AI classification
Categorise with AI processes text, uploaded images or both into one of the categories defined for your service. Use it to suggest a report type, priority or handling team while retaining the original evidence.
Define classification buckets
Each category has a stable ID, label and optional description. Explain what belongs in it and distinguish overlapping categories. For example, a facilities service could define water/plumbing, heating/ventilation and access/doors.
The action returns one category or an explicit successful no-match result. It does not choose several categories or calculate a numeric priority score. Preserve IDs when changing labels or descriptions.
Choose the source and result
Add the action after its source answers are available. Select a text field or Liquid text template, accepted uploaded images, image URLs, or a combination. Use actions.categoryAction.response.categoryId for the selected ID and response.status to distinguish matched from no_match. See the Categorise with AI settings reference for limits and exact result fields.
Use classification in the journey
Keep a deliberate fallback for no match and processing failures. A suggestion can guide a route or help staff triage; it does not prove eligibility or make a review decision. If respondents should confirm it, collect that confirmation as an ordinary answer.
Use Add result to form to save a read-only category field. New fields are visible to authorised reviews by default and hidden on the service page and applicant answers until enabled.
Prepare examples for testing
Test every category, mixed topics, vague/out-of-scope content, image-only and combined inputs, no match and failure. User view uses free mocks. Test AI uses saved settings for a real, potentially chargeable example call without running other actions or saving answers. Real QA calls also use account AI capacity.
See AI summarisation and AI capabilities overview.
