentences —
and flag whether the answer is strong, weak, or "I don't know".
Output: The memo, then the 4 Q&As. Do not soften the "I don't
know" answers — those are the ones to prepare for.
Real output (excerpt — the hard question):
The hard question (from the financier): "Your CAC has risen 40% in two quarters. At what CAC do you stop acquiring?"
Answer: We model break-even on a 14-month payback. At current trend we hit that ceiling in Q3. We have not yet decided whether to throttle paid acquisition or shift the mix to outbound — I don't know the answer yet, but I will bring a recommendation to the next board meeting.
The "I don't know" is the most important sentence in the memo. Boards trust founders who name what they don't know. They distrust founders who fake confidence.
10. The Investor Pitch That Survives the Coffee Chat
Formal pitches are easy. Coffee chats are hard — because there is no deck to hide behind. This prompt constructs a 90-second spoken pitch that works at a coffee shop, with no slides.
Role: A founder delivering a 90-second pitch in a coffee chat
(no deck, no slides). The company: [NAME]. The wedge: [WEDGE].
The proof: [ONE NUMBER]. The ask: [AMOUNT] for [STAGE]. The
founder's unfair advantage: [WHY YOU].
Constraints: The pitch must be spoken, not written — no bullet
points, no headers. It must open with a specific number or a
specific story (not a vision statement). It must contain exactly
one metaphor. It must end with a question that invites the
listener to engage, not a request for money. Under 200 words
spoken aloud.
Output: The pitch as a single paragraph. Then, in italics, the
single sentence the founder should cut if the listener interrupts.
Real output:
"We cut the time from a support ticket to a shipped fix from 14 days to 3. The wedge is simple: 70% of support tickets mention the same 12 bugs, and most teams fix them in the order they were reported, not in the order they lose revenue. We surface the ones that are actually churning customers. I spent four years running support at a company that hit $40M ARR — I watched us lose six-figure accounts because the bugs that lost them were never the bugs we prioritized. We have 11 paying customers in 9 weeks, including two names you would recognize. Raising a $1.4M seed to hire two engineers and reach $30k MRR by Q2. The question I keep getting from operators is: how do you decide which bug to fix first when everything feels urgent? I'd love to hear how you'd think about that prioritization."
Cut if interrupted: "Raising a $1.4M seed to hire two engineers and reach $30k MRR by Q2."
The pitch ends with a question — not an ask. The ask comes when the listener leans in.
The Structure Behind These Prompts
Every prompt above uses the same four-part structure. Once you internalize it, you can write your own in under a minute.
Role — Tell the model who is writing. "A skeptical seed investor" produces a different output than "an encouraging mentor". Pick the role that gives you the harsh truth, not the validation.
Context — The specific situation, in concrete nouns. Not "a startup" — "a seed-stage SaaS with $8k MRR and 9% monthly churn". Specificity is the difference between generic advice and a usable output.
Constraints — What the output must NOT do. This is where most prompts fail. "Write a pitch" gives you a generic pitch. "Write a pitch with no bullet points, under 200 words, that ends with a question" gives you something you can actually use.
Output — The exact format. "A numbered list, max 5 items, the last one labeled 'the uncomfortable one'." The more specific the format, the more usable the result.
How I Test Every Prompt
Before a prompt goes into my library, I run it through four checks:
- Run it 3 times with the same inputs. If the outputs vary wildly, the prompt is too vague — tighten the constraints.
- Change one input (the customer persona, the metric, the stage). The output should change meaningfully. If it doesn't, the prompt is producing a template, not a real answer.
- Read it aloud. If you can't say it in a real conversation, it's not done. Rewrite until it sounds like a human.
- Cut anything generic. If a sentence could appear in any founder's update, cut it. The sentence that only you could write is the one that matters.
Where to Get All of These (And 120 More)
These 10 prompts are a slice of a larger library I use to run the operational side of a startup — idea validation, customer development, pitch materials, hiring, growth, and the day-to-day writing that founders do every week.
The full library lives here:
- The Complete AI Prompt Library for Founders — all 5 packs (130+ tested prompts and 12 workflows) in one bundle.
- Developer Product-Launch Prompt Pack ($9) — the 7-prompt entry point, if you want to start small.
- SaaS Marketing Copy Prompt Pack ($49) — 25 prompts for landing pages, cold emails, and ad copy.
- Developer Productivity Prompt Library ($49) — 30 prompts for engineering work.
- AI Startup Operations Prompt System ($199) — 50+ prompts and 10 workflows for running the whole company.
- AI Business Transformation Playbook ($999) — the flagship, 100+ prompts and 12 workflows for a full operating system.
Every prompt in every pack was tested with the four-step method above. No filler.
If this was useful, the $9 pack is the cheapest way to see if the rest of the library is for you.
— A founder who writes too much at 11pm
Top comments (0)