How do I upload interview transcripts to create a persona?

The short answer: PDF format, drag-and-drop into the persona creation page, AI extracts in 1–3 minutes. Quality depends on transcript depth: 5,000+ words gives you a rich persona, 2,000 words gives you a usable one, under 1,000 words is thin.

What format does the upload accept

PDF only. The full transcript in one file.

If your data is in Word, plain text, or audio: convert to PDF first (any tool works; for audio, transcribe then save as PDF).

What makes a transcript work well

Ideal (5,000+ words):

One full 1-hour in-depth interview
Mix of demographics, behaviors, attitudes, pain points
Q&A format preserved
Includes the participant's own words, not interviewer summaries

Usable (2,000–5,000 words):

A shorter interview (30–45 minutes)
Multiple short interviews combined into one document
Customer interview with structured notes

Thin (<1,000 words):

Quick survey responses without depth
Single-question interview
Mostly interviewer notes, few direct quotes

You'll get a persona either way, but the depth scales with the source material.

The 4-step upload flow

Go to Persona management → Create Persona → Upload Interview Transcript
Drag-drop or select your PDF
AI extracts in 1–3 minutes — you'll see a preview of the persona
Edit any field that's wrong, then save to My Library

If the upload fails: check file size (3MB max per file), check it's actually a PDF, check it isn't password-protected.

What AI extracts

Seven standard dimensions:

Demographics (age, gender, city, occupation, income)
Values (what matters to them, priorities)
Lifestyle (daily patterns, work-life balance)
Consumption habits (how they buy, what they buy, decision factors)
Decision logic (how they evaluate options, what tips them)
Information channels (where they get advice, what they read)
Social relationships (how they talk to others about products, who influences them)

A 5,000-word transcript usually fills all 7. A 2,000-word transcript fills 4–5. A 1,000-word transcript fills 2–3.

After extraction, AI flags which dimensions are missing or thin. You can edit to fill gaps or accept the thinner version.

How to prepare transcripts for better output

Three things help:

Keep the original Q&A format. "Q: How often do you buy coffee? A: Daily, only weekdays." preserves more signal than "The participant is a frequent coffee buyer."

Include the participant's words, not just the interviewer's summary. AI extracts more from direct quotes than from third-person descriptions.

One persona per transcript. If you have 10 customer interviews, upload each as a separate PDF. AI builds 10 personas; you pick the ones that match your audience. Combining interviews into one PDF forces AI to merge into one averaged persona, which loses distinctiveness.

What about audio recordings?

Audio isn't accepted directly. The flow is:

Transcribe the audio (any transcription service works)
Save the transcript as PDF
Upload

Transcription takes about 25% of the audio length (a 60-minute interview = ~15 minutes of transcription time). Most transcription services handle Chinese, English, and major languages.

Privacy

Uploaded transcripts:

Are stored privately
Are not visible to other users
Are not used for AI training
Can be deleted at any time (deletion is immediate)

If you're working with sensitive customer data, this is the right place. After the persona is extracted, you can also delete the source transcript — the persona itself is what you keep.

Limits to know

3MB max per file
50MB total upload size across all personas
PDF format only
Single file per persona creation (batch is not supported)

For very long transcripts (10,000+ words), split into multiple PDFs and create multiple personas. AI will let you pick which personas to keep.

After upload: what's next

The persona is saved to My Library. From there:

Use it in any Interview or Discussion
Add it to a Panel
Edit fields if anything is wrong
Upload additional transcripts to enrich it
Share with team (team plans)

For most users, the next step is to build 3–5 personas from 3–5 transcripts, then run an Interview with them as a Panel.

Common mistakes

Uploading a summary instead of the transcript. If you wrote "Sarah is a 32-year-old PM who values quality," AI has nothing to extract from. Upload the actual conversation.

Combining multiple interviews into one PDF. You lose distinctiveness. Upload separately.

Accepting the persona as ground truth. The persona is a model, useful for testing — not a definitive portrait.

Stopping at one persona. A single custom persona gives you a narrow view. Build a small set (3–10) to cover the variety in your real audience.

Related: Persona Library, Scout, Panel, Interview