| UI or UX | UI + UX - AI output correction feedback flow with selected claim, reason, scope, consent, receipt, and review state | UI + UX - Submitted search query correction and did-you-mean result state | UI + UX - Field-level validation feedback | UI + UX - Object-attached comment composer and comment list with authorship, replies, state, permissions, and moderation | UI + UX - AI response regeneration and same-turn retry control with version and context preservation | UI + UX - Editable generated draft with provenance, review, and apply controls |
| UI guidance | Render correction feedback as an AI-output control that anchors feedback to a specific answer, claim, source, recommendation, tool result, or message instead of a detached thumbs signal. | Show the original submitted query and the corrected query in the result state, for example Showing results for benefit appeal with a Search instead for benifit appeel control. | Render a labeled field with hint text, field-adjacent error text, invalid styling, preserved value, and corrected state. | Render comments as anchored contributions with author identity, timestamp, body, optional attachment or selection context, edited state, reply target, and state labels such as open, resolved, hidden, deleted, or assigned. | Render regenerate / retry as a response-level control that names whether it will rerun the same prompt, continue after failure, regenerate a new answer version, or retry failed tool and source work. | Render editable AI output as a generated draft with a clear boundary between original generated text, user-edited text, tracked changes, source mapping, citation preservation, and final applied output. |
| UX guidance | Separate correction capture from regeneration and manual editing: users should know whether their feedback updates this answer, future answers, reviewer queues, model improvement data, or none of those. | Use query correction to recover from likely spelling, spacing, transposition, or phrase mistakes after a user submits a free-form search. | Help the person reach a clear task outcome. | Preserve comment drafts, anchors, authorship, notifications, edit history, deletion outcomes, and resolved or hidden states so conversation context remains trustworthy across reloads, permissions, moderation, and return visits. | Use regenerate / retry when users need another AI attempt for the same submitted request, or need recovery from a failed, stopped, low-quality, stale, blocked, or partially generated response. | Use editable AI output when users need to revise generated content after creation while retaining provenance, citations, source coverage, and a deliberate apply or save contract. |
| Good UI | An assistant answer lets a user select a suspect claim, choose Wrong answer, add the expected answer and corrected source, opt out of training use, and see Feedback received with review routing. | A results page says Showing results for benefit appeal, marks benefit and appeal as corrected tokens, and offers Search instead for benifit appeel. | Error text appears next to the field with readable spacing, persistent label, hint text, and the invalid value still visible. | A document margin comment shows the selected paragraph, author, timestamp, body text, Reply, Resolve, Assign, and Copy link actions with the composer focused on that selection. | A chat answer shows Regenerate answer, Retry failed sources, and Compare versions with the original prompt, source scope, model, and tool changes visible. | A policy assistant shows a generated draft answer with citation chips, user-edited spans, tracked change controls, source mapping indicators, and Apply output disabled until unsupported edits are reviewed. |
| Bad UI | The feedback form captures only sentiment and loses the correction reason or expected answer. | The search field silently changes benifit appeel to benefit appeal and the original query is gone. | Error message appears far from the field it describes. | A Notes textarea sits under a record and calls itself comments even though every user overwrites the same field. | A Try again button silently changes the prompt, model, sources, and tools, then overwrites the previous answer and citations. | Citations vanish or remain attached to unsupported edited claims. |
| Good UX | A user flags a wrong eligibility answer, attaches the correct policy source, applies the correction to the current answer only, and receives a reviewer-routed status. | A user searches benifit appeel, sees corrected results for benefit appeal, checks the changed tokens, and can switch back to the original query. | Validation appears after blur or submit when it helps correction. | A reviewer comments on a selected line, adds an action item for Dana, receives a reply, resolves the comment, and can reopen it from the resolved filter. | A user sees answer v1 has stale citations, chooses Regenerate with refreshed sources, compares v1 and v2, then restores v1 because v2 lost a required caveat. | A reviewer changes one sentence, sees it marked as user edited, accepts the tracked change, reviews a stale source warning, and applies the output only after unsupported text is resolved. |
| Bad UX | A user reports that a cited source is wrong, but the feedback is stored as an undifferentiated dislike and future answers keep using the same bad source. | The UI reports zero results for an obvious typo even though a high-confidence correction is available. | Showing errors before users type. | A user writes a long comment, loses network connection, and the draft disappears when the page reloads. | Changed source scope, model, tools, or context are hidden between versions. | Citations vanish or remain attached to unsupported edited claims. |
| Best fit | Users can identify wrong, unsupported, stale, unsafe, biased, irrelevant, or wrongly sourced AI output after it is generated. | Users submit free-form searches where spelling and typing mistakes commonly block useful results. | A single field has a specific correctable problem. | Feedback must preserve author, timestamp, reply, resolution, assignment, or moderation context. | A user needs another AI-generated answer for the same request or a visible recovery path after response failure. | Generated content is expected to be revised before it is copied, saved, sent, published, or applied. |
| Avoid when | Users are editing a generated draft directly rather than reporting what the AI got wrong. | The input is a constrained value selector better served by autocomplete. | The main recovery task is finding several errors across a submitted page. | The user is simply entering a long answer into a form field. | The task is a simple non-AI operation retry already covered by the Retry pattern. | Users only need to write or revise the request before generation. |
| Required state | Initial AI answer state with feedback entry points attached to the answer, individual claims, citations, recommendations, or tool-derived values. | Auto-applied correction state showing corrected results, original query, changed tokens, and search-original action. | Invalid field with preserved user value and adjacent correction text. | Empty comment list and first-comment composer. | Initial answer state with prompt snapshot, response version, source scope, model or mode, and available regenerate or retry controls. | Generated draft state with original generated content, creation time, model or run reference, and source coverage visible. |
| Accessibility burden | Expose selected claim, selected source, reason, scope, privacy choice, submission state, receipt, review state, and outcome as text. | Announce correction status and result counts through a polite status region after submission. | Expose invalid state on the input and connect error text to the field description where possible. | Label the comments region with the object or selection being discussed. | Expose same prompt, changed context, regenerating, retrying, version created, compare available, cooldown, exhausted, blocked, and restored states as text. | Expose generated draft, user edited, tracked change, unsupported edit, unsafe edit, stale source, review required, accepted, rejected, saved, copied, regenerated, and applied states as text. |
| Common misuse | Using a thumbs-down or smiley reaction as the only mechanism for correcting wrong AI output. | Treating every fuzzy or semantic match as a corrected query. | Showing field errors before users have interacted with the control. | Using one shared Notes field as a comment system and overwriting prior contributors. | Using one Try again button for same-prompt retry, prompt edit, tool retry, source refresh, and alternate answer generation. | Dropping all citations when one paragraph is edited. |