Create personas with AI: step by step, backed by evidence instead of invented

Create personas with AI in four steps

• 1. Collect material: reviews on comparison sites, posts in forums and communities, your own customer conversations, support requests, search queries from your campaigns. AI can search and summarise; check every quote you keep at its source.
• 2. Find patterns: which problems, triggers and objections come up more than once, from different people? Let the AI group them and check each group against the quotes behind it.
• 3. Write personas: for each persona, who they are, what they want to achieve and what makes them hesitate, backed by quotes. Mark what you infer as an assumption.
• 4. Check: do the personas really differ, or are they variants of the same person? Are there any who would push back on your offer? Is a group missing that buys from you?

Typical mistakes with AI personas

• Invented quotes: a language model writes statements that sound real. Without a source, a quote is not evidence.
• All the same: ten personas with the same situation and the same doubt show an average, not an audience.
• Only agreement: personas that approve every piece of copy are useless for checking. Language models tend to agree with their users; a study by Anthropic calls this “sycophancy”.
• Profile instead of decision: check whether the persona says what matters for their decision, or mainly age, hobbies and favourite brands.

Profile or persona that reads along

Classic personas are profiles a team reads. Synthetic users react to copy and pages themselves: you show them two versions and get objections and a choice. That is fast and works before launch, but does not replace measurement in the market. UX studio concludes that ChatGPT does not replace usability tests with real people; AIMultiple sees the main use in hypotheses and early tests. More under synthetic users.

Create personas with 99takes in Claude

• 99takes researches your competitors and public customer voices in Claude and builds your audience from them: 10 to 14 personas with a wide spread, every piece of evidence with its source. Where evidence is missing, it says so, and an audience with no evidence at all is marked as a hypothesis.
• Set up once for 15 credits. After that the personas read your copy and pages in every check. More under audience research.
• The verdict follows the personas the offer is made for; the edge is reported separately (core and edge).
• To try it out there are ready-made audiences, for example for software buyers in mid-sized companies or for consumers.

Frequently asked questions