| UI or UX | UI + UX - Editable generated draft with provenance, review, and apply controls | UI + UX - In-place value or row editing | UI + UX - Primary editable input surface for composing and submitting an AI request | UI + UX - Incremental generated-output surface that renders response chunks before final completion | UI + UX - Inline claim-to-source evidence display for generated or summarized content | UI + UX - Whole-answer source coverage and grounding evidence display |
| UI guidance | 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. | Render a read-state value with a visible edit affordance, then replace only that value, cell, or row with the appropriate input and adjacent Save and Cancel controls when edit mode starts. | Render the prompt box as a labelled, editable composer with visible draft area, send control, context chips, attachment controls, model or mode indicator when relevant, character or token boundary feedback, and clear disabled or blocked-send reasons. | Render streamed output with a visible generation state, partial-answer label, stop control, final-complete state, and clear distinction between text that is still arriving and content that has passed final citation, safety, tool, or format checks. | Render citation markers beside the claims they support, and connect each marker to a selected source preview with title, source type, excerpt, date or version, permission state, and open source action. | Render source grounding as an answer-wide evidence panel that separates source scope, searched sources, retrieved sources, used sources, supported claims, partially supported claims, unsupported claims, and unresolved source states. |
| UX guidance | 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. | Use inline edit when users need to make frequent, low-impact changes to one visible property while preserving surrounding list, table, or record context. | Use a prompt box when users need to author a natural-language request for AI generation, transformation, analysis, or automation and must remain in control of the exact request being submitted. | Use a streaming response when showing partial generated output helps users start reading or monitoring work before the model finishes, and when the product can explain that early chunks may still change, be filtered, or lack final sources. | Use citation display when users must verify where a generated claim, summary, or recommendation came from without leaving the answer context. | Use source grounding display when users need to judge whether an AI answer is backed by the right body of evidence, not merely open one citation. |
| Good UI | 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. | A resource table shows an edit icon beside the owner cell; activating it turns that cell into a text field with Save and Cancel buttons while other cells remain read-only. | An assistant composer labels the selected source as Contract draft, shows Attach file, Use selected text, Format: table, and Send, and blocks sending when the referenced file is no longer available. | A policy assistant shows Answer generating, streams paragraphs into a stable answer region, marks citations pending, exposes Stop generation, then changes to Complete when citations and safety checks finish. | A policy assistant places numbered citation chips after each sourced claim; selecting a chip opens a source preview with the document title, section, quoted excerpt, updated date, and Open source action. | A policy answer includes a Grounding panel showing 4 sources searched, 3 retrieved, 2 used, 5 supported claims, 1 partially supported claim, and 1 unsupported claim with a Review action. |
| Bad UI | A final-looking answer becomes editable with no generated-versus-user-edited distinction, no citation preservation state, and no undo to the generated draft. | Every table cell is always an input, so users cannot tell which values changed or whether a row is still just being read. | A blank AI field shows only Ask me anything and sends vague requests with hidden page context. | A generated answer appears word by word with no partial label, no stop control, and a Copy button that looks ready before sources arrive. | An answer ends with five links under Sources but no marker shows which link supports which claim. | The answer shows a green Grounded badge even though only one citation supports one paragraph. |
| Good UX | 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. | A user clicks Edit owner, changes the value, sees Save become enabled, enters an invalid short value, fixes it beside the field, saves, and stays on the same row. | A user selects a policy paragraph, writes Summarize risks for a non-lawyer, sees Selected text and Output: bullets chips, submits, then edits and resends the exact prompt after the first answer is too broad. | A user sees the first-token state quickly, reads early outline bullets while the answer continues, stops generation after enough detail, and sees the result labelled Partial with Continue and Regenerate options. | A user checks a claim, opens its source preview, compares the quoted excerpt with the answer text, and copies the citation with the source title included. | A reviewer opens the grounding panel, sees that the answer used the current policy but not the outdated FAQ, and flags one unsupported claim before publishing. |
| Bad UX | A user edits a generated compliance summary and all citations disappear, leaving no way to know which claims remain source-backed. | A user types a new value and clicks elsewhere; the product silently discards the draft with no warning. | A user presses Enter expecting a new line and accidentally sends an unfinished prompt to an external model. | A user copies an early legal recommendation before the final paragraph reverses the conclusion after a tool result arrives. | A user trusts a generated compliance claim because it has a number beside it, but the number points to an unrelated source. | A user trusts a generated answer because the product says Grounded, but the source scope was only web search and did not include internal policy. |
| Best fit | Generated content is expected to be revised before it is copied, saved, sent, published, or applied. | Users frequently update one visible value or one row property. | Users must write or revise an AI request before generation, analysis, transformation, or automation begins. | Generated text or structured content can be read or monitored before completion. | Users need to verify generated claims, summaries, recommendations, or extracted facts against source material. | Users need answer-wide evidence coverage before trusting generated content. |
| Avoid when | Users only need to write or revise the request before generation. | The edit affects multiple dependent fields or needs a review step. | The task is better expressed as a fixed form, button, or command with known parameters. | Intermediate chunks may expose unsafe, private, or misleading content. | The product cannot reliably map claims to sources or label unresolved citations honestly. | The system cannot determine source scope, retrieval status, or claim support reliably. |
| Required state | Generated draft state with original generated content, creation time, model or run reference, and source coverage visible. | Read state with displayed value and discoverable edit affordance. | Empty prompt state with label, helpful instruction, and no implied hidden submission. | Queued or receiving state before first output arrives. | Default answer with cited claims and inline citation markers. | Default grounded state with source scope, searched sources, retrieved sources, used sources, and supported-claim count. |
| Accessibility burden | 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. | Give edit, save, cancel, confirm, and dismiss controls accessible names, especially when icons are used. | Provide a programmatic label for the prompt editor and named controls for send, attach, remove context, clear, retry, and cancel. | Expose stream milestones such as started, still generating, stopped, failed, citation ready, and complete as status messages. | Give citation markers accessible names that include their selected state and source status, such as Citation 2, verified source, or Citation pending. | Expose grounding summary, source scope, status counts, unsupported claims, and source groups as text. |
| Common misuse | Letting users edit a final-looking AI answer without generated-versus-user-edited status. | Using inline edit for high-impact changes that need a review or confirmation step. | Showing Ask anything as the only instruction while hiding what sources and tools the model can use. | Showing a blinking cursor with no state, stop control, or elapsed feedback. | Displaying a link dump below the answer instead of mapping sources to specific claims. | Showing a global Grounded badge when only some claims have evidence. |