How do I create my own AI persona?
Two main paths: upload an interview transcript, or run a Scout task. Most people who need custom personas already have some interview data, so upload is the default. If you don't have transcripts, Scout builds personas from observed social media behavior.
Custom personas are private to you and reusable across studies.
Two paths at a glance
| Upload transcripts | Scout | |
|---|---|---|
| Source | Your existing interview data | Public social media |
| Best for | You have real customer/user data | You have no data, niche audience |
| Time | 5–10 minutes per transcript | 1–2 days |
| Quality | Depends on transcript depth | Built from observed behavior |
Path 1 — Upload an interview transcript
What works:
- A 1-hour recorded interview, transcribed to text.
- A customer research transcript with depth (5,000+ words ideal).
- Multiple short interviews combined into one persona.
The flow:
- Go to Persona management → Create Persona → Upload Interview Transcript.
- Upload a PDF (or TXT, Word — supported formats vary).
- AI extracts basic info, values, lifestyle, consumption habits, pain points.
- Preview, edit if needed.
- Save to your private library.
Processing time:
- 2,000 words → about 1 minute.
- 5,000 words → about 2–3 minutes.
Tips for better output:
- Keep the original Q&A format if possible. "Q: How often do you buy coffee? A: Daily, but only on weekdays." preserves more signal than summaries.
- A mix of demographics, behaviors, and attitudes produces a richer persona than demographics alone.
- If the transcript is short (under 1,000 words), the persona will be thinner. Either upload more data or accept it.
Path 2 — Use Scout
When you don't have transcripts but want to study a specific audience.
The flow:
- Go to Scout → Create New Scout Task.
- Describe the audience: "Camping enthusiasts on Xiaohongshu" or "Young environmentalists on Weibo."
- Scout runs 1–2 days, observes public posts, infers patterns, generates personas.
- Personas are saved to your private library automatically.
Best for niche audiences the public library doesn't cover, or when you want personas that reflect current behavior — not behavior from interviews done six months ago.
What you can edit after creation
Anything. Go to the persona detail page and edit:
- Basic info (age, city, job, income)
- Behavioral traits
- Pain points
- Tags and notes
You can also upload additional interviews to enrich an existing persona. AI merges the new data with the existing persona.
Privacy and reuse
Custom personas are private by default — only you (and your team, on team plans) can find them. They're not searchable by other users and not used for shared AI training.
Once in your library, you can use them in any Interview or Discussion, add them to a Panel for repeated studies, or share with team members on a team plan.
Common mistakes
Uploading transcripts that are too short. Under 1,000 words gives AI limited material. The persona will be thin.
Uploading summaries instead of transcripts. "Sarah is a 32-year-old product manager who values quality" gives AI nothing to work with. The actual conversation transcript is the source.
Building one persona when you need five. A single custom persona is rarely useful alone. Build a small set (3–10) so you have variety for studies.
Treating the extracted persona as ground truth. It's a model of a real person, useful for testing, not for representing that specific person in any context.