szintetikus persona

in English: synthetic persona

advancedspreading

A fictional buyer or user created with artificial intelligence, whom companies question about their ideas, advertisements, or products instead of, or before, asking real people.

In plain terms

Marketers have long worked with personas: these are fictional but representative customers, such as 'Kati, 38, a two-child accountant from Győr'. Until now, this was a character sketch written on paper and pinned to the wall by the team. A synthetic persona is the same, but it speaks: a language model plays Kati and answers your questions.

It is like when an actor receives a detailed character description, then responds to anything improvisationally, in character. You can ask them what they think of the new packaging, whether they would buy it at that price, or what they would type into a search engine if they were looking for such a product.

It is good for getting first feedback in minutes and cheaply before asking real people. It usually represents one type of person, not a copy of a real person, and is not a survey: the model says what it considers probable for such a person based on learned patterns.

It is increasingly common in search and in measuring AI visibility: synthetic personas are used to generate the questions that different types of people would ask a chatbot, and then the machine's answers are checked to see whom it recommends.

How it works

First, the persona is assembled: age, residence, occupation, income, habits, value system. This can come from actual research data, a customer database, or simply the team's assumptions.
The description is received as an instruction by the language model, with the request to respond on behalf of this person.
The researcher asks: conducts an interview, has them fill out a questionnaire, shows them ad copy, or asks them to formulate search queries.
This is usually repeated not once but with hundreds or thousands of different personas, so that ratios and patterns emerge from the responses. In more advanced systems, each persona is an agent that remembers what came before and can even converse with the other personas.
Careful teams ultimately compare results with real human responses, because only then can the reliability of synthetic responses be determined.

An everyday example

A web store plans a new shipping fee and has no time for weeks of market research. The marketer shows the three price variations to a thousand synthetic shoppers and sees by afternoon which one would annoy the most. The winning variation is then tested on a small group of real buyers.

Who is affected

marketerYou can quickly and cheaply test messages, prices, and advertising materials with it, and map what your target groups ask AI search engines. → Marketing

entrepreneur, executiveYou can get initial market feedback with a small budget, but you must know this is an estimate, not a promise from real buyers.

developerYou can test the interface or chatbot with synthetic users before it is exposed to real users. → Code

everyoneDecisions about which ads and products are aimed at you are increasingly based partly on the opinions of 'buyers' who do not exist.

A common misconception

MythThe synthetic persona says the same things real customers would say, thereby replacing market research.

In realityThe model returns the average and the probable, but often not the surprising, contradictory behavior of humans. It is suitable for pre-screening and brainstorming; real people are needed for important decisions.

Limits and risks

Responses tend to be too uniform, too polite, and too reasonable; real people are more extreme and inconsistent.
The model's training data contains little material on buyers from Hungary and generally smaller markets, so the 'Hungarian pensioner' easily becomes a stereotype or a translated American.
What the persona says is not proof: it can confidently be wrong about new products, recent events, or situations the model does not know.
If personas are created from real customer data, data protection issues arise, and it is misleading to present synthetic results as a real survey.

Similar, but not the same

buyer persona – A traditional persona is a static character profile written by humans, whereas a synthetic persona is queryable because it is played by a model.
synthetic data – Synthetic data is an artificially generated dataset, such as a table or images, while a synthetic persona is a fictional person who can be interviewed or made to speak.
digital twin – A digital twin is a virtual copy built from data of a specific, existing thing or person, while a synthetic persona typically represents a type of person. The line can blur: if the persona is built from interviews with a real person, it is more of a digital twin.

Related concepts

Broader

synthetic data, market research, generative AI

Narrower

synthetic user, synthetic respondent

On the edge

buyer persona, digital twin, large language model, AI visibility, agent-based simulation, silicon sample

A brief history

The persona as a design tool spread in the software industry with Alan Cooper's 1999 book. The version voiced by large language models appeared in the early 2020s: a 2022 study called the simulated respondents "silicon samples," and a 2024 study built agents imitating them from interviews with more than a thousand real people.

Where things stand

As of late 2026, a range of market research and marketing tools offer synthetic personas, and larger companies regularly use them for pre-testing. However, their reliability is the subject of lively professional debate, and they have not replaced research with real people.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

AB InBev launches AI-personalised Budweiser Sessions ad platform in 14 marketsSep 21, 2026 · Marketing

Also known as

synthetic persona · synthetic persona · AI-persona · synthetic user · synthetic user · synthetic respondent · synthetic respondent · silicon sample

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.