Gains:
- Ability to prepare customer interview questions without directing and synthesize interview notes according to theme with artificial intelligence
- Ability to detect the difference between what is said and what is done (confirmation bias) and use artificial intelligence to separate the real signal from the polite sentence
- Ability to verify that the AI summary is representative of the raw data, which may be misleading if the sample is small or biased
No matter how good an idea looks on paper, it's just a guess until a real person says "yeah, that's my thing and I'll pay for it." Market validation is the process of testing this prediction against real human behavior. The most powerful tool for this is the customer interview - a structured conversation with the potential customer to understand his problem, without selling the product. In this unit, we will use AI (artificial intelligence) for two things: crafting good interview questions and synthesizing dozens of interview notes to extract meaningful patterns. Let's set the limit from the beginning: AI summarizes and organizes your notes; But you make the call, and human behavior determines the real signal, not the AI's sentence.
The golden rule of interviewing: don't sell, learn
The most common mistake is to turn the conversation into a sales pitch. "If I made an application like this, would you use it?" It is useless to ask; because people say "yeah, great" to be nice. This is called confirmation bias and the politeness trap. The correct approach is to ask about actual behavior in the past, not a dream question about the future. "When was the last time you had this problem, what did you do, what did it cost you?" Past behavior doesn't lie; It tells the future intention.
The essence of the interview method is the principle known as "The Mom Test": your questions must be so good that even your loving mother cannot lie to you. Three rules for this: (1) talk about the other person's life, not your opinion; (2) ask concrete past rather than general future; (3) don't take compliments and "I'll use" words for granted — only actual actions (did they spend money, did they invest time, did they look for a solution) are signals.
Note: "I would use this" and "great idea" are not data. Real signals are: the person has already spent money/effort to solve the problem; directed you to someone else; have agreed to prepay or sign up for the waiting list. It is considered an action, not a word.
Step by step: Conversation and synthesis with AI
- Prepare your question set. Generate non-leading questions from the AI that don't sell and focus on past behavior.
- Make the calls. 8-15 people are usually enough to see patterns. The other party should do most of the talking (you 20%, him 80%).
- Save raw note. Write without distorting one's own words; leave the comment for later.
- Anonymize. Remove name, contact information, recognizable details (privacy).
- Synthesize with AI. Divide notes by theme: recurring pains, available solutions, words used, objections.
- Separate signal from noise. Ask the AI "which ones are evidence of actual behavior, which are just polite phrases?" parse it as follows.
- Decide. Does the pattern confirm pain, or is a pivot needed? The founder makes this decision.
The difference between what is said and what is done
People talk about what they want to do, not what they have done. Separating these two is critical in AI synthesis. Someone who says "I want to eat healthy" may be eating fast food every evening. In the interview synthesis, divide each statement into two: statement of intent (“I want, I do, I use”) and evidence of behavior (“I did this 3 times last month, I gave 200 liras to that”). Only the latter carries weight for the decision. Make the AI make this distinction explicitly; Otherwise, it will mislead you by summarizing the intent sentences as real demands.
three mini cases
Case 1 — The mask of politeness is falling. One founder interviewed 12 people; 10 said "great idea, I will definitely use it". He was happy. But if you give the notes to the AI and ask "how many people are actually spending money/effort on this problem right now?" When we parsed it, the number dropped to 2. Ten people had acted kindly, only two people were in real pain. The founder escaped the trap of early joy and deepened the profile of those two individuals.
Case 2 — Words changed the product. One team conducted 14 interviews with freelance designers. When they synthesized it with AI, they found that everyone was using the same word: “collection nightmare.” Their product idea was “project management” but the real pain was payment tracking. AI's word frequency analysis provided positioning that spoke to the customer's own language and significantly increased landing page conversion.
Case 3 — Where the small sample misleads. A founder talked to only 4 people and had them synthesized by AI; The AI “found” a strong pattern. But three of the four people were from the same friend group (sampling bias). When the founder realized this, he added 10 more people through different channels; the pattern has changed greatly. Lesson: No matter how well the AI summarizes, the output will be misleading if the input is small or biased.
Four copyable templates
1) Non-guided interview questions:
Your role: customer discovery coach. Segment: [target customer].Write 8 open-ended interview questions that don't sell to me, don't give away my opinion, and focus on past behavior. The questions should ask about the concrete past, such as "when was the last time, what did you do, what did it cost you", not future intent such as "will you use/want to use it?" Using leading questions.
2) Interview synthesis (theme extraction):
Below are [n] anonymous notes of the conversation. Task: (1) group recurring pains by theme and count how many conversations occur, (2) list common words/phrases clients use, (3) extract current solutions (what they are doing now). Do not add anything that is not in the note; If you generalize, please specify.
3) Signal/noise separation:
Use the same notes. Divide each key phrase into two:(a) evidence of behavior (money/time/effort actually spent),(b) intent/courtesy phrase (“I use”, “great”). List (a) group separately; I will make the decision based on these. Count how many people show evidence of real behavior.
4) Sample health check:
Profile of interviewers as follows: [short, anonymous summary]. Tell me about possible biases in this sample (are they all from the same channel, same age/sector). Suggest how many more meetings I should have with different profiles before I trust the pattern.
Weak prompt / Strong prompt
Weak prompt:
Summarize these interview notes.
This request produces a superficial summary that confuses intent and behavior, making polite phrases seem like genuine requests.
Powerful prompt:
Synthesize the following 12 anonymous interview notes. (1) Sort pain into themes and count frequency, (2) list behavioral evidence and intent statement separately, (3) omit client's own words, (4) warn of risk of sampling bias. Don't add anything that isn't in the note.
Common mistakes
- Turning the conversation into a sale. Explaining your idea and asking for approval; It hides real pain.
- Taking future intentions as data. “I would use” is a guess; Ask about past behavior.
- Making decisions with very few people. 3-4 interviews are insufficient for the pattern and open to bias.
- Talking from the same circle. The friend/same channel sample validates you but is not representative of the market.
- Mistaking the AI summary as raw data. The summary is as good as the notes; Bad/less input gives bad results.
Tip: Ask one question after each conversation: “What has this person done so far to solve the problem?” If the answer is "nothing," you're probably in vitamin territory, not painkillers, no matter how enthusiastic he or she sounds.
In summary
Market validation is testing the idea with real human behavior, and the most powerful tool is the customer interview. The rule of the interview is to learn, not sell: ask for concrete past behavior, not future intentions; Compliments and "I'll use" words are not considered data. AI is very powerful at preparing unguided questions and extracting word patterns by separating dozens of notes into themes. But AI synthesis is input-dependent: if the sample is small or biased, if intent statements are not separated from behavior, the summary will mislead. It is up to the founder to separate the signal from the noise — to filter the evidence of actual behavior through politeness — and make the final decision.
Application task
Plan to meet with at least 5 people from your target segment (through different channels). Prepare your question set with the "Non-guided interview questions" template. After conducting the interviews, anonymize the notes and feed them to the AI with "Interview synthesis" and "Signal/noise separation" templates. Report the result as follows: how many people showed evidence of real behavior, what words were repeated, what bias was present in your sample, and what the decision should be to continue or pivot based on that.
checklist
- [ ] Are my questions not selling, focusing on past behavior?
- [ ] Did I meet with enough and diverse people (through different channels)?
- [ ] Have I anonymized the notes?
- [ ] Have I separated the evidence of conduct from the intent/politeness clause?
- [ ] Did I base the decision on the actual behavioral signal and not the AI summary?