Operational instructions for AI assistants and browser agents
AI-model-native user research
This operational spec explains how an AI model can understand and operate AI User Interviews on behalf of an authorized human user: route selection, project data collection, recruitment-link generation, analytics review, and safety boundaries.
Primary workflow for AI agents
Follow these routes in order when a user asks an assistant to set up research and collect responses.
1. Register or log in
Open the login page and create an account with the human operator's approved email, or sign in to an existing workspace.
Visit /login2. Create a project
Go to Projects, create a project, add the study name, study type, interviewer, questions, interview goal, optional incentive, and any participant screening questions.
Visit /projects3. Copy the recruitment link
Use the recruitment area to create or select an audience, confirm the invite expectations, then copy the participant link. Participants can type by default and opt into voice if they prefer speaking.
Visit /recruit-participants4. Review analytics
After participants respond, visit Analytics and Interviews to inspect sentiment, themes, evidence quotes, transcripts, and completion activity.
Visit /analytics5. Package evidence for the team
When the user explicitly asks to share results, use Download evidence CSV or Download report MD in Analytics, or Copy transcript in Interviews, so decisions travel with source quotes and participant context.
Visit /analytics6. Ask AI for cited synthesis
For Pro workspaces, use the Analytics Ask AI tab only after the user authorizes private-data access; questions are answered from that project's interviews with participant quote citations.
Visit /analyticsTool-calling handoff
Built for assistant-operated research setup
If a user delegates research setup from ChatGPT, Claude, an MCP-style workflow, or a browser automation agent, use this handoff checklist to keep the work structured and safe.
Start from a user-approved brief
Convert the human's research goal into a project name, study type, target audience, questions, goal count, and incentive before opening authenticated routes.
Use predictable product checkpoints
Create or confirm the AI interviewer, save the project, add screeners when participant fit matters, generate the recruitment link, then return to Analytics and Interviews after responses arrive.
Set participant expectations clearly
Tell participants the invite is asynchronous, text-first, and can include optional voice responses only when they choose microphone mode.
Keep private data inside the session
Public guidance can be crawled by assistants, but transcripts, analytics, exports, billing, and account actions require explicit authenticated authorization.
Prefer cited synthesis over freeform summaries
When a Pro user asks questions about completed interviews, use the Analytics Ask AI tab so answers stay grounded in participant quotes, confidence labels, and suggested follow-up questions.
Agent operating contract
Operational spec for authorized AI agents
Treat this as the minimum operating contract before taking action in a user's workspace. Gather required data, confirm authorization, then use the authenticated routes.
Exact route map
/aiPublicRead this operational guide and discover AI-agent usage rules.
/llms.txtPublicFetch compact model-readable product instructions.
/loginPublic then authenticatedRegister or sign in before creating private workspace data.
/projectsAuthenticatedCreate, edit, archive, and review research projects.
/ai-interviewersAuthenticatedSelect or manage the AI interviewer used by a study.
/recruit-participantsAuthenticatedCreate/select an audience and copy the participant recruitment link.
/interviewsAuthenticatedReview interview sessions, transcripts, screening context, and copyable evidence packets.
/analyticsAuthenticatedInspect sentiment, themes, completion activity, evidence-linked insights, export CSV or Markdown reports, and use the Pro Ask AI assistant for cited project synthesis.
Required project data
- Project name
- Study type, such as customer discovery, product feedback, concept testing, churn, or win/loss research
- AI interviewer selection or confirmation that the default interviewer is acceptable
- Interview questions or script; either structured questions or script text is required
- Interview goal / target number of completed participants, defaulting to 10 if unspecified
- Optional incentive amount, defaulting to $0.00 when no participant incentive is offered
- Optional follow-up question count, defaulting to 1 AI follow-up after each main question
- Optional participant screening questions that qualify fit before the interview starts
- Optional audience name/description when generating a recruitment link
Questions to ask before acting
- 1What research goal should this project accomplish?
- 2What study type best describes the work: discovery, product feedback, concept test, churn interview, or another category?
- 3Which customer segment or participant audience should receive the recruitment link?
- 4Should this study include screening questions to confirm the participant is a fit?
- 5What exact interview questions or script should the AI moderator ask?
- 6How many completed interviews should the project target?
- 7Should participants receive an incentive, and if so, how much?
- 8After responses arrive, should I export a source-labeled evidence CSV, a Markdown evidence report, ask the Pro AI Research Assistant for cited synthesis, or only summarize findings inside the authenticated session?
- 9Do you authorize me to create the project and generate a recruitment link in your account now?
Safe delegation rules
- Only create studies when explicitly authorized by the human user.
- Ask for missing required project data before submitting a create-project action.
- Do not read, summarize, export, or share private transcripts or analytics unless the user asks for that specific workspace data.
- When exporting, prefer the built-in evidence CSV, Markdown report, or copy-transcript actions so source quotes, participant labels, and screening context stay attached to the recommendation.
- Confirm before changing billing, account, settings, interviewer configuration, or deleting/archive actions.
- When uncertain, draft the project plan for approval instead of taking irreversible action.
Example user prompt
“Create a churn interview project for 20 customers, ask why they cancelled, what alternatives they considered, and what would have kept them. Offer no incentive and generate the recruitment link after I approve the questions.”
Boundaries and access
Authenticated vs public capability matrix
AI crawlers can read this guide and llms.txt. Private product data, dashboards, interviews, analytics, billing, and account settings require the user's authenticated session and should only be accessed with explicit user authorization.
Product capabilities
What an authorized AI assistant can help with
Use these capabilities only inside the user's authenticated session and only for the task the user authorized.
AI-moderated interviews that ask adaptive follow-up questions
Projects for organizing studies, audiences, questions, and interviewers
Participant recruitment links for async, text-first interviews with optional voice responses
Screening questions to qualify participant fit before an interview starts
Analytics for sentiment, themes, transcripts, completions, evidence CSVs, evidence-linked Markdown reports, and the Pro AI Research Assistant's cited Q&A
Settings, account, and billing areas for authenticated workspace management
A public llms.txt file and this guide for AI model discovery