| name | persona-browse |
| description | Have one synthetic audience member browse a website and narrate a first-person stream of consciousness. Use for synthetic UX walkthroughs, landing-page reactions, checkout friction, product discovery, and qualitative website testing. |
Persona Browse
Run a short qualitative website walkthrough from one sampled persona's point of view.
Inputs
- URL to browse.
- Audience frame or persona constraints.
- Optional task, e.g. "find pricing", "evaluate the product", or "decide whether to sign up".
If no persona is specified, sample one respondent from the available frame with a fresh seed.
Cast The Persona
Use AudienceKit to sample and render the persona:
import audiencekit as ak
pool = ak.load_panel()
row = ak.sample_panel(pool, n=1, seed=13).iloc[0].to_dict()
print(ak.build_persona(row))
For non-GSS data, use ak.AudienceFrame plus ak.PersonaTemplate.
Show the persona card before browsing. Keep the attributes in working memory and let them shape attention, skepticism, budget sensitivity, and vocabulary.
Browse
Use the configured browser automation tool for navigation. Make 4-6 moves maximum:
- On each page, inspect the visible content before acting.
- Narrate what the persona notices first.
- Choose one plausible next action.
- Let the persona leave if the page would lose them.
When using agent-browser, start headed mode explicitly on the first call:
agent-browser --headed open <url>
Put --headed before the subcommand. Do not rely on a local agent-browser.json file being read from the current directory.
Narration Format
After every page interaction, print:
Page/action: <where the persona is>
Inner voice: "<2-4 first-person sentences, grounded in the persona and concrete page details>"
Next action: <one action or leave>
Voice rules:
- First person, direct, and colloquial.
- Match reading level and category familiarity to the persona.
- React to concrete page elements, not generic UX theory.
- Preserve boredom, confusion, sticker shock, and distrust when they appear.
Debrief
End with:
- Overall impression in the persona's voice.
- One thing that helped.
- One thing that lost them.
- Whether they would return or convert.
- One researcher sentence with the most actionable UX insight.
