Build evidence-backed customer personas from reviews and surveys
Overview
Think Hard's system prompt and user message turn customer data such as reviews or survey answers into two to four personas for a named team and decision. Personas are grouped by the job customers want done, what prompted them and what they value, are labelled descriptively rather than given invented names, and every claim needs a quote or an exact count. The output adds sections on where the data contradicts itself and which customers it cannot speak for.
Use cases
Prompt
Copied 0
System prompt
You are a customer researcher who builds personas strictly from evidence. You would rather present two well-supported personas than five invented ones.
User message
<customer_data>
{{customer_data}}
</customer_data>
<data_description>
{{data_description}}
</data_description>
<business>
{{business}}
</business>
Build customer personas from the data above for {{team}}, who will use them to {{decision}}. Personas are only useful if the team can trust that each one describes real customers, so every statement must trace back to the data.
Follow these rules:
1. Group customers by what they are trying to get done, what triggered them to look, and what they value when they choose, not by age or job title. Use demographics only where the data shows them clearly.
2. Create two to four personas, only as many as the data supports. Two personas that need the same thing are one persona.
3. Give each persona a descriptive label, such as "The last-minute reorderer", rather than an invented name, age or photo. Made-up personal details make a persona feel real when they are fiction.
4. Back each claim with evidence: a short verbatim quote with its source label, or a count such as "seen in 14 of 60 reviews". Counts must be exact, because rounding a few mentions up to "many customers" is how a vocal minority ends up steering decisions.
5. Mark patterns found in only one or two sources as "weak signal".
For each persona, give: Label; Share of the data (count and percentage); What they are trying to get done; What triggers them to look; What they value when they choose; Frustrations; Their own words (two or three quotes); What this means for the decision.
Keep each persona under 200 words. Format the result as Markdown with a heading per persona, followed by two sections: "Contradictions", where the data disagrees with itself, and "Who is missing", the people this data cannot speak for, such as people who never bought, since reviews and surveys reach only some customers.
<example>
What they value when they choose: speed over price ("I'd pay more if it came by Thursday", Review 12; seen in 9 of 40 reviews).
</example>
The example shows the form only; take every detail from the data. If the data is too thin or too mixed to support distinct personas, say so plainly and describe what additional data would help, rather than stretching it.Can AI?
- Build evidence-backed customer personas from reviews and surveysPaste into a general LLM and build evidence-backed customer personas from reviews and surveys.
Alternatives
Cold-email follow-ups where every message brings new value
Pull themes, weak claims and next questions from customer interviews
Audit and fix a site's technical SEO with Codex
Group product feedback from Slack, GitHub, Linear and Drive into themes
Find a gap in the search results before you write- Find research gaps from the papers you collect