Categorise with AI action settings
Add Categorise with AI from Page actions → Add action. It selects one category from your list, or returns no match. It does not supply numeric scores, weights or several selected categories.
Configure inputs and categories
Use the shared AI text/image inputs and limits. The stored action type is aiCategorise.
| Property | Meaning |
|---|---|
aiCategories | Between 1 and 100 categories. |
Category id | Unique, stable identifier, up to 100 characters. Conditions and integrations should use it. |
Category label | Visible name, up to 200 characters. |
Category description | Optional inclusion criteria, up to 1,000 characters. |
For example, a facilities service might define water/plumbing, heating/ventilation and access/doors. Describe their boundaries, and preserve each ID when improving its wording. Optional additional instructions can explain the handling of overlapping reports.
Read match and no-match results
{{ actions.categoryAction.response.categoryId }}
{{ actions.categoryAction.response.categoryLabel }}
{{ actions.categoryAction.response.status }}| Outcome | Result | Handle it by |
|---|---|---|
| Matched | status="matched", with the configured category ID and label. | Use the ID for a suggestion or route, with appropriate review. |
| No match | status="no_match", categoryId=null, categoryLabel="No matching category". | Provide a deliberate fallback. This is a successful result, not an action failure. |
| Processing failed | Normal action error and response code. | Use the service's failure route; do not treat it as a category. |
No-match results are chargeable on metered accounts like other successful AI results. Review AI suggestions against the original content; they do not establish eligibility or a staff decision.
Save a category and control visibility
Use Add result to form for response.categoryId, response.categoryLabel or response.status. These are read-only calculated fields, with independent page, applicant-answer and review visibility. To ask someone to confirm a suggestion, collect their confirmation in a separate ordinary answer.
Test the boundaries
Test each category, multiple possible categories, vague/out-of-scope content, image-only input, combined input, no match and a failed call. User view is mocked; saved-settings Test AI is real, separate from other actions and uses account capacity. The shared 15-second deadline, input limits and charging rules are described in Summarise with AI.