Validate Your Startup Idea with AI Before Spending Big on a Product｜Coach Jiang

What this article is about

The biggest risk with a startup idea is that once you build it, no one actually wants to use it. This article lays out a "Real Needs Investigation Method": three questioning principles, the VJPD validation framework, commitment-signal scoring, plus a ready-to-copy AI advisor prompt, so you can see the risk before you invest. The principles come from Rob Fitzpatrick's classic book The Mom Test.

Who this is for

· Founders: afraid no one will buy the idea and the money goes to waste
· Product managers: need to surface users' real pain points, and are done with polite talk
· Marketing and project people: to design an offer that moves people, you first have to know what customers really think

What you can take away

· Three questioning principles that stop you from fishing for polite talk
· The VJPD framework: a complete flow from hypothesis to interviews, surveys, and iterative decisions
· A 0-to-4 commitment-signal scoring scale, and an AI advisor prompt

Admit one thing firstThe biggest enemy of founders and product managers is often themselves. We fall in love with our own ideas far too easily, and once we slip into that self-hyping state, we unconsciously look only for evidence that supports us, then collect a pile of false positive signals from the well-meaning white lies of friends and family.

Three questioning principles: stop fishing for polite talk

Principle 1Talk about their life, not your idea

Users are experts on their own life, but amateurs on your product. Talk about what has already happened and you get facts; talk about your idea and you get only guesses.

Principle 2Ask about the concrete past, not the abstract future

People are bad at predicting their own future behavior. The money and time they actually spent in the past to solve the problem is the hard evidence.

Principle 3Dig into problems and costs, don't collect opinions and praise

A question that costs the user nothing is a fake question. Praise won't build a business; complaints will.

The contrast is where it really lands:

Bad question: "What do you think of an app that automatically organizes meeting minutes?" Good question: "How did you process the meeting minutes after the last important meeting? How long did it take?"
Bad question: “If this feature were available, would you be willing to pay for it?” Good question: “In the past year, have you paid for any tools to solve this problem? What is your budget?”
Bad question: “Do you think this feature is important?” Good question: “What were the direct consequences of this problem last time it happened?”

A real case: I almost jumped straight to a solution

A director at a cultural and educational foundation came to me: "Can you help us design a more automated system to collect audience feedback after our talks?" My immediate answer was sure, collect it through the official LINE account, have AI summarize it, and I would write up a proposal when I got back.

Then I stopped to think: was it really right to hand over a solution like that? Do audiences actually have thoughts after a talk? Do they even want to share them? Who do they actually share them with? I realized that before designing any proposal, I had to figure out what was really going on. So I switched to doing interviews first, with one core mindset: I am here to understand the problem, so I put the sales pitch down.

Interview opening (to the client):
"Before we decide which tool to use, we probably need to
understand the audience's real situation and thinking more deeply,
so that what we build is something they will actually use."

Dig for motivation: what specific event first made you feel this mattered?
Look at the cost: if this feedback never came in, what is the biggest headache or regret?
Check the status quo: how do audiences most often share their thoughts right now?
Define success: suppose it is working well in three months, what does that look like?

The core sequence of the whole strategy is: validate the problem first, then design the process, and only then choose the tool. Exactly the opposite of most people's instinct.

Schedule for the Chiayi Youth Entrepreneurship course series; 'Validating Startup Ideas with AI' is one of the sessions — Schedule for the Chiayi Youth Entrepreneurship course series: "Validating Startup Ideas with AI" is one of the sessions, and this article lays out exactly that method

VJPD Framework: Turning Verification into Four Steps

VJPD is four validation dimensions: V, Validate the problem (is the pain point real); J, Judge the impact (how big is the cost); P, Probe the behavior (how do they solve it now); D, Demographics (who is this). In practice it is four steps:

Step zero: define the core hypothesis in one sentence, spelling out the target user, pain point, frequency, time cost, and existing solution:

I assume that for college instructors or lecturers,
compiling and summarizing after-class feedback happens at least once a week,
takes more than 90 minutes on average,
is currently handled with Excel and message screenshots,
and has a duplicate-and-useless content rate above 50%.

Step one, qualitative interviews: find 5 to 8 target users and run 30-to-60-minute semi-structured interviews, focused on capturing their exact words and concrete situations. Step two, a quantitative survey: turn the interview findings into checkbox options for frequency, time cost, and pain points, and test them against more than a hundred people. Step three, analyze the commitment signals:

0 points (worthless): "What a great idea!"
1 point (interest): "Sounds good, I might use it."
2 points (time commitment): "I'll set aside 30 minutes next week to see a demo."
3 points (reputation commitment): "I can introduce you to other teachers in the department."
4 points (money commitment): "Is there an education plan? I can pay to try it right now."

Step four, iterate the decision: if signals of 2 to 4 points make up more than 60%, keep pushing forward; if 1-to-2-point signals dominate but point to a different pain point, adjust direction; if 0-to-1-point signals are the majority, drop it decisively and save the money for the next idea. And an MVP doesn't have to be a finished product first. A survey, a sign-up page, a small class, a single interview, anything that lets you read the commitment signals counts.

Hand the whole method to an AI advisor

The framework above is a bit heavy to take in on a first read, let alone apply straight away to the problem in front of you. So I packaged it into an AI advisor prompt. Copy and paste it, and the AI will walk you through the thinking as you go (a model with reasoning is recommended):

You are an AI advisor specializing in the "Real Needs Investigation Method."
Your mission is to keep me from falling into the feel-good trap during user research.

When I give you an interview outline or a draft survey:
diagnose which questions violate the three principles, say which one each breaks,
and give me a drop-in improved version.

When I only have a vague idea:
guide me step by step to fill in the background, and use the VJPD framework
(Validate the problem, Judge the impact, Probe the behavior, Demographics)
to generate a professional draft set of interview questions.

When I bring back feedback, score it on the commitment signal (0 to 4 points),
and recommend: keep pushing, adjust direction, or drop the idea.

It does two things: as your "question quality inspector," it rewrites each polite question into one that digs out real behavior; as your "interview strategist," it builds you an interview plan from scratch. The full prompt and examples are in the Real Needs Investigation Method skill package (free to download).

AI simulation cannot replace real peopleAI is great for helping you break down assumptions, design questions, organize interview notes, and play the skeptic for a first round of stress testing. If you want to go further and use AI to simulate consumers' purchase intent (the SSR method), see AI Is a Microcosm of the Entire Market; its companion AI Consumer Validation skill package is open source too. The two packages are a twin design: AI stress testing first, real-person validation after. But the model's answers can only serve as a thinking aid; market validation still has to come back to real people, real situations, and real commitments. The pragmatic order is: use AI to get the idea clear first, then use real-person interviews and small experiments to gather signals, and only then decide whether to scale up your investment.

How to start: write down three hypotheses first

Before building any product, do these four things:

Break the idea into three testable hypotheses, each written as one sentence in the format "who, what pain point, how often it happens, how big the cost is, and how they solve it now."
Paste the hypotheses to the AI advisor, ask it to design interview questions, and weed out the ones that violate the three principles yourself first.
Talk to 5 to 8 real target users about their past experiences, and write down their exact words.
Score with the commitment signals and face the results honestly: did you get praise, or a commitment of time, reputation, or money?

This is far steadier than pouring everything into a finished product from the start. Having an idea rejected by a validation framework is much cheaper than having a product rejected by the market.

Startup ValidationReal Needs InvestigationVJPDCommitment SignalsMVPAI AdvisorUser Interviews