AI form drafting and feedback reports
Use AI to draft a survey, search feedback and summarize a bounded set of inbox responses.
Choose the right workflow
Pollenate has three distinct AI-assisted workflows:
| Workflow | Input | Result |
|---|---|---|
| Draft a form | Your instructions and optionally existing questions | A proposed title, description and question set to review |
| Search customer feedback | A natural-language query | Matching inbox feedback; a text fallback is available |
| Generate an insight report | A selected feedback period, optionally a brand or inbox | A summary, themes, recommendations and highlights |
Connecting an external assistant through MCP is another way to operate forms and read data. It is separate from recording ratings about your own chatbot.
Draft and review a survey
In Feedback Pages, choose the AI creation option and describe the audience, decision and question count. For example:
Draft three onboarding questions for developers who just installed our API. Ask how easy setup was, which step was unclear, and whether a code example would help.
Review wording, required fields and answer options before saving. Generation itself does not publish a form or collect feedback.
For an authorized server integration, POST /organizations/ORG_ID/feedback-pages/ai/generate accepts:
{
"prompt": "Draft three questions about API onboarding.",
"existingQuestions": []
}prompt is 1–2,000 characters. At most 50 existing questions can be supplied. The response contains the proposed form and ordered questions; save it through form management. Machine credentials need admin on this generation route; the UI uses the signed-in person's form-management permission. Generation can return 502 for an invalid model result or 504 when the request times out.
Request a feedback report
curl https://api.pollenate.dev/organizations/ORG_ID/feedback/insights \
-H "X-Pollenate-Key: $POLLENATE_ADMIN_KEY" \
-H "Content-Type: application/json" \
--data '{ "days": 30, "inboxId": "INBOX_ID" }'Alternatively provide since and until timestamps; optional brandId narrows the input. Reports use inbox feedback events, not every answer in Feedback Pages. The current report samples up to 200 newest events and uses up to 100 nonempty comments, each shortened to 300 characters, for its text analysis. It is a bounded view, not an exhaustive analysis of an unlimited dataset.
Introductory access (trial) cannot generate insight reports. Starter, Team and Enterprise are not rejected by that plan check; account permissions still apply. The current API error text says “paid plan” when it rejects a trial account, although Starter also passes the check.
Interpret the output
AI themes and recommendations are suggestions. Check the source comments, dates and sample size before making decisions. A high rating with no comment gives less explanatory evidence than a detailed response. No data returns an empty report rather than invented findings.
Feedback comments are sent for embedding, and report/form-generation inputs are processed by the configured AI service. Avoid adding credentials or information unnecessary for the task. The product does not promise automatic issue prioritization, automatic voice transcription or a custom-trained model.
Next: semantic feedback search or measure assistant quality with the collection API.