🎒 Welcome to angela builds · for high school founders
Build something real — with AI as your thinking partner.
This is home base for Entrepreneurship & Innovation: every class, every prompt, every tool from our sessions — free, open, no sign-up. AI here argues back, asks the next question, and pressure-tests your thinking. The interviews, the judgment, and the insight stay yours.
Every class. Every phase. There's a companion for it.
Built around the sessions we actually ran — the interviews, the synthesis, the build studio, the pitch. Same anatomy every time: The Arc · Handouts · AI Companion · The Numbers · Landmines.
The starting line · Class 2 · Free · No sign-up
Become a Customer Detective
Everything downstream — your problem statement, your UVP, your MVP, your price — is built on what you hear in interviews. This companion turns Claude into your devil's advocate, your practice interviewee, and your bias-checker: nine cards, every prompt from class expanded and ready to copy. The through-line: a problem statement is a hypothesis, not a fact.
A · Think like a detectiveB · Find the real jobC · Go talk to people
Before anything else, do the pre-work
Every companion assumes three things are in your hands before you open it:
Your problem statement — one sentence, from class. Rough is fine; that's the point.
Your notes, anonymized — strip names and schools from interview notes before they touch any AI tool. Every card repeats it.
The last class's output — the cards link forward: your answers from one class become your paste-ins for the next.
The companions are the layer on top of class — not a replacement for showing up, interviewing real people, and owning the insight.
Class Tools
Honest answers in five minutes.
No login. No email. One for your idea, one for you.
🎯
Idea Signal Scorer
Six honest questions about your current idea — validated, hypothesis, or guessing? Score, verdict, and the next card to open.
Score my idea →
🧭
Interview Readiness Check
Solo or with your team — four questions on interview craft, and the exact companion card that closes your gap.
Check my readiness →
From the builders
Real stories from real builders.
Go further · Stanford Continuing Studies · Fall 2026
Ready for the full course?
Everything on this page comes from a real course. BUS 81 — Launching from Idea to Reality takes first-time founders through the whole arc live: customer discovery, validation, MVP, unit economics, go-to-market, and the pitch — with AI as accelerant, never as judgment.
When
Thursdays, 5:30–7:00 pm PT Sep 24 – Dec 3 · 10 weeks No class Nov 26 · recordings available
Where & what
Online · 1 unit · no letter grade Enrollment capped at 35 Tuition $685 · refund deadline Sep 26
You’ll leave with
A validated concept and clarity on your next step — discovery, MVP, unit economics, GTM, pitch.
Strip names and schools from your interview notes before they touch any AI tool. Real people trusted you with their problems; protect them. Anonymized quotes carry all the signal you need — the pattern matters, the name never does.
The capability check
Did it work?
Completion is not comprehension, and comprehension is not changed behavior — so we check behavior. After you've done the real work — the interviews, the synthesis, the build — come back and check only what you have actually done, in your own words.
In class we asked: are you solving a real problem, or the problem you want to exist? It's easy to invent a problem that fits the solution you already dreamed up. This prompt gives Claude a skeptical role and the context to poke holes in your thinking.
Copy into Claude →
You're a friendly but honest startup mentor. Help me figure out if I'm chasing a real problem.
Here's my idea:
- the problem I want to solve: [your problem in 1-2 sentences]
- who I think has it: [the people you'd serve]
- why I care about it: [1 sentence]
Please do this in order, in plain language:
1. Tell me honestly: does this sound like a real problem people have, or one I might just want to exist so my idea works?
2. Ask me 3 questions that would help me figure out the difference.
3. Tell me what I'd need to hear from real people to know this problem is real.
Be kind but don't just agree with me – I want the truth.
⚠ The trap to watch for ▶
The danger isn't a bad idea – it's a problem you never checked. If you can't name a real person who recently felt this pain, that's a sign to go talk to someone before building anything.
✅You'll get back: an honest read on your problem and what to go check with real people.
🔗Links to #2: keep Claude's answer here – you'll feed your sharpened problem straight into the validation stress-test.
prompt engineering · seek disconfirmation
Discovery is the method; validation is the goal. Before you talk to anyone, have Claude play devil's advocate and surface what you're assuming – and exactly what evidence would prove you wrong. This is the fastest way to stop confirmation-shopping your own idea.
🎓
Straight from class:“here's my problem statement + why I think it's real (no names/schools). what am I assuming? what evidence would prove me wrong? who would push back?” The prompt below is that same move, expanded.
Copy into Claude →
You're a sharp but friendly startup mentor who plays devil's advocate. I might be in love with my own problem, and I want you to find the cracks now – not after I build.
Here's my problem statement and why I think it's real (no names, no schools):
- problem: [paste your sharpened problem from #1]
- why I believe it: [what makes you think it's real]
Do this in order, in plain language:
1. List the assumptions I'm making without proof.
2. Name the single piece of evidence from a real person that would prove me WRONG.
3. Who would push back on this problem being real, and what would they say?
4. Give me 2 things to listen for in interviews that would count as REAL validation.
Be kind, but don't agree just to be nice.
✅ What counts as a validated problem? ▶
You heard it from more than one person.
It's based on a real past action – they already tried to solve it (spent time or money).
There's emotion or a workaround attached to it.
One polite “sure, I'd use that” is noise, not validation. Weigh patterns over single opinions.
✅You'll get back: a list of your hidden assumptions and exactly what would count as proof – so you know what to listen for.
🔗Links to #7: the assumptions Claude surfaces here become the things your 5 interview questions need to test.
prompt engineering · output format
"Confirmation shopping" means secretly asking questions that fish for a yes. Real discovery welcomes bad news instead – it seeks disconfirmation. No questions written yet? No problem – this prompt works either way.
Copy into Claude →
You're a friendly customer research coach. I'm a high school founder doing my first interviews about this problem: [your problem].
If I already have draft questions, here they are: [paste them, or write "none yet"].
If I wrote "none yet," first suggest 4 questions someone might naturally ask about this.
Then, for each question, make a simple 3-column table:
| the question | is it secretly fishing for a "yes"? | a better version that asks about what they ACTUALLY did before |
Explain each in one short sentence I'd understand.
✅You'll get back: a clean set of honest questions you can actually use – even if you started with nothing.
✕ confirmation shopping
seeking a yes
asks about hypothetical futures
avoids bad news
✓ real discovery
curious and open
asks about past behavior
welcomes bad news
🔗Links to #7: the rewritten, honest questions Claude gives you here become the starting point for your 5 homework questions.
prompt engineering · quick self-check
The core principle of good discovery: look for patterns, not individual opinions. That means talking to people before you build, focusing on their problem not your solution, and hearing the same thing from a few people before you trust it. This is a fast gut-check before you go interview.
Copy into Claude →
I'm a high school founder about to interview people about this problem: [your problem]. I plan to talk to [who you'll talk to].
Be my friendly discovery coach. Check my approach against these 3 rules and just tell me "looking good" or "watch out" for each, in one sentence:
1. Am I talking to people BEFORE deciding on a solution?
2. Am I focused on THEIR problem, not selling my idea?
3. Am I planning to talk to a few people (to spot patterns), not just one?
Finish with the single most useful tip for my first interview.
✅You'll get back: a quick thumbs-up or heads-up on your plan, plus one tip to start strong.
prompt linking · context engineering
Everything we use is "hired" to do a job – physical, emotional, or social. Once you name the job, you see who you're really competing with (often not who you'd guess) and where the opportunities are. This one builds on the problem you validated in #1 and #2.
Copy into Claude →
Here's the problem I sharpened and stress-tested earlier:
[paste your answers from #1 and #2]
Act as a product strategist. Help me name the "job to be done" for the people I want to serve.
1. Write the job in this exact frame: "When [situation], I want to [motivation], so I can [expected outcome]." Give me 2 versions to choose from.
2. List ALL the things people currently "hire" to get this job done today, including low-tech, free, or surprising ones – that's my real competition.
3. Point to one gap where every current option is annoying. That's my opportunity.
✅You'll get back: a clear "job" sentence, a list of who you're really competing with, and one opportunity to explore.
☀ Why this matters ▶
Reveals your real competition – the job "unwind after school" competes with TikTok, a nap, and a snack, not just other apps.
Uncovers opportunities – gaps where every current option is annoying.
Helps you prioritize – build the part that does the job best.
🔗Links to #6: the job and the people you named here tell you exactly who to go interview next.
prompt linking · output format
The best interviewees are people who just went through the problem – not people with opinions about it. This finds them, and tells you where to find them this week.
Copy into Claude →
Here's the job to be done I landed on:
[paste your answer from #5]
I'm a high school founder and I can realistically only interview classmates, friends, family, and people in my community. Make me a table:
| type of person | why they RECENTLY lived this problem | where I'd find them this week | one opening line to ask for 10 minutes |
Only include people who actually experienced the problem – not people who just have opinions about it.
✅You'll get back: a ready-to-use list of real people to interview, plus an easy way to ask them.
🔗Links to your homework: pick 2 people from this table to interview for your Class 2 assignment.
prompt linking · self-check
Your homework needs 5 interview questions that cover three goals. This drafts them from everything you've learned so far, then makes Claude grade its own work so nothing sneaks past.
🎓
Straight from class:“here are 5 questions I want to ask people about [your problem]. which ones lead the witness or ask them to predict the future? rewrite each as a question about a real past action.” The prompt below does this and drafts your five for you.
Copy into Claude →
You're a customer interview expert. Here's what I've figured out so far about my problem and the job to be done:
[paste your answers from #5 and #3]
Write me exactly 5 open-ended interview questions that cover these three goals:
1. their current situation and challenges
2. how they hope this problem could be solved
3. what "success" looks like for them
Then grade your own questions: for each, label it GOOD (asks about real past behavior) or RISKY (leads them or asks them to guess the future), and rewrite any RISKY ones. Give me a clean final list of 5 at the end.
✅You'll get back: your 5 interview questions for homework – already checked and cleaned up.
✓ What good vs. bad looks like ▶
"Would you use my app?" → "Walk me through the last time you dealt with this."
"Don't you hate when X?" → "What was the most annoying part of that?"
Ask about the last time something happened, don't lead them, ask "why," and let them talk more than you.
🔗Links to #8: paste your final 5 questions into the rehearsal prompt to practice them out loud.
prompt linking · role-play
Get comfortable before the real thing. Claude plays your interviewee using the actual questions you wrote in #7, so you practice following the conversation instead of reading off a script.
Copy into Claude →
You're playing a high schooler I'm interviewing about [your problem]. Stay fully in character – give realistic, slightly messy answers like a real teen would, not perfect ones.
These are my interview questions:
[paste your 5 questions from #7]
Run it like a real interview: I ask a question, you answer in 1-2 honest sentences, then quietly suggest in [brackets] the best follow-up "why" I could have asked. When we finish all 5, tell me which of my questions got the weakest answers and why. Ready when I am.
✅You'll get back: a full practice run, plus tips on which questions to improve before the real thing.
context engineering · self-check
Once you've done your interviews, it's dangerously easy to remember only the parts that agreed with you. Use Claude to check your own confirmation bias and separate real signal from polite noise – without letting it do your thinking for you.
🔒
Anonymize first: before you paste your notes, remove names, school names, and anything identifying. Keep it general.
🎓
Straight from class:“here are my interview notes (anonymized). where might I be hearing what I want to hear? what did I forget to ask? which quotes are signal vs. noise?” The prompt below expands that move.
Copy into Claude →
You're a customer research coach helping me stay honest after my interviews.
Here are my anonymized interview notes (no names, no schools):
[paste your raw notes]
Do this:
1. Point out where I might be hearing what I WANT to hear (confirmation bias) vs. what people actually said or did.
2. Separate signal from noise: which quotes describe a real past action, cost, or workaround (signal) vs. a polite opinion or guess about the future (noise)?
3. What did I forget to ask that I should follow up on?
4. In one line, based ONLY on this evidence: does my problem look validated, partly validated, or not yet?
Don't invent insights that aren't in my notes – just help me read what's really there.
✅You'll get back: an honest read on which of your findings are real signal, and what to chase next.
🔗Links to tomorrow: this signal-vs-noise read is exactly the synthesis work we do in Class 3. Bring it with you.
cluster & critique · context engineering
In class you flag by hand first – circling what made you pause, starring what you heard more than once. That judgment is yours; keep it. Then let Claude group the flagged quotes into themes fast, and immediately interrogate what it got wrong. AI finds the pattern; you decide what's real.
🎓
Straight from class:“paste anonymized quotes → group these into themes. then ask: what did it over-cluster, miss, or flatten?” The prompt below is that same move, expanded.
Copy into Claude →
You're helping me synthesize customer interviews. I've already flagged the quotes that made me pause – now help me find patterns.
Here are my anonymized interview quotes (no names, no schools, no employers):
[paste your flagged quotes]
Do this in order, in plain language:
1. Group these into 3-5 themes. Give each theme a short name and list which quotes belong to it.
2. Then critique your own clustering: what did you over-cluster (forced together), miss, or flatten (lost the emotion or the specifics)?
3. Point to the ONE theme with the strongest signal – repeated across people, with real emotion or a workaround attached.
Don't invent quotes or themes that aren't in my notes.
✅ Signal vs. noise – how to weight a theme ▶
You heard it from more than one person.
It's tied to a real past action – they already spent time or money trying to solve it.
There's emotion or a workaround attached.
One polite “that's interesting, I'd probably use that” is noise, not signal. Weight patterns over one-offs.
✅You'll get back: a set of named themes – plus an honest note on where the machine grouping is misleading you.
🔗Links to #2: take your strongest theme into the next prompt to distill the real need underneath it.
prompt engineering · output format
A theme isn't an insight yet. Step 3 asks the detective question: “what is this actually about?” Look beneath the quote for the real need, feeling, or problem. This turns a bucket of quotes into one clean sentence – and keeps you honest about which parts you actually heard.
Copy into Claude →
You're a product strategist helping me distill an insight. Here's my strongest theme from synthesis, with its quotes:
[paste your theme + quotes from #1]
Fill in each of these three frames using ONLY what's in the quotes – and show me the exact quote you based each one on:
1. Users want [...] but feel [...] because [...]
2. The real problem isn't [...], it's [...]
3. This tells us that people are struggling with [...]
Then flag any blank you had to guess at – that's a spot where I still need more interview evidence. Rank the three by frequency (how many people) x intensity (how much it hurts).
✅You'll get back: three sharp ways to name the real need, each traceable to a real quote, ranked by how much they matter.
🔗Links to #3: your clearest distillation becomes the raw material for your problem statement.
make AI argue · seek disconfirmation
Turn the insight into a problem statement – then, before you fall in love with it, have Claude play devil's advocate and find the cracks now, not after you build. This is the whole point of validation: you're stress-testing a bet, not proving your idea is amazing.
🎓
Straight from class:“here's our problem statement + evidence. what's the weakest assumption a skeptic would attack?” The prompt below expands that move.
Copy into Claude →
You're a sharp but friendly startup mentor who plays devil's advocate.
Here's my insight and the evidence behind it (anonymized):
[paste your distilled need from #2 + a few supporting quotes]
Do this in order:
1. Write my problem statement in this exact frame: "Our target users struggle with [problem] because [reason], which results in [situation]." Give me 2 versions.
2. Now argue against it: what's the single weakest assumption a skeptic would attack?
3. Of the people I interviewed, how many actually SHOWED this problem (did something about it) vs. just agreed it exists?
4. Name the ONE thing I'd need to hear in my next interview to be more sure.
Be kind, but don't agree just to be nice – I'd rather find the crack now.
⚠ Before you fall in love – is it valid? ▶
How many interviewees actually showed it (not just agreed)?
What did they already try to solve it?
What's the ONE thing you'd need to hear next to be sure?
Remember: a problem statement is a hypothesis, not a fact.
✅You'll get back: a tight problem statement and a clear list of what still needs proof.
🔗Links to #4: a well-framed problem is halfway to a solution – carry it into How Might We.
prompt linking · insight language
A well-framed problem is halfway to a solution. The frame is “How might we [solve pain] for [target user] so that [outcome]?” The trick is to write several from different angles – so you don't lock onto your very first idea. (In class, Alex's one problem opened into agency, durable skills, mentorship, communication, and clarity.)
Copy into Claude →
Here's my validated problem statement:
[paste your problem statement from #3]
Help me open the solution space with "How Might We" questions. Use this exact frame every time:
"How might we [solve pain] for [target user] so that [outcome]?"
Write me 5 HMWs, each from a DIFFERENT angle – for example: giving the user more agency, building durable skills, connecting them to people or mentors, improving how they communicate, or giving them clarity and feedback. Label each with its angle.
Pull the wording from my interviews where you can, and keep them broad enough to brainstorm from – don't jump to a specific product yet.
☀ Tips for writing great HMWs ▶
Start broad, then narrow – don't jump to a specific solution.
Use insight language – pull directly from what you heard in interviews.
Write multiple HMWs per insight – each problem has many solution frames.
✅You'll get back: five HMWs that each open a different direction to brainstorm from.
🔗Links to #5: pick the one you're most excited about and have AI widen it even further.
widen the space · divergence
This is where AI earns its keep: it's great at generating many options fast. Push for a wide menu of HMW variations, then you choose which one to brainstorm solutions against. Don't let it pick for you – convergence is your job.
🎓
Straight from class:“here's our How Might We + evidence. give us 10 more variations based on the shared context.” The prompt below expands that move.
Copy into Claude →
Here's the How Might We I want to explore, plus the context behind it:
[paste your favorite HMW from #4 and your problem statement]
Give me 10 more variations of this How Might We – same core problem, different framings. Vary the angle (emotional vs. practical, individual vs. community, prevention vs. recovery, self-serve vs. supported) and keep the exact frame:
"How might we [solve pain] for [target user] so that [outcome]?"
After the list, point to the 2 that open the most UNEXPECTED solution spaces and say why in one line each. Don't tell me which to pick – that's my call.
✅You'll get back: a wide menu of directions, plus a nudge toward the surprising ones you might have missed.
📝Your homework: submit your interview synthesis, problem statement, and How Might We statements – that's everything from steps 1–5. Steps 6–8 are a preview of what's ahead.
context engineering · grounding
Your UVP is a clear statement of what your idea is, who it's for, and why it's valuable. The frame is “For [target user] who [struggle with problem], we offer [solution] that [benefit].” The strongest ones borrow the actual phrases your interviewees used – ground it in your notes, not your imagination.
Copy into Claude →
Help me draft a unique value proposition. Use this exact frame:
"For [target user] who [struggle with problem], we offer [solution] that [benefit]."
Here's what I have:
- who I'm serving: [target user]
- the problem, in their words: [paste a real quote or phrase from your interviews]
- my solution idea: [your solution]
- the benefit / how it makes them feel: [benefit]
Write me 3 UVP drafts. In each, stay as close as possible to the actual quote I pasted. Then tell me which draft is the most SPECIFIC and why – vague, could-be-anyone benefits are the enemy.
☀ Why UVPs work ▶
They show why your idea matters, not just what it is.
They connect to real people with real problems.
They help your idea stand out when others sound the same.
They make it easy for someone to say “that's for me!”
✅You'll get back: three UVP drafts grounded in real interview language, with the sharpest one flagged.
🔗Links to #7: now test whether your grounded UVP actually beats one the AI writes blind.
the class activity · AI vs. interviews
This is the activity we run in class: a UVP the AI writes with no context vs. the one you built from real interviews. The gap between them is the entire lesson – and the reason discovery can't be outsourced.
🎓
Straight from class:“AI-generated UVP vs. interview-derived UVP – what's the difference, and why does it matter?” Run it yourself with the prompt below.
Copy into Claude →
I want to compare two value propositions.
FIRST, without any of my interview notes, write a value proposition for a product that helps [one-line description of your space, e.g. "young professionals grow at work"]. Use the frame "For [user] who [problem], we offer [solution] that [benefit]."
THEN here's mine, built from real interviews:
[paste your best UVP from #6]
Now compare them side by side:
1. What does your blind version get generic or wrong because it never talked to anyone?
2. What does my interview-derived version know that you couldn't have guessed?
3. What's the ONE specific detail from my interviews that would make mine hard for a competitor to copy?
✅
The takeaway: AI can phrase your UVP. Only your interviews can ground it. Specificity is the moat.
✅You'll get back: a clear picture of why your discovery work – not the AI – is your real edge.
🔗Links to #8: take the winning UVP and sharpen it into your final one.
make AI argue · self-check
Like your problem statement, your UVP is a bet – the value you're claiming to deliver, not yet proven. Sharpen it now, then you'll test it when you design and build. This one also spits out the 3-sentence pitch your homework asks for.
Copy into Claude →
Here's my current unique value proposition:
[paste your winning UVP from #7]
Be my honest editor:
1. Which part is a claim I HAVEN'T validated yet – a benefit I'm hoping for vs. one I actually heard in interviews? Flag it.
2. Rewrite it to be more specific – swap any generic word ("easy," "better," "simple," "seamless") for something concrete from my interviews.
3. Give me a 3-sentence version I could say out loud: "The problem is... / Our solution is... / It stands out because..."
Keep my voice – don't make it sound like an ad.
✕ vague UVP
"a simple app that makes learning easier"
could be anyone's product
benefit you hope for
✓ sharp UVP
uses a real phrase they said
names the specific user & pain
benefit you actually heard
✅You'll get back: a sharper, more honest UVP and a 3-sentence pitch you can actually say out loud.
🔗When you're ready: refine this with your teammate using exact phrases from your interviews.
ground it · groundingSlides 14–17 · 9:25–9:50
In class you built your persona by hand, from the real quotes you heard – that judgment is yours, and it stays yours. Now bring Claude in to catch where you quietly drifted into assumption. A persona is only as strong as the quotes underneath it.
🎓
Straight from class:“personas must come from your real interviews, not made-up assumptions.” This prompt is how you check that you actually did that.
Copy into Claude →
You're helping me check that my user persona is grounded in real evidence, not assumptions.
Here's my persona (anonymized – no names, schools, or employers):
[paste your persona: name, age, goal, frustration, habits, quote]
Here are the real interview quotes I actually heard:
[paste your anonymized interview quotes]
Do this in order:
1. Go line by line through my persona. For each trait, show me the exact quote that supports it.
2. Flag every trait that has NO quote behind it – that's an assumption I invented, not something I heard.
3. Suggest one question I should ask in my next interview to fill the biggest gap.
Don't invent quotes. If the evidence is thin, tell me plainly.
✅ What makes a persona real, not made-up ▶
The goal is something they told you they want – not what you'd want in their place.
The frustration is a real moment they described, with emotion attached.
The quote is word-for-word from an actual interview.
If you can't point to the interview behind a line, it's a guess. Guesses are fine as questions – just don't design on them yet.
✅You'll get back: a persona where every line is either backed by a quote or honestly flagged as an assumption to test.
🔗Links to #2: take your grounded persona into the journey map.
feel it · perspective-takingSlides 19–23 · 10:00–10:30
In class you mapped the actions – what your user does, step by step. This is where you go deeper into the thoughts, emotions, and pain points that make it empathy and not just a flowchart. Let Claude draft the emotional arc, then you correct it against what you actually heard.
Copy into Claude →
You're helping me turn a bare list of steps into an empathetic journey map.
Here's my user (from my persona):
[paste your grounded persona from #1]
Here are the steps they take to do one key task they struggle with:
[paste your actions, step by step]
For each step, add three things:
- Thought: what's probably going through their head
- Emotion: how they likely feel (one word)
- Pain point: where it breaks down
Mark each one [heard] if my quotes support it, or [guess] if you're inferring it. Then name the single LOWEST emotional point in the whole journey – the moment of most frustration – and say why.
Only use what fits my persona and quotes. Where you're guessing, say so.
🗺 The five layers of a journey map ▶
Actions – what users actually do, step by step.
Thoughts – what's going through their head.
Emotions – how they feel at each step.
Pain points – where it breaks down.
Opportunities – where you could help.
🕵
Keep the human in charge: Claude can guess the feeling, but only your interviews can confirm it. Cross out any [guess] that doesn't match what you actually heard.
✅You'll get back: a journey with an emotional arc, each beat labeled heard vs. guessed, and the worst moment flagged.
🔗Links to #3: that lowest point is exactly where your biggest opportunity is hiding.
argue · pressure-testSlides 24–25 · 10:30–10:45
In class we defined it simply: opportunity = pain point + gap. Your journey map surfaced several painful moments – now find the one worth designing for, and have Claude argue about whether it's really the biggest before you commit.
Copy into Claude →
Here's my user's journey map with the pain points marked:
[paste your journey + pain points from #2]
Help me find the single best opportunity to design for:
1. List every pain point, and rank them by frequency (how often it happens) x intensity (how much it hurts).
2. Name the #1 opportunity as a "How might we..." statement – "How might we [help with pain] for [this user] so that [outcome]?"
3. Now argue against my ranking: what's the weakest assumption in calling this the biggest opportunity? What would change the order?
Use only what's in my journey map. Don't invent new pain points.
✅You'll get back: a ranked shortlist, your #1 opportunity framed as a How Might We, and an honest challenge to it.
🔗Links to #4: now put the value of solving that moment into words – your UVP.
ground it · groundingSlides 26–34 · 10:30–10:45
Your UVP is a clear statement of what your idea is, who it's for, and why it's valuable – the empathy you built, put into words. The frame from class: “For [target user] who [struggle with problem], we offer [solution] that [benefit].” The strongest ones borrow the actual phrases your users said.
🎓
Straight from class:“for [target user] who [struggle with problem], we offer [solution] that [benefit].” This prompt fills that frame from your real interviews.
Copy into Claude →
Help me draft a unique value proposition. Use this exact frame:
"For [target user] who [struggle with problem], we offer [solution] that [benefit]."
Here's what I have, grounded in my persona and interviews:
- who I'm serving: [target user]
- the problem, in their words: [paste a real quote or phrase from your interviews]
- my solution idea: [your solution]
- the benefit / how it makes them feel: [benefit]
Write me 3 UVP drafts. In each, stay as close as possible to the actual quote I pasted – use their words, not fancier ones. Then tell me which draft is the most SPECIFIC and why. Vague, could-be-anyone benefits are the enemy.
☀ Why UVPs work ▶
They show why your idea matters, not just what it is.
They connect to real people with real problems.
They help your idea stand out when others sound the same.
They make it easy for someone to say “that's for me!”
✅You'll get back: three UVP drafts grounded in real interview language, with the sharpest one flagged.
🔗Links to #5: now test whether your grounded UVP actually beats one the AI writes blind.
AI vs. interviews · the moatSlides 26–34 · head-to-head
Here's a test worth running: a UVP the AI writes with no context vs. the one you built from real interviews. The gap between them is the reason discovery can't be outsourced – and why your empathy work is your real edge.
Copy into Claude →
I want to compare two value propositions.
FIRST, without any of my interview notes, write a value proposition for a product that helps [one-line description of your space, e.g. "students find good coffee shops"]. Use the frame "For [user] who [problem], we offer [solution] that [benefit]."
THEN here's mine, built from real interviews and my persona:
[paste your best UVP from #4]
Now compare them side by side:
1. What does your blind version get generic or wrong because it never talked to anyone?
2. What does my interview-derived version know that you couldn't have guessed?
3. What's the ONE specific detail from my interviews that would make mine hard for a competitor to copy?
✅
The takeaway: AI can phrase your UVP. Only your interviews can ground it. Specificity is the moat – and it comes from the human you took the time to understand.
✅You'll get back: a clear, concrete picture of why your discovery work – not the AI – is your edge.
🔗Links to #6: take the winning UVP and sharpen it into your final one.
argue · self-checkSlide 35 → bridges to Class 5
Your UVP – and the solution behind it – is a bet, not proof. Sharpen it, then name the one thing you're most afraid is wrong. That assumption is exactly what you'll test tomorrow, when we build a validation prototype (does the problem hold?) before a product one.
Copy into Claude →
Here's my current unique value proposition and the persona behind it:
[paste your winning UVP from #5 + your persona]
Be my honest editor:
1. Which part is a claim I HAVEN'T validated yet – a benefit I'm hoping for vs. one I actually heard in interviews? Flag it.
2. Rewrite the UVP to be more specific – swap any generic word ("easy," "better," "simple," "seamless") for something concrete from my interviews.
3. Name the SINGLE assumption I'd most need to test before building anything – the one that, if it's wrong, the whole idea falls apart. Phrase it as a yes/no question a quick prototype could answer.
Keep my voice – don't make it sound like an ad.
✕ vague UVP
"a simple app that makes finding things easier"
could be anyone's product
benefit you hope for
✓ sharp UVP
uses a real phrase they said
names the specific user & the exact painful moment
benefit you actually heard
✅You'll get back: a sharper, more honest UVP – plus the exact assumption to test in tomorrow's build.
📝Bring to Class 5: the assumption you named here becomes the question your validation prototype has to answer.
advise · skeptical advisorSlide 14 · build studio
You scoped your MVP as a team – assumption, in/out, success line. Before Claude builds a thing, make it attack that scope and hand you options. The assumption is the anchor: your test and your success line both have to tie back to it.
Copy into Claude →
Be a skeptical advisor, not a cheerleader. Here's my MVP scope:
- assumption: [what must be true or the idea fails]
- in / out: [what I'm building / deliberately leaving out]
- success line: [the keep/kill number]
For each question, hand me 2 options to pick from – not one answer:
1. Is that my riskiest assumption, or a safe one? ("users want this" = safe · "users switch from [X] to get it" = risky) → give me 2 sharper versions, riskiest first.
2. Does my in/out actually test that assumption? If it could pass without proving it, it's the wrong build. → give me 2 ways to retarget it so it can only succeed if the assumption is true.
3. Does my success line prove or kill the assumption, or is it vanity? ("signups" = vanity · "came back day 2 unprompted" = real) → give me 2 sharper metrics tied to the assumption, each with a number.
4. What's the fastest way this test gives me a false positive? → name 2 ways I'd get fooled + a guard for each.
End with my tightened scope in 4 lines. Don't agree with me to be nice.
🎯
Straight from class:“the assumption drives both – if your test or success line doesn't tie back to it, fix that first.”
✅You'll get back: a tightened 4-line scope where your build and your success metric both trace to your riskiest assumption.
🔗Links to #2: carry that sharp assumption straight into finding your core loop – same chat.
interview · one at a timeSlide 15 · build studio
The core loop is the single repeated interaction your whole solution orbits – trigger → action → payoff. Don't let Claude guess it for you. Have it interview you, push back when you're vague, and pull the loop out of your own head.
Copy into Claude →
Work with me as my product advisor. I'm finding the CORE LOOP of [your solution] for [user] – the single repeated interaction that delivers the value.
Ask me these one at a time, in sequence. After each answer, push back if it's vague or too big, and ask your own follow-up before moving on. Don't advance until my answer is sharp:
a. trigger – what moment makes a user open this?
b. action – the single thing they do?
c. payoff – the second they feel the value?
d. if I deleted everything but this loop, what dies? what survives?
Don't design UI or write code yet – just interview me.
🔄 What makes a loop worth repeating ▶
The trigger is a real moment in the user's day, not "they remember to open the app."
The action is one step – if it takes three, the loop is too heavy.
The payoff lands fast enough that they'd come back and run it again tomorrow.
✅You'll get back: a rough core loop – trigger → action → payoff – with the hand-waving caught.
🔗Links to #3: now hold that rough loop up against loops that already work.
borrow & bend · compareSlide 16 · build studio
The fastest way to see your own loop is next to loops that already work. Have Claude reverse-engineer the closest proven consumer apps, show where yours is weaker, and offer sharper versions – then land on one committed sentence you'll build from.
Copy into Claude →
Using the rough loop we just landed on:
1. Show me the core loop of 3 successful consumer apps closest to [your solution] – each as [trigger] → [action] → [payoff] in one line, and what makes it worth repeating.
2. Compare mine to them – where is mine weaker, slower, or less repeatable?
3. Give me 2 sharper versions of my loop modeled on the best of them, and ask me which fits my user better.
Then help me commit to one final sentence – the loop, its "aha," and the assumption it proves.
💡
Straight from class:“the best loops are borrowed then bent to your user.” You're not copying – you're learning what makes a loop repeatable, then making it yours.
✅You'll get back: your core loop in one committed sentence, benchmarked against loops that already work.
🔗Links to #4: design the one screen that runs this loop.
interview · master UI designerSlide 17 · build studio
Now Claude switches hats to master UI designer – but still no code. It interviews you on how the one screen looks and feels, then writes it back as a tight text wireframe you'll hand straight to the build step.
Copy into Claude →
Continue as my product advisor and master UI designer. Using the core loop we just locked, help me design the one screen – still no code.
Ask me these one at a time, and push back if my answer is vague or adds scope:
a. what's on the screen at each step of the loop, top to bottom?
b. what's the first thing a user sees in 5 seconds – and what do they tap first?
c. what does each state look like: empty, in-progress, success?
d. where does the payoff show up so they actually feel it?
When we're done, write it back to me as a tight text wireframe – and flag any scope creep that snuck in.
✅You'll get back: a tight text wireframe of the one screen – designed in words, ready to build.
🔗Links to #5: hand that wireframe back and turn it into a working prototype.
hold the line · refuse scope creepSlide 18 · build studio
Now you build – only the loop, nothing around it. Real content, a payoff that lands, and a tiny feedback capture at the end so you learn something when a real person touches it.
Copy into Claude →
Now build the core loop as a working artifact from the wireframe we just wrote.
- Build ONLY the loop. Use real content, not lorem – it has to be reactable.
- Make the payoff moment obvious and satisfying.
- Bake in a tiny end-of-loop capture (1 rating + 1 line of text), logged so I can review after class.
- Refuse to add login, onboarding, settings, or nav – if I ask, remind me they don't test the loop.
When it's built, tell me the ONE thing to watch a real user do that proves the loop works.
🚀
Straight from class:“rough and real beats polished and hypothetical – get it into a human's hands.”
✅You'll get back: a working prototype of your core loop – real enough to hand another team.
📝Bring to the test round: hand it to another team, stay quiet, and watch the ONE thing that proves the loop works.
Copy into Claude blank template
Act like Angela: a direct, sharp startup mentor who doesn't let vague
answers slide. Run this as a real back-and-forth, not a report.
My idea: [one sentence — what it does + who it's for, no real names].
My gut answer to "who pays" is: [B2C / B2B / B2G / B2B2C / C2C].
Don't analyze it for me yet. First ask me to explain WHY in my own
words — who exactly hands over the money, how long the sales cycle is,
how many people have to say yes. Wait for my answer.
Then, based on what I actually said, ask me the ONE question that pokes
the biggest hole in my reasoning. Wait for that answer too before you
go further. Only after I've defended myself at least twice, tell me
honestly whether my pick holds up or whether a different type fits
better — and why.
If my first answer is fuzzy, don't move on — make me sharpen it first.
✅ You'll get back: a real challenge to your first instinct, not just a rubber stamp — plus the questions you need answered before you can defend it.
🔗 Next → Pick a revenue model: who pays shapes which of the three revenue categories will actually fit.
Copy into Claude blank template
Act like Angela: direct, a little skeptical, allergic to founders picking
a model because it's trendy instead of because it fits. This is a live
back-and-forth — ask, wait for my answer, then respond to what I said.
My idea: [one sentence]. My customer type from Step 1: [B2C / B2B / B2G /
B2B2C / C2C]. My gut pick: [transactional / recurring / facilitation, +
freemium layer if relevant].
Don't evaluate my pick yet. First ask me to argue FOR it myself: why
this model, and specifically why not the other two. Wait for my answer.
Then push on whatever's weakest in what I said — is this really about my
customer, or is it what everyone else is doing? What does it assume
about my cash flow that I glossed over? Ask me one hard question at a
time and make me answer before the next one.
Only after I've defended my pick through at least two rounds, tell me
straight whether it holds up, and name the single hardest question I
still need to answer before I lock this in.
✅ You'll get back: your pick pressure-tested against the trend trap, the cash-flow assumption named, a real counter-model, and the one question you can't dodge.
🔗 Next → Price it two ways: now put a real number on it.
Copy into Claude blank template
Act like Angela: you think most founders under-price out of fear and
call it "being reasonable." Don't let me do that — and don't just hand
me a number. Make me work for it.
My idea: [one sentence]. My revenue model: [from Step 2].
First ask me to state, in my own words, what I think it costs to
serve one customer AND what it's actually worth to them — what it
saves them, replaces, or means to them. Wait for my answer before you
calculate anything.
Once I've answered, give me the cost-plus floor and the value-based
ceiling from what I said. Then ask me: which number inside that range
would I actually charge — and why that one? Push back on my answer if
it sounds like I'm playing it safe. If I flinched while describing what
it's worth, say so — that flinch is usually the real number trying to
get out.
End by giving me the one objection a real customer would throw at
whatever I land on, unsoftened. I decide the final number.
✅ You'll get back: your floor, your ceiling, the number you're most likely to underprice, and the real objection to expect.
🔗 Next → Pressure-test: a price from a spreadsheet is still a guess. Now put it in front of a skeptical user and see if it holds.
Copy into Claude blank template
First ask me who exactly I'm pricing this for and why I picked the
number I did. Wait for my answer.
Then, fully in character, become that persona — not "an AI helping
with pricing," an actual person with their own budget, priorities, and
skepticism. Stay in character until I say "break character."
We're a [one-line idea]. We're considering charging [your price] on a
[your revenue model + any secondary category or freemium layer].
As that persona, answer honestly, with real reasoning, not just a
number:
- what's a fair price for this?
- what would feel like too much?
Then let me push back and defend my price to you, still in character —
argue like a real person would, don't cave just because I pushed.
After a few rounds, break character. As Angela, give me the honest
debrief: did my price survive contact with a real (if simulated) user,
where did I get defensive instead of listening, and what's the ONE
thing from this conversation I should actually go verify with a real
person before I trust it?
✅ You'll get back: a persona that pushes back for real, a price tested under real objections, and an honest debrief on where you got defensive.
🔗 Next → Understand your costs: you have a price. Before you can know if it works, get honest about what this actually costs you to run.
Copy into Claude blank template
Act like a startup financial coach for first-time, young/student
founders — someone who's seen a hundred teenage founders forget the
same costs. Blunt, practical, no jargon. Don't dump a list on me yet.
My idea: [one sentence]. My revenue model + price: [from earlier steps].
First, ask me to write my OWN first-pass list of everything I think
this costs me to run — fixed, variable, and my own time. Wait for my
answer before you add anything.
Then, as my coach, go through what I missed. Specifically check for the
costs young/student founders almost always forget:
- payment processing fees (Venmo/Stripe/PayPal cuts on every sale)
- what happens when I outgrow a free tier (Canva, Claude, hosting,
app store fees)
- materials, packaging, and shipping if it's physical
- my own unpaid hours — building, testing, messaging users, making
content — and what happens to my costs if I actually valued that time
- one-time costs that sneak in later (a domain renewal, a template,
a logo)
- taxes or fees if money changes hands through an app
For each one you flag, ask me whether it applies to MY idea before
adding it — don't pad my list with costs that don't apply. End with my
full list sorted into fixed / variable / time, each with a real dollar
estimate.
✅ You'll get back: your own first-pass list, coached and filled in with the costs founders your age typically forget — sorted into fixed, variable, and time, with real numbers.
🔗 Next → Unit economics: bring this cost list with you — now find out if one customer actually makes you money.
Copy into Claude paste your cost list from Step 5
Act like Angela: revenue is vanity, profit is sanity. Be a skeptic, not
a cheerleader — and don't do the math until I've tried first.
My idea: [one sentence].
My cost list from Step 5 (fixed, variable, time): [paste it].
What I'd earn from one customer: [your price from Step 3/4 × how often].
First, ask me to do the subtraction myself — revenue-per-user minus
cost-per-user — and tell you whether I think it's positive or negative
before you check my work.
Then go after what I said:
1. Check my math against MY numbers. Did I get it right?
2. Ask me to defend the single number in my cost list I'm least sure
of — the one the whole thing rests on. Push until I either justify
it or admit it's a guess.
3. If it's negative: ask ME first which lever I'd pull — raise the
price, lower the cost, or add value — before you weigh in on whether
that's realistic for my idea.
I'd rather you find the hole now than I find it after I build.
✅ You'll get back: your math checked, your weakest assumption named, and the one realistic fix if it doesn't work — using your own cost list, not a generic one.
🔗 For your Canvas homework: this is exactly the reflection you'll submit — does revenue cover costs, and if not, what changes?
"High schoolers" is a demographic, not a customer. A beachhead is a group so specific you could physically reach every one of them this week. Your persona describes a kind of person. This step turns that into a list of actual humans — and that list is the only thing a GTM plan can point at.
the differencePersona: "Maya, 16, overwhelmed by studying, uses her phone constantly." → Beachhead: "the ~40 students in my school's AP Bio group chat, plus the 6 people in my study group." One you can describe. The other you can text.
Prompt 1 — narrow it until it hurts blank template
Act like Angela: a direct, sharp startup mentor who doesn't let vague
answers slide. Run this as a real back-and-forth, not a report.
My idea: [one sentence — what it does + who it's for, no real names].
My user persona from class: [paste it — behaviours and frustrations, no real names].
My gut answer to "who exactly" is: [your first attempt at a group].
Don't analyze it for me yet. First ask me: could I name TEN of these
people by name, right now, without looking anything up? Wait for my
honest answer.
If the answer is no, don't move on — make me narrow it. Keep asking
"which ones, specifically?" until my group is small enough that I could
reach every single person in it before Friday. Cut me off if I widen it
back out to sound more impressive.
Then, and only then, ask me the ONE question that would prove I actually
know this group — something a real insider could answer and a stranger
couldn't. Wait for my answer before you judge it.
A smaller group I can reach beats a bigger one I can only describe.
✅ You'll get back: a beachhead narrow enough to act on — and an honest read on whether you actually know these people or just imagined them.
Prompt 2 — turn it into the 20-list paste your narrowed group from Prompt 1
Same session. My narrowed beachhead: [paste it].
I need a list of 20 real, reachable people or groups — but I have to
write it, not you. So coach me through it instead of guessing for me.
1. Ask me to list every PLACE this group already gathers — physical and
online. Wait for my list. Then ask what I left out.
2. For each place, ask me roughly how many of my people are there and
whether I can actually get in. Push back if I'm assuming access I
don't have.
3. Ask me who the 2-3 GATEKEEPERS are — the people who could put me in
front of the whole group at once. Wait.
4. Now ask me to count: across all of that, can I realistically get in
front of 20 of these people before Friday? Make me say a real number.
If my number is under 20, don't reassure me — tell me straight whether
the problem is that my group is too narrow, or that I don't actually
have access to it. Those are different problems with different fixes.
Do NOT invent names or places for me. Ask, wait, then push.
✅ You'll get back: your own 20-list, plus the honest verdict on whether you have real access to this group or you've been assuming it.
🔗 Next → What do we say: now that you know exactly who you're talking to, you can stop talking about yourself.
Your message is not a description of your product. It's a description of their problem, in their words — the ones you actually heard in interviews. Then it compresses into a hook: the first 7 words a real person sees before they decide to keep reading or scroll past.
the one-liner formula
For [who], who [the problem, in their words], our thing [does the one thing]. Unlike [what they do today], it [the difference that matters].
Prompt 1 — sharpen the UVP into the one-liner paste your beachhead + your interview quotes
Act like Angela: direct, allergic to founders describing their product
when they're supposed to be describing their user's problem. Real
back-and-forth — ask, wait, respond to what I actually said.
My beachhead: [paste it — no real names].
My UVP: [paste it].
Exact phrases real users said in my interviews: [paste 3-5 quotes, anonymized].
Step 1: read my UVP back to me and tell me how much of it is about MY
PRODUCT versus THEIR PROBLEM. Give me a rough percentage. Be harsh.
Step 2: ask me which of my interview quotes I'd be willing to put in the
message word-for-word. Wait for my answer. If I don't pick one, ask me
why I trust my own phrasing more than my user's.
Step 3: now make me fill in the formula out loud, one blank at a time —
"for [who], who [problem in their words], our thing [does one thing];
unlike [what they do today], it [the difference]." Push back on any blank
where I slip back into product language.
End with my one-liner in MY words, using at least one phrase a real user
actually said. Not your words. Mine.
✅ You'll get back: a one-liner grounded in real interview language instead of founder language — and a clear read on how much of your UVP was secretly about you.
Straight from class"AI is fast at options and terrible at voice. So use it in two moves — never one." Move 1: get 5 hooks and ask which is the most generic. Move 2: paste how you actually talk, and make it sound like you. Move 2 is the part that matters.
Prompt 2 — the hook: generate wide, then take it back the in-class prompt, expanded
Two moves. Don't skip the second one.
MOVE 1 — options.
My one-liner: [paste it from Prompt 1].
My user: [beachhead — no real names or school names].
Write 5 hooks — first 7 words only. Every one must open with THEIR
PROBLEM, not my product. Make them sound like a real teenager texting a
friend, not an ad. No exclamation marks. No "revolutionise." No
"game-changer."
Then tell me which of the 5 is the most GENERIC — the one that could
belong to any app — and why.
MOVE 2 — take it back.
Now here's how I actually talk: [paste 2-3 real messages you've sent —
group chat, caption, anything, names removed].
Pick my favourite hook: [the one you chose].
Rewrite it in MY voice. Cut any word I'd never say out loud. Keep it
under 7 words.
Then tell me exactly what you had to strip out to make it sound human —
I want to see the difference between AI-generic and me.
Last: give me one test I could run this week to find out if the hook
works, that doesn't involve asking a friend if they like it.
✅ You'll get back: 5 hooks, the generic one named, and your favourite rewritten in your own voice — plus a visible before/after of what "generic" actually looks like.
🔗 Next → Through what path: you have a message. Now find the road it travels down.
Every channel on earth is one of four things, sorted by who takes the last step toward the user. Pick two — from different letters. If both your channels are the same mechanism, you've made one bet twice.
A · you go to them
You make the first move.
DMs, tabling, announcements, cold email, paid ads
B · they come to you
You leave something findable.
Reels, YouTube, posters, newsletter, search
C · someone brings them
You borrow trust.
a teacher, a club lead, a creator they follow
D · the product brings them
Using it spreads it.
referrals, invite-a-friend, shareable outputs
"Word of mouth" is not a channel — it's a scoreboard. It's what happens when C or D works. You can't pick it, you can only earn it. If it's on your list, replace it with the mechanism that would actually cause it.
Prompt 1 — where are they already, and what does it cost to try? paste your 20-list
Act like Angela: direct, and deeply suspicious of founders who pick the
channel that's easiest for THEM instead of the one their user is
actually on. Ask, wait, respond — don't hand me a plan.
My beachhead + where they gather: [paste it, no real names].
My message/hook: [paste it].
The four mechanisms are:
A – you go to them (DMs, tabling, announcements, ads)
B – they come to you (content, posters, search)
C – someone brings them (a teacher, club lead, creator they trust)
D – the product brings them (referrals, invites, shareable outputs)
1. Ask me which mechanism I'm instinctively drawn to. Wait. Then ask me
the uncomfortable question: is that where my USER is, or is it where
I'M comfortable? Make me actually answer.
2. Walk me through all four, one at a time. For each, ask ME to score it
1-5 on: REACH (how many of my 20), TRUST (do they believe it there),
and COST-TO-TRY (could I test it this week for under $10 and 2 hours).
Don't score them for me — ask, wait, then challenge any score I can't
justify.
3. Make me pick TWO, from DIFFERENT letters. If I pick two from the same
letter, tell me I've made one bet twice and make me choose again.
End with my two channels and one sentence each on why my 20-list is
ALREADY there — not why they should be.
✅ You'll get back: two channels from different mechanisms, scored by you, with the "is this where they are or where I am?" question answered honestly.
Prompt 2 — now attack them skeptical advisor
Be a skeptical advisor, not a cheerleader. Same session.
My two channels: [paste them + their letters].
My user: [beachhead]. My hook: [paste it].
Argue AGAINST me. Be specific and be harsh:
1. Give me the strongest reason each of these two channels will FAIL to
reach my user. Not generic risks — reasons specific to this group.
2. Which of the two is the lazy pick — the one I chose because it's easy
for me? Name it.
3. What's the ONE channel I dismissed that might actually be the best
one, and what am I afraid of about it?
4. If I only get ONE shot this week, which of the two should I run, and
what's the cheapest possible version of it — under $10, under 2 hours?
Then ask ME whether I still stand by my two picks. Don't let me answer
with "I think so" — make me give a reason.
Don't agree with me to be nice.
✅ You'll get back: the specific reason each channel could fail, your lazy pick named out loud, and the cheapest real version you could run this week.
🔗 Next → How do they get it: the channel gets them to look. Something else has to get it into their hands.
The channel gets them to look. The motion gets it into their hands — self-serve, sales-led, or partner-led — and teams lose almost everybody in the gap between those two things. Then the ask: the single action you want, small enough to say yes to and costly enough to mean something.
A compliment is not signal
"That's so cool." "I'd totally use that." Likes. Views. A friend saying yes to be nice. None of it costs them anything — which is exactly why it tells you nothing.
Real signal costs them something
Money — a $3 pre-order. Time — a booked 10-minute slot. Data — their actual email, typed. Reputation — they invited a friend.
Prompt 1 — walk the four steps from "sees it" to "using it" paste your channels
Act like Angela: direct, and obsessed with the step where founders
quietly lose everyone. Ask, wait, respond.
My user: [beachhead, no real names]. My channel: [the one you'd run first].
My product: [one sentence].
The three motions are:
- self-serve (they find it and start alone — it must be dead simple)
- sales-led (you have a conversation and they say yes)
- partner-led (someone they trust hands it to them)
1. Ask me which motion I'm using and WHY that one fits this user. Wait.
If my user is a parent, teacher, school, or business, push me hard on
whether self-serve is realistic or wishful.
2. Now make me write the EXACT four steps from "sees it" to "using it."
One step at a time. After each one, ask: what does the user have to
DO here, and how long does it take them? Don't let me skip the boring
steps — those are where people leave.
3. When I've got all four, ask me to point at the step where I'll lose
the most people. Then tell me if I picked the right one, and why.
4. Find any step that asks a stranger to do something annoying for free.
Name it. That's my leak.
End with my four steps rewritten so the leak is smaller.
✅ You'll get back: a four-step path with the leak found and named — usually a step you didn't think counted.
Prompt 2 — is your ask real signal, or a compliment? signal vs. noise
Be a skeptical advisor. In this course we've spent two weeks separating
signal from noise. Now do it to my go-to-market.
My four steps from "sees it" to "using it": [paste them].
The single action I want from a user: [write your ask, word for word].
1. Tell me straight: does my ask cost the user MONEY, TIME, DATA, or
REPUTATION? If it costs them none of the four, say so plainly — I'm
collecting compliments, not evidence.
2. If it's a compliment, give me 3 alternative asks that cost one of the
four currencies but are still small enough that a stranger would
actually say yes. Rank them by how much they'd tell me.
3. Ask ME which one I'd be nervous to send. Wait for my answer — then
tell me whether the nervous one is the honest one.
4. Finally, write my ask as the exact words I'd say or type. Under 25
words. No hedging, no "would you maybe possibly."
The ask I'm scared to make is usually the only one worth making.
✅ You'll get back: an honest verdict on whether your ask produces evidence or applause — plus the exact words to send.
🔗 Next → How do we know: you have a plan. Now make it falsifiable.
You wrote your MVP as an experiment. Do the same here. If there's no number that would disappoint you, you're not testing — you're hoping.
the test sentence
If we [do this, in this channel, by this day], we predict [this many see it] → [this many act]. Below [number], we were wrong — and we change [what].
Prompt 1 — commit to a number you could fail paste your answers to Q1–Q4
Act like Angela: I'd rather you find the hole now than a real user find
it on Friday. Don't do the thinking for me — make me commit first.
My GTM so far:
who: [beachhead, no real names]
message: [one-liner + hook]
channel I'm running first: [your pick]
the ask: [the one action]
1. Before you say anything, ask ME to predict the numbers: how many
people will SEE it, and how many will ACT? Make me say two real
numbers out loud. Don't let me say "some."
2. Then ask me the number that would make me admit I was WRONG. Not the
number I'm hoping for — the number that would make me change the plan.
Wait. Push if I pick something so low I couldn't possibly fail.
3. Now check my prediction against reality: given my channel and my
20-list, is my "sees it" number even physically possible? Say so if
it isn't.
4. Write my whole test as one sentence, with my numbers and a day of the
week: "If we ___ by ___, we predict ___ see it → ___ act. Below ___,
we were wrong, and we change ___."
If I can't fail this test, it isn't a test. Tell me if that's what I've
written.
✅ You'll get back: one falsifiable test sentence, with your numbers, runnable this week — and a straight answer on whether it's actually a test.
Straight from class"Be a skeptical advisor, not a cheerleader." Argue why these channels won't reach my user. Where will I lose the most people? Is my ask real signal or a compliment? What's the cheapest test I could run this week? "If it only says nice things, your prompt was too soft. Ask it again, meaner."
Prompt 2 — the pre-mortem: read the result before you get it the in-class prompt, expanded
Be a skeptical advisor, not a cheerleader. Here's my full go-to-market:
who: [beachhead — no real names or school names]
message: [one-liner + hook]
channels: [2 channels]
motion + ask: [how they get it + the one action]
test sentence: [paste it from Prompt 1]
It's Friday. The test FAILED. Walk me through the four ways that could
have happened, and for each one, tell me which part of my plan was
wrong and what I'd change:
nobody saw it → wrong CHANNEL
saw it, didn't click → wrong MESSAGE
clicked, didn't act → wrong ASK, or the motion is broken
acted, never came back → the PRODUCT (and that's a bigger problem)
For each: what would that look like in MY numbers specifically, and
what's the ONE thing I'd change first?
Then tell me which of the four you think is most likely for my plan, and
why. Be specific. Don't hedge.
Finally: what's the cheapest thing I could do BEFORE Friday — under $10
and under 2 hours — that would tell me which of those four is coming?
Don't agree with me to be nice.
✅ You'll get back: a diagnostic you can run the moment the result lands — so a failed test tells you which of the five questions to fix, instead of just feeling bad.
Before you upload your GTM — the five-question check
Question
You pass if…
Who exactly
☐I can name 10+ of them without looking anything up.
What we say
☐a stranger can repeat the problem back after hearing my hook.
Through what path
☐two channels, from different mechanisms — and my users are already there.
How they get it
☐four steps written down, and I've named the one where I'll lose people.
How we know
☐a number that would disappoint me, and a day of the week.
🔗 For your homework: this is the one-page GTM you upload — plus your post drafted with AI, then rewritten in your voice. Submit both versions so we can see the difference.