Feedback
Collect quick, specific feedback on AI output that users actually give, and that the team can actually use.
5 states4 edge casesDesktop + Mobile
01The problem
Teams building AI features need to know when output is wrong, but users rarely explain. A thumbs-down with no context says something failed, not what. Long feedback forms get skipped. The most valuable signal — what the user changed after receiving the output — usually isn't captured at all.
02When to use
Use when
- Chat assistants and copilots
- Generated drafts, summaries and suggestions
- AI classifications users can see and correct
Avoid when
- Output users can't evaluate (they lack the knowledge to judge correctness).
- Every single interaction — prompting feedback constantly reduces both response rate and quality.
03Recommended UX
Statecraft recommendation
Offer a one-tap rating on every completed response, ask for a reason only after negative ratings using a short list of specific chips with optional text, and capture implicit signals — copy, edit, regenerate, undo — automatically.
04Desktop behaviour
On desktop, rating controls sit in the response's action row and the follow-up appears inline below the response — never as a modal.
Desktop rules
- Thumbs up/down alongside Copy and Regenerate, visible on the latest response.
- After thumbs down, show 3–5 reason chips inline with an optional text field and Send.
- Show the selected state persistently so users know their rating was saved.
- Allow changing or removing a rating.
05Mobile behaviour
On mobile, keep rating one tap and present reasons in a small bottom sheet that can be dismissed without answering.
Mobile rules
3 mobile rules are included in Pro.
Statecraft Pro
Unlock the full pattern
Get the complete interaction flow, states, edge cases, mobile rules and implementation prompt.
- Interaction flow · 5 steps
- 3 mobile rules
- 5 states
- 4 edge cases with recommendations
- 4 microcopy comparisons
- Accessibility notes
- Developer notes
- Claude Code / Codex implementation prompt
Edge cases covered in this pattern
- Feedback contains sensitive data
- Rating a regenerated response
- Negative rating on a refusal
- Low response rate