Side HustlesBeginner22 min read

Talk to ten customers: interview scripts with AI

What to learn from an interview, finding ten people without spamming, rules of a good question, a reusable script, running the call and taking notes, spotting polite yeses, and turning ten conversations into an evidence table.

Talk to ten customers: interview scripts with AI

Talk to Ten Customers: Interview Scripts With AI

The cheapest way to test a business idea needs no website, no product and no money: ten honest conversations with the kind of person you want to serve. Most beginners skip it, because talking to strangers is uncomfortable, and because AI makes it tempting to "research" the market from a chat window instead. AI can tell you what people in general tend to say. It cannot tell you what your ten people actually did last month, what it cost them and whether they would pay you. The other danger is the polite conversation: friends and strangers say "that sounds great!" and then never buy. This tutorial teaches you to run short interviews that collect facts instead of compliments. You will learn to ask about the past rather than the future, to find ten people without spamming anyone, to use AI to prepare a script and review your notes (but not to replace the people), to spot the polite yes that really means no, and to turn ten conversations into a simple evidence table.

Difficulty: ★★☆☆☆ (no technical background; some courage)
Tools needed: A notes app, a calendar and any chat AI for preparing and reviewing
Last updated: November 2026

Overview

An interview in this tutorial is a fifteen to twenty minute conversation with one person who might have the problem you want to solve. Your goal is not to pitch. It is to learn three things: whether they have the problem, what they do about it today and what that costs them in time, money or frustration.

This works because behaviour in the past is much better evidence than opinions about the future. If you ask "Would you use an app that tidies your receipts?", nearly everyone says yes: it costs them nothing to say so. If you ask "Tell me about the last time you did your receipts, what did you do and how long did it take?", you hear what actually happens, including the workaround they built, the money they already spend and the moments they are irritated. Someone who has built a workaround or paid for a solution has a real problem. Someone who has never done anything about it probably does not.

There is a second reason to talk first: it is cheap to be wrong early. A week of conversations can save you months of building the wrong thing.

AI is useful here in specific ways, and harmful in others. Useful: drafting a short script, finding where your questions are leading, turning messy notes into a tidy table and finding patterns across ten sets of notes you have written. Harmful: inventing "customers" to interview, answering as a "typical customer" and treating the result as research. A model's answers describe a statistical average of text, not the people you can reach. Use AI as a coach and note-taker. The evidence has to come from humans.

This tutorial has seven steps. First, what you are trying to learn and how to write it as questions. Second, how to find ten people respectfully. Third, the rules of a good question, with before-and-after examples. Fourth, a script you can use. Fifth, how to run the conversation and take notes. Sixth, how to spot polite yeses and other weak signals. Seventh, how to turn ten conversations into an evidence table and a decision.

A note on honesty and privacy. You are asking people for their time. Be clear that you are not selling, ask permission before taking notes, do not record without consent, and do not paste identifying details into an AI tool without their agreement and a check of the tool's privacy terms. This tutorial gives no guarantee about outcomes. Nothing here is legal advice. Rules on contacting people, especially for marketing, differ by country, so check them where you live.

Who This Is Useful For

  • People with an idea who have not yet spoken to a possible customer.
  • People who have asked friends "what do you think?" and got nothing useful.
  • Anyone who wants evidence before spending money or weeks of evenings.
  • What You Will Learn

  • What to learn from an interview, and what not to ask.
  • How to find and approach ten people without spamming.
  • The rules of a good question: the past over the future, specifics over opinions.
  • A reusable interview script, adapted for service, product, content and software.
  • How to use AI to prepare and review, without replacing real people.
  • How to spot compliments, "polite yeses" and fake demand.
  • How to turn ten conversations into an evidence table and a decision.
  • What You Need

    1. Your lane choice and limits sheet from the previous tutorials, if you have them. If not, one sentence on who you want to help and with what.
    2. A notes app and a calendar.
    3. About three hours spread over a week for the first round.
    4. A chat AI, optional but useful.

    Information as of 14 November 2026. Tool features and privacy terms change. This tutorial names no specific tools. Check the current terms of any tool before you share notes with it, and check local rules on contacting people for business.

    Step 1 — Decide what you need to learn

    Before writing any question, write what you are trying to find out. Interviews without a goal turn into friendly chats that teach you nothing.

    Your starting belief. Write one sentence: "I believe that [specific group of people] have the problem of [problem], and would pay to solve it." Make the group specific. "Freelancers" is vague. "Freelance designers in my city who invoice by hand" is something you can find.

    Three questions the interviews should answer.

    1. Is the problem real? Does it happen often enough and hurt enough that they have done something about it?
    2. What do they do today? The current solution may be a spreadsheet, a paid tool, an assistant, a workaround or nothing. This is your real competition, including "do nothing".
    3. What does it cost them? Time, money, stress, missed opportunities. Costs show whether a paid solution is plausible.

    One more thing to watch for: who pays. The person with the problem is not always the person who pays. A parent pays for the child's tutor. An employee has the problem, and the employer pays. Notice who decides and who holds the money.

    What interviews cannot tell you. They cannot prove people will buy your specific solution. They can tell you whether the problem is worth solving. Later tests, covered in the next tutorials, check whether strangers will act on an offer.

    Write your learning goals.

    INTERVIEW GOALS
    My starting belief: I believe that ___ have the problem of ___ and would pay to solve it.
    What I want to learn: 
    1. Is the problem real and frequent? 
    2. What do they do today, and what does it cost? 
    3. Who decides and pays? 
    My biggest assumption that could be wrong: ___


    Step 2 — Find ten people without spamming

    You need about ten people who match your target group. Ten is a good number: enough to see patterns, few enough to be manageable.

    Who to talk to.

  • People who have the problem now, not people who "might". Someone who did their expenses this week is a better source than someone who vaguely could.
  • Strangers and weak ties, as well as friends. Friends are kind, which makes them less honest. Use them for practice, and for introductions.
  • A mix of situations. A few who cope well and a few who struggle will show you the range.
  • Where to find them. Think about places where your group already gathers.

  • People you know, asking for introductions: "Do you know someone who runs a small shop and does their own accounts?"
  • Communities and groups related to the problem, where you are a member and where rules allow it. Read the group's rules first. Many ban promotion.
  • Local places and events: meetups, markets, workshops, co-working spaces.
  • Your own network at work or school, careful to keep your employer's information and contacts out of it.
  • Customers of related businesses, reached through legitimate public channels.
  • How to ask. Keep the message short, honest and easy to decline. A template:

    Hi [name], I'm researching how [group] handle [problem]. I'm not selling anything. Could I ask you a few questions over a 15-minute call or chat? I'll share what I learn if you like. No problem at all if you're busy.

    What to avoid.

  • Mass messages, scraping or bought lists. They are impolite, often against platform rules and in some places against the law. They also produce the lowest quality conversations. Ten personal, honest requests beat a thousand automated ones.
  • Pretending to be someone else, or hiding that you are testing an idea.
  • Pressure. Accept a "no" kindly.
  • Using your employer's contacts or information without permission.
  • How many to ask. Expect many people not to reply. To get ten conversations you may need to ask thirty. That is normal, and it is one of the first tests of your effort and your idea: if the group is hard to reach, that is information about how hard selling to them will be.

    Offer something small if appropriate, such as sharing the results, a short summary of what you learn or a simple thank-you. Be careful with payment, since it can attract people who are only interested in the reward. Keep it honest.

    Where AI helps. Ask it to help you write several versions of the request in different tones, to brainstorm places your group gathers, or to suggest the questions a person might ask before agreeing. Do not ask it to find individuals' contact details or to send messages for you.

    Step 3 — The rules of a good question

    Most of the value is in the questions. Here are the rules, with examples.

    Rule 1: Ask about the past, not the future. People are poor at predicting their own behaviour and tend to be optimistic and polite. They are much better at recalling what they did.

    Weak (future, opinion)Strong (past, fact)
    "Would you use an app that does X?""Tell me about the last time you did X. What happened?"
    "Would you pay for this?""What have you paid for, or tried, to deal with this?"
    "Do you think this is a good idea?""What made you decide to handle it that way?"

    Rule 2: Ask about specific events, not general habits. "How do you usually do your accounts?" produces a tidy, idealised answer. "Walk me through the last time you did them" produces the real story, with the untidy parts.

    Rule 3: Do not mention your idea early. If you describe your solution, people will react to it, and be nice. Start with their life and their problem. If you show the idea at all, do it near the end and treat it as a question: "How would that compare to what you do now?"

    Rule 4: Ask open questions. Avoid questions answered with yes or no. "How…", "What…", "Tell me about…" and "Why…" open the door.

    Rule 5: Do not lead. "Don't you find it annoying when…?" suggests the answer. Ask neutrally: "How do you feel about that part?"

    Rule 6: Follow the energy and the cost. When someone shows frustration, a workaround or a number, ask more. "You said it takes a whole Sunday. What happens if you skip it?" The follow-up is where the insight is.

    Rule 7: Ask about money and commitment, gently. "What do you currently spend on this, in money or time?" "Have you ever paid someone to help?" "What would you want to be true before you paid for something like this?" These reveal whether payment is realistic without asking for a promise.

    Rule 8: Listen more than you talk. Aim to speak about a quarter of the time. Silence is useful: people often add the real answer after a pause.

    Rule 9: Do not fish for compliments. If you leave a conversation feeling pleased, check whether you learned any facts. Compliments feel good and teach nothing.

    Rule 10: Ask who else to talk to. The end of an interview is a good moment for "Who else deals with this, who I should speak to?" Referrals give you the next conversations.

    Use AI as a question reviewer. Paste your draft questions and ask it to mark any that are leading, future-focused or yes/no, and to rewrite them as past-focused, open questions. This is a good use: it checks your wording, and does not pretend to be a customer.

    Step 4 — A script you can use

    Here is a flexible script for a 15 to 20 minute conversation. Adapt the wording to your own voice.

    INTERVIEW SCRIPT (15-20 minutes)
    
    1. OPENING (1 minute)
    "Thanks for your time. I'm learning how people handle [topic]. I'm not selling anything today. There are no right or wrong answers. Is it OK if I take notes? I won't share anything with your name on it."
    
    2. CONTEXT (2 minutes)
    "Tell me a little about what you do and how [topic] fits into your week."
    
    3. THE LAST TIME (5-7 minutes)
    "Think about the last time you had to [do the task / deal with the problem]. Can you walk me through it?"
    Follow-ups: "What happened next?" "How long did that take?" "What was the most annoying part?" "What did you try?" "What did you use?"
    
    4. WHAT THEY DO TODAY (3 minutes)
    "What do you do about this now?" "What have you tried before?" "What did you like or dislike about it?"
    
    5. COST AND IMPORTANCE (3 minutes)
    "What does this cost you, in time, money or stress?" "What happens if you don't deal with it?" "Have you ever paid for help or a tool? What made you decide to?"
    
    6. LIGHT TOUCH ON THE IDEA (optional, 2 minutes)
    "I'm thinking about [one-sentence description]. How would that compare with what you do now?" "What would have to be true for you to use something like that?"
    
    7. CLOSE (1-2 minutes)
    "Is there anything I should have asked?" "Who else deals with this, who I could speak to?" "May I follow up if I have another question?" Thank them.

    Adjusting for your lane.

  • Service: focus on the last time they hired help or did it themselves, what they paid and what went wrong.
  • Product: focus on what they already own or use, what they searched for and why they were not satisfied.
  • Content: focus on where they get information on the topic, what they trust and what they wish existed.
  • Software: focus on the exact steps they take, which tools they combine and where the process breaks.
  • Keep it short. Respect the time you asked for. If the person wants to carry on, let them.

    Prepare, but do not read it out. Know the sections and a few good follow-ups. Use the script as a map, not a recording.

    Practise once. Run the script with a friend, and ask them to tell you where you led them or talked too much. Better to find your bad habits there.

    Ask AI to be your practice partner, not your subject.

    Role: an interview coach. You are not a customer and you must not invent customer answers.
    Task: I will paste my interview script. (1) Mark any question that is leading, asks about the future, or can be answered yes/no, and rewrite it as an open question about the past. (2) Suggest three follow-up questions for the "last time" section. (3) Tell me which part is likely to make people give polite answers.
    [paste the script]

    Step 5 — Running the conversation and taking notes

    Before. Confirm the time. Prepare the script, your notes sheet and a quiet place. Decide how to take notes. Writing is simplest. If you want to record, ask first and respect a "no". Do not record secretly.

    During.

  • Start warm and clear. Say what you are doing and that you are not selling.
  • Let them talk. Resist the urge to explain your idea. If you catch yourself pitching, stop and ask a question.
  • Stay on the past. If they drift into "I would…" or "people generally…", gently return: "And the last time that happened, what did you do?"
  • Dig into surprises. If they say something unexpected, ask about it. Unexpected is where the learning is.
  • Stay neutral. Do not argue, defend your idea or correct their view. You are there to understand.
  • Note exact words. Quotes are more useful than your summary. Mark them clearly.
  • Notes format. Use the same one for every interview so you can compare later.

    INTERVIEW NOTES
    Number: ___   Date: ___   Person (first name or code): ___   Role/situation: ___
    Last time they dealt with the problem (what happened, when): ___
    What they do today / tools / workarounds: ___
    Cost (time, money, stress) in their words: ___
    Have they paid or tried to fix it? What: ___
    Exact quotes worth keeping: ___
    Surprises: ___
    Signals (see Step 6): problem real? ___ money spent? ___ urgency? ___
    Follow-up agreed? ___

    After. Within an hour, tidy your notes while memory is fresh. Mark what was a fact (something they did) and what was an opinion (something they think or predicted). Thank the person, and send a short thank-you.

    Using AI after the call. If the person has agreed and you have removed names and identifying details, you can paste your notes and ask AI to tidy them into the format above. Ask it not to add anything you did not write. Check the output against your own memory. Do not paste anything that identifies a person, or any private business information, into a tool whose terms you have not checked.

    Dealing with nerves. It is normal to feel awkward. The first one is the hardest. People usually enjoy being asked about their experience, and many will tell you more than you expected. Do two practice rounds before you start counting.

    Dealing with a person who wants to buy. If someone asks, "When can I get it?", note it, thank them and be honest about the stage: "I'm still learning. May I contact you when I have something to try?" Do not take money for something that does not exist.

    Step 6 — Spotting polite yeses and other weak signals

    Many conversations end with warm words and no facts. Learn to tell the difference.

    Weak signals (do not count these as evidence).

  • Compliments: "That's a great idea!" "I love it."
  • Hypothetical enthusiasm: "I would definitely use that." "I'd pay for that."
  • Feature requests with no story: "It should also do X." Ask when they last needed X.
  • General statements: "People always struggle with that."
  • Politeness: nodding, "interesting", "let me know when it's ready".
  • Friends' support: they like you, so they say yes.
  • Vague promises: "Sure, keep me posted." No commitment of any kind.
  • Strong signals (count these).

  • A recent, specific story of the problem, with details.
  • A workaround they built: a spreadsheet, a checklist, a person they pay, a duct-tape process. People rarely build one for a minor annoyance.
  • Money or time already spent on the problem. Real numbers are best.
  • Searching for solutions on their own: "I looked for a tool last month."
  • Emotion tied to a fact: irritation, worry, relief when describing the issue.
  • Commitment: agreeing to a next step that costs them something, such as a time to try a prototype, an introduction to a colleague, a pre-order or a deposit.
  • The commitment ladder. From weakest to strongest: a compliment, a promise to "keep in touch", an email address, a scheduled follow-up, an introduction, a deposit or pre-order, a payment. The higher up, the stronger the evidence. In early conversations, aim to get at least a scheduled follow-up or an introduction from a real prospect.

    Reading mixed results. If five people describe a strong, recent problem and five describe nothing, you may have a real problem for a subgroup. Look at what the five have in common: their role, size, location or habits. That is your target.

    Be honest about your bias. You want your idea to be good. This makes you hear what you hope for. Write down the facts first, interpret later, and ask someone else to read the notes with you.

    A note on AI-generated "customers". Some tools will play a customer persona. These can help you practise the conversation, as long as you remember it is a rehearsal with a language model, not evidence. Do not count anything an AI said as a data point in your count of ten.

    Step 7 — From ten conversations to an evidence table and a decision

    After about ten conversations, put your notes side by side.

    Make an evidence table. One row per person.

    #SituationLast time (specific?)Workaround or toolMoney or time spentUrgencyCommitment
    1Yes / NoLow / Med / HighNone / Follow-up / Intro / Pre-order

    Fill every cell from your notes. Leave a cell empty if you did not learn it. The gaps tell you what to ask in the next round.

    Count real things. How many of the ten:

  • described a specific, recent occurrence of the problem?
  • built a workaround or used a tool?
  • spent money or significant time?
  • rated it a high priority in their own words?
  • agreed to a concrete next step?
  • Look for patterns.

  • What did the ones with a strong problem have in common?
  • What words did they use to describe it? Those are your future marketing words.
  • Which cost came up most: time, money, stress or risk?
  • Who paid, and who decided?
  • What did people already do, and why did they not like it?
  • Write a one-page summary.

    INTERVIEW SUMMARY (after ___ conversations)
    Starting belief: ___
    Number with a real, recent problem: ___ of ___
    Number who built a workaround or paid: ___ of ___
    Most common pain, in their words: ___
    Who pays and decides: ___
    Current alternatives and why they fall short: ___
    Strongest commitment I got: ___
    What surprised me: ___
    What I still do not know: ___
    Decision: continue as is / change the idea or audience / stop. Because: ___

    Decide, using the rules you wrote earlier. If at least a few people with a clear, recent problem showed real commitment, continue to a smoke test, the subject of the next tutorial. If people liked the idea but showed no real problem, change the audience or the problem, or stop. If you could not find enough people with the problem, that is an answer too. Compare your result with the kill criteria on your limits sheet.

    Ask AI to challenge your reading.

    Role: a sceptical reviewer. You do not have access to my customers and you must not invent evidence.
    Task: I will paste my evidence table and summary. (1) Tell me which of my conclusions go beyond what the facts show. (2) Point out where I may have counted a compliment as evidence. (3) Suggest the one question I should ask in the next five conversations to settle my biggest uncertainty. Be direct.
    [paste the table and summary, with names removed]

    Worked Example: Dani's ten conversations

    Dani, the invented meal-plan founder from the previous tutorial, believes "busy parents of school-age children will pay for ready-made weekly dinner plans". Dani dislikes talking to strangers, so starts with a simple goal: ten short chats in two weeks.

    Finding people. Dani asks three friends for introductions to parents who "always complain about dinner", and joins a local parents' group whose rules allow questions, after reading them. Dani sends 22 personal messages and gets 11 replies. Eight agree to talk, and Dani completes nine conversations including one friend used for practice. The practice one is not counted.

    First draft of questions. Dani asks AI to review them. Two were leading ("Don't you find planning dinners stressful?"), and one asked about the future ("Would you buy a meal plan?"). Dani rewrites them: "Tell me about the last weeknight dinner you planned. What happened?" and "What have you tried to make dinners easier?".

    What Dani hears. In six of nine conversations, parents described a specific, recent dinner struggle, usually on a Thursday or Friday when they were tired. Four had built workarounds: a shared note on a phone, a fridge whiteboard, a rotation of five meals, a weekly shopping list on repeat. Two had paid for a meal-kit subscription and cancelled it, one because of cost and one because of waste. Three said "a plan would be great!" but could not describe the last time they planned. Dani marks those as weak.

    A surprise. Planning was not the pain. The pain was the shopping list and food waste. One parent said, "I don't mind choosing meals, I hate throwing out half a cabbage." Dani had not expected that.

    Commitment. Two parents agree to try a sample week and give feedback. One offers to introduce a friend. No one is asked for money yet.

    Summary. Real recent problem: six of nine. Workaround or paid: five of nine. Most common pain, in their words: "wasting food" and "Thursday panic". Strongest commitment: two agreed to test a sample week. Decision: continue, but change the offer from "meal plans" to "plans built around one shopping list with less waste". Dani writes a hypothesis for a smoke test, and uses the results to set a number before the test.

    Dani learned something that would have been missed by building the original product, and spent no money.

    Common Mistakes to Avoid

    Three habits waste most interviews.

    Mistake #1: Pitching instead of listening. When you describe your idea, people respond to the idea, politely. Fix: ask about the last time, stay quiet, and bring up your idea last or not at all.

    Mistake #2: Counting compliments as evidence. "I'd love that!" costs people nothing to say. Fix: count recent stories, workarounds, spent money and real commitments. Treat everything else as a weak signal.

    Mistake #3: Letting AI play the customer. A model's answer is an average of text, not your market. Fix: use AI to prepare questions, review wording and tidy notes. Get the evidence from real people.

    Going Further

    Do a second round. After a first round, change your questions to settle the biggest unknown, and talk to five more people, ideally from a different corner of your group.

    Offer a trade. If people are short of time, offer a 10-minute chat in exchange for a summary of what you learn.

    Keep a swipe file of words. The exact phrases people use are your future headlines and offers.

    What to learn next. The next tutorial covers smoke tests: a simple page, a waitlist or a pre-order, which show whether strangers will act. After that comes what counts as validation, with numbers to set before the test. Then the quest has you write a validation plan.

    Key Takeaways

    In this tutorial you learned:

  • Ten honest conversations are the cheapest early test, and they can save you months of building the wrong thing.
  • Ask about the past, not the future. "Tell me about the last time…" gives facts. "Would you use…?" gives politeness.
  • Find people respectfully. Use introductions, communities where it is allowed and personal messages. Avoid mass messages, scraping and bought lists.
  • Follow good-question rules: specific events, open questions, no leading, no early pitch, listen more than you talk, ask about money and ask who else to speak to.
  • Use AI as a coach and note-taker, to review wording and tidy notes. Never use it to invent customers, and protect people's privacy.
  • Count strong signals: a recent specific story, a workaround, money or time spent, searching for solutions and a real commitment. Compliments and hypotheticals are weak.
  • Turn ten conversations into an evidence table and decide: continue, change or stop, using the rules you wrote earlier.
  • Before you move on, a challenge. Write your starting belief and your interview goals, then draft the script. Ask AI to mark any leading or future-focused question, fix them, and book your first three conversations this week.

    Learn AI, after work

    Track your progress, earn XP, and unlock more free tutorials in the AfterWork Academy app.

    Open this tutorial in the app

    More AI tutorials