BLOG 1: From 10 interviews to a clearer user picture

You have a few conversations with users. Then a few more. There are transcripts, interview notes, feedback forms, sales notes, and customer support comments. There is more than enough data. And yet, when someone asks: “What did we actually learn?” the answer is often not that simple.

Every PM/PO knows that feeling. The problem is not collecting data. The problem is synthesizing it.

AI can help exactly here. But not by having the AI decide who your user is or what you should build, but by going through a large amount of information much faster and helping you find patterns.

AI as a research assistant, not as a research oracle

Based on transcripts, AI can identify:

• recurring problems;
• behavioral patterns;
• user motivation;
• existing workarounds;
• differences between user groups;
• potential signals of willingness to pay;
• questions to which you don’t have an answer yet.

This can save a PM/PO hours of manual synthesis. But there is an important trap. AI is very good at filling in the gaps. If the data does not provide an answer, AI may be tempted to construct it. That is why a good prompt must explicitly say:

Do not invent what does not exist in the data.

The original approach to this problem clearly distinguishes between what the user directly said, what can be inferred from multiple responses, and what is simply not known. This is one of the most important things to transfer into any AI-assisted research workflow.

Proposed prompt to get started

Role: You are an experienced Product Researcher with experience in [industry/niche].

Data source: Interview transcripts are the primary data source. Do not invent or supplement information that is not supported by the materials.

Clearly distinguish between:

• what the user directly said;
• conclusions that can be drawn from multiple responses;
• information that is not known..

Product context:
Product: [describe]
Target group: [describe]
Problem we are solving: [describe]
Product stage: [idea/MVP/existing product/growth]
Research goal: [ICP / MVP / positioning / understanding the problem / other]

Task: Based on the provided user interviews, identify patterns in user behavior, needs, problems, and motivation. Analyze all interviews and propose a primary user persona.
The persona should be based on:

• problem intensity;
• problem frequency;
• consequences of the problem;
• motivation to solve the problem;
• existing workarounds;
• relevance of the problem to our product.

Do not assume willingness to pay if there is no evidence in the interviews.

At the end, state:

• 3 biggest pain points;

• existing workarounds;

• up to 2 key quotes;

• why this persona is a candidate for the primary persona;

• 3 things we don’t know yet;

• one piece of information that, if obtained, would change the conclusion the most.

Don’t just ask: “Who is the persona?” A better question is: “What evidence supports the conclusion that this is our primary persona?”
Even better: “What would have to be wrong with our assumptions for this persona to not be the best initial segment after all?” This changes the way you use AI. Instead of looking for a ready-made answer, you use it to challenge your thinking.

And what about willingness to pay?

This is particularly important. If a user says: “This causes a big problem for me.” That does not automatically mean: “I am ready to pay for a solution.”
Therefore, AI should not convert problem intensity into willingness to pay. Instead, look for concrete signals:

• whether the user is already paying for an alternative solution
• whether they invest significant time or resources;
• whether they have already tried to solve the problem;
• whether they have a budget;
• whether they have explicitly expressed a willingness to pay.

If there is not enough evidence – the answer should be: “I don’t know.” And that is a good research result.

Try it today

Don’t start with 50 interviews. Take the last three transcripts you have. Put them into the AI and ask for an analysis. Then compare the result with your own conclusions.
But don’t just ask: “Is the AI right?” Ask: “What did the AI notice that I might have missed?”
That is a much more interesting experiment.

In addition to the basic prompt to start experimenting, we also provide a more advanced prompt that you can try as well.

Proposed advanced prompt

Role

You are an experienced Product Researcher with experience in [enter industry/niche, e.g., B2B SaaS / FinTech / E-commerce…]. Your task is to identify patterns in user behavior, needs, problems, and motivation based on the provided user interviews, and based on them, propose one primary user persona.

Data source

Attached are transcripts of conversations with users [or detailed interview notes]. Transcripts are the primary and most important data source. Do not invent or supplement information that is not supported by the provided materials. Clearly distinguish between:

• what the user directly said;
• conclusions/inferences that can be drawn from multiple responses;
• information that is not known.

If there is not enough evidence for a certain conclusion, clearly state that it is not possible to reliably conclude it.

Product context

We are developing a product that helps [target group] solve the problem with [key problem].
Current product stage: [idea / discovery / MVP / existing product / growth].
[Optional]
The primary user of the product is [enter].
The buyer of the product is [enter].
The person making the purchasing decision is [enter].

Research goal

We are doing this analysis to [e.g., define ICP / identify the best initial segment / prioritize MVP / understand the problem / define positioning / verify product-market fit].

Task

Analyze all provided interviews and identify recurring patterns. Based on evidence from the interviews, propose one primary user persona. The primary persona should be a user group that, based on available evidence, shows the best combination of:

• problem intensity;
• problem frequency;
• consequences of the problem;
• motivation to solve the problem;
• existing attempts to solve the problem;
• potential willingness to pay for the solution;
• relevance of the problem to our product.

Do not assume willingness to pay just because the user considers the problem important. If the interviews do not contain enough evidence of willingness to pay, clearly state that this conclusion cannot be reliably made. Also, do not define the persona solely by demographics, title, or position. Focus primarily on the problem, behavior, context, and motivation.

Output format

Primary persona

Persona name: Put a short, descriptive name based on the persona’s problem or behavior, not a generic name like “Sales Manager”.

Profile & role: Who is this person, in what context do they work, and what are their main responsibilities?

Context in which the problem occurs: When, how, and under what circumstances does the problem occur?

Main goal / Jobs-to-be-Done

What is this person trying to achieve? Formulate the most important JTBD in the format: When [situation], I want to [motivation/action], so that [expected outcome]. If there are multiple JTBDs, isolate only the one most relevant to the problem our product solves.

Biggest problems / Pain points

List up to 3 biggest pain points. For each, state:

• Problem
• Evidence from the interview
• Frequency: how many interviews support this conclusion, if possible to determine
• Consequence: what this problem causes for or prevents the user from doing

Do not mix what the user said with your assumptions!

Current behavior and workarounds

How do users solve this problem today? List:

• existing tools;
• manual processes;
• workarounds;
• alternatives;
• what currently works and what does not.

This is important because a user who is already investing time, money, or resources into solving a problem can be a stronger signal than a user who only verbally expresses frustration.

Signals of willingness to pay

Based on the interviews, extract concrete evidence indicating:

• existing spending of money;
• use of paid alternatives;
• significant expenditure of time/resources;
• previous attempts to solve the problem;
• expressed intention to pay;
• existence of budget or purchasing authority.

If there is not enough evidence, write: “Willingness to pay cannot be reliably concluded from the available interviews.”

Key quotes

Extract up to 2 literal quotes from the interviews that best illustrate the core of the problem. Do not paraphrase or reconstruct quotes. If an interview or user label is available, state it alongside the quote.

Why this persona?

In a few sentences, explain why this group is the best candidate for the primary persona compared to other users from the interviews. State concrete patterns/evidence from the research.

What we don’t know yet?

List up to 3 most important pieces of information that are missing to make the persona more reliable. For each, write:
What is missing → why it is important → how we could verify it in the next interviews.
At the end, specifically isolate one piece of information that, if obtained, would change or confirm the conclusion about the primary persona the most.

Rule for the entire analysis

Accuracy and fidelity to the research are the priority, not the completeness of the persona. If the data does not support a certain conclusion, do not invent it just to make the persona look complete. It is better to write “I don’t know” than to present an assumption as a fact.
After generating this proposal, write down which piece of information you miss the most in my transcripts to make this profile even more accurate.

Recommended tools to try: ChatGPT · Claude · Perplexity AI