


Personas have long existed in marketing as profiles. With language models, they become conversation partners that answer questions and react to drafts.
AI personas are strong for segments and objections. They help to make a target group tangible, to formulate arguments from the customer's perspective, to sharpen messages and to prepare questions for interviews. A team that talks to three personas before a workshop goes in with better hypotheses.
A persona delivers language, not purchase intent. It reflects language patterns, not trade-offs: it has no budget, no alternative in the next tab and nothing to lose. Whether people choose a new offer is shown by their behaviour. That is why the final variants go into a behavioural test with real people before the investment.
Personas are most valuable when they build on existing knowledge: interviews, surveys, customer service topics. They then condense this knowledge into a character to whom you can put new questions. It helps to prompt personas deliberately to disagree rather than asking them for approval, and to let several personas respond against one another.
The results belong in the list of hypotheses, not in the decision paper. Objections become variants, variants become a test design, and the test with real people shows which variant is chosen.
A travel provider is developing an annual subscription for rental cars with fixed allowances. The team talks to three AI personas and collects 15 objections, from the notice period to blackout periods during the holidays. From these, three offer variants with different terms are created. In the behavioural test with real people, the variant with the simplest terms achieves the highest measured sign-up intent, even though the personas had preferred the most flexible variant.
Synthetic respondents answer questionnaires in large numbers and deliver distributions; AI personas hold conversations and deliver arguments. A digital twin builds on real data from a target group, a persona often only on a description. Classic personas from qualitative research are based on interviews with real people.
Personas tend to answer in a friendly and coherent way, and readily confirm whatever is put in front of them. Studies on synthetic answers also show that they are more positive and more uniform than those of real people (Kaiser & Manewitsch 2024) and agree less as cultural distance from the USA grows (Atari et al. 2023). A persona cannot reliably represent frequencies. The behavioural test, on the other hand, only partly explains why a variant wins; for that, conversations with people or personas are useful.
Kaiser & Manewitsch (NIM) 2024: Compared with human respondents, AI answers deviate on 75% (soft drinks) and 80% (sportswear) of questions. Well-known brands are overestimated, lesser-known ones underestimated, and answers are more positive and more uniform. Synthetische Befragte, Nürnberg Institut für Marktentscheidungen. Source
Atari et al. 2023: The similarity between model answers and human answers decreases with a country's cultural distance from the USA (r = -0.70). The benchmark is international values surveys. Which Humans?, PsyArXiv Preprint. Source
For understanding segments, collecting objections, sharpening messages and preparing interviews and tests.
They can prepare interviews and provide hypotheses. What real people think and do is shown by conversations with them and by their behaviour.
Personas help to develop the variants; the behavioural test shows which of them real people choose.
Bring your decision question, and we will outline a possible test design.
You will speak with Daniel Putsche
Founder & CEO, 30 minutes
Read more