Synthetic users: what they are and how to use them to test an idea

Simulated people built on a market’s data: what they can tell you, what they can’t, how to interview them without fooling yourself, and when you still need the real customer.

Updated October 6, 2026 · MAPS editorial team

In brief

Synthetic users are people simulated by an AI model, built on a market’s data — segments, conversations, reviews, real customers — with a backstory, habits and objections. You can interview them, put them in a focus group, and use them to test an idea before spending. They are a pre-test: they help you prepare and discard weak assumptions, but they don’t replace real customers.

What they do well

React to a stimulus. A concept, a message, a page, a price: they tell you what they understand, what convinces them, what holds them back.
Surface objections. Often ones the team hadn’t thought of.
Compare alternatives. Two messages, three concepts, in front of the same profiles.
Prepare the real research. They help you write better questions for customer interviews.

What they can’t do

Predict the numbers. They can’t tell you how many people will buy, or at what price, with any certainty.
Replace a representative sample. When you need a measurement that holds up, you need real people.
Surprise you beyond the data. They know what the data they are built on contains, plus what the model knows in general.

How to use them well

1. Build them on real data

A synthetic user is only as good as the data it comes from: analyzed segments, reviews, conversations, CRM customers. Characters dreamed up from scratch produce the answers the team expects.

2. Know what they rest on

Every profile should say which data it comes from and what is missing. A profile built on little data should be read with more caution.

3. Ask neutral questions

Leading questions work on simulated people too: “don’t you think it would be convenient…?” gets a yes. Ask what they would do, not whether they like it.

4. Show the real stimulus

An image, a PDF, a page, a website: reacting to something concrete tells you far more than reacting to a description.

5. Put people up against each other

In a focus group, positions contradict each other and shift: tensions emerge that a one-on-one interview doesn’t reveal.

6. Check against the market

Whatever holds up in front of synthetic users must then be tested with real customers, against a criterion decided in advance.

How MAPS does it

In MAPS, synthetic users are built from the project’s data — a brand and its reviews, the segments, the niches found — or from your real customers in the CRM, always and only as aggregates. Each profile says which data it rests on. You interview them while showing images, PDFs, links or an entire website, and MAPS warns you if a question is leading; you put them in a focus group run by an AI moderator; and with “Test with Users” you can test almost any tool on seven or eight dimensions. In B2B there is the Buying Committee. Reactions are presented for what they are: useful, not infallible.

A focus group with synthetic users in MAPS: the moderator presents two suits and the people respond, each from their own point of view

How it works: Synthetic Users →

Frequently asked questions

Are synthetic users reliable?

As a preparation tool, yes, if they are built on real data and used with neutral questions. Not as a prediction of real behavior: for that you need tests with real customers.

How are they different from buyer personas?

A buyer persona is a profile described on paper. A synthetic user is a profile you can talk to: it answers, reacts to stimuli, changes its mind in a focus group.

Can they be built from customer data?

Yes, starting from groups of real customers — from the CRM, for example — using only aggregated data, without exposing any individual’s information.

What does the research say?

Several studies have shown that language models, suitably conditioned, can partly reproduce the responses of groups of real people; others have shown their limits and biases. That is why they should be used as a pre-test, not as a substitute.

Origin and references

Among the first studies on the subject: Lisa P. Argyle et al., “Out of One, Many: Using Language Models to Simulate Human Samples”, Political Analysis (2023).

Written by the MAPS editorial team with the help of AI, checked and reviewed by the team.

Methodology·10 min di lettura

Strategic validation: putting an idea to the test before you invest

Find the riskiest assumptions, choose the right test, decide up front what counts as evidence: a method for failing small instead of failing big.

Read →

Strategic Tools·8 min di lettura

Empathy Map: what it is and how to use it to understand the customer

What your customer says, thinks, does and feels, what they see and hear, what holds them back and what they want: how to fill it in without making things up, and how to turn it into decisions.

Read →

Strategic Tools·9 min di lettura

Value Proposition Canvas: how to find fit with the customer

The customer profile — jobs, pains, expected gains — next to what you offer: how to build it on evidence and tell whether your proposition really answers.

Read →