AI answer without sources

Problem Context

• The answer may be generated from web search, file search, enterprise knowledge sources, selected documents, uploaded files, chat history, model prior knowledge, tool outputs, or no retrieval at all.
• The answer contains claims users may rely on for policy, compliance, support, operations, money, safety, medicine, law, customer communication, product decisions, or public publication.
• The system may have skipped retrieval, failed retrieval, hidden permission-limited sources, mixed model knowledge with retrieved evidence, or returned a no-source answer after a source-backed request.
• Users need to know whether the answer is sourced, not searched, unsupported, permission-limited, stale, or draft-only before acting on it.

Selection Rules

• Flag this anti-pattern when an AI answer makes evidence-dependent claims and provides no source link, citation marker, grounding panel, source scope, or verification action.
• Flag it when the UI presents a polished answer as final even though source search was not run, failed, or returned no supporting source.
• Use source grounding display when the answer needs answer-wide source scope, searched/retrieved/used evidence, and unsupported-claim coverage.
• Use citation display when individual claims need inline citation markers and source previews.
• Use confidence / uncertainty display when the issue is model reliability or calibrated uncertainty; confidence is not a replacement for sources.
• Use streaming response when source events are still pending during generation; keep the answer labelled pending until sources settle.
• Use correction feedback when users need to report that an answer is wrong, unsupported, or uses the wrong source after the answer is shown.
• Use AI limitation onboarding when users need upfront expectations about what the system can and cannot source before they ask.
• Do not show related links, suggested reading, or a generic Learn more list as if those links were evidence used by the answer.
• If no source is available, label the answer as not sourced, not searched, no supporting source found, or draft from model knowledge, and provide a path to search, upload, cite, ask a human, or continue with explicit risk.

Required States

• Sourced answer state with citations or grounding visible.
• Not searched state where retrieval was not run and the answer is labelled accordingly.
• No supporting source found state with unsupported claims separated from answer text.
• Retrieval failed state with retry source search path.
• Permission-limited source state that explains an access gap without pretending the answer is verified.
• Pending source search state while streaming or retrieval is still resolving.
• Draft from model knowledge state that blocks source-dependent apply or publish actions.
• User-provided source needed state with upload, connect, or choose source action.
• Human verification state for high-stakes answers that remain unsourced.
• Copy/apply gated state when sources are missing.
• Bad confident answer state with no source route.
• Bad related links as evidence state.
• Bad hidden retrieval failure state.
• Bad permission omission state.
• Bad citation promised but missing state.
• Bad stale source claim state.
• Bad copy final state that strips no-source warning.

Interaction Contract

• Evidence-dependent answers must expose whether sources were searched, retrieved, used, unavailable, permission-limited, failed, or not required.
• Copy, publish, apply, send, export, and automate controls must preserve or surface no-source status when the answer lacks evidence.
• No-source answers must offer at least one recovery path when possible: search sources, choose corpus, upload file, request access, retry retrieval, ask a human, or mark as draft.
• When users intentionally continue without sources, the UI records the no-source state and does not convert it into a verified answer.
• If sources arrive after streaming, answer status updates from pending to sourced, unsupported, permission-limited, or no supporting source found.
• Related links are clearly separated from evidence used by the answer.
• Accessible names and status text include the source state, not only a visual badge.

Implementation Checklist

• Track answer provenance as structured state: source mode, source scope, retrieval status, used evidence, unsupported claims, permissions, freshness, and no-source reason.
• Render no-source and not-searched labels beside the answer summary, not only in hidden metadata or logs.
• Provide search, retry, source selection, upload, request access, ask human, and draft-only actions when those routes can change the source state.
• Block or warn before copy, publish, apply, send, or automate when evidence-dependent claims remain unsourced.
• Preserve source state in copied text, exports, audit logs, handoffs, and regenerated answer versions.
• Do not replace missing sources with confidence percentages, related links, generic disclaimers, or decorative citation numbers.
• Test no retrieval, retrieval failure, zero results, permission-limited sources, stale sources, streaming pending, mobile layout, high zoom, keyboard flow, and screen reader status.

Common Generated-UI Mistakes

• Showing a polished answer with no citations or grounding after the user asked for a policy-backed answer.
• Adding a generic disclaimer while still letting users copy or send the unsourced answer as if it were verified.
• Hiding retrieval failure and returning an answer from model memory.
• Showing related links below the answer without saying they were not used as evidence.
• Promising citations in the prompt or product copy and then returning an answer with no source state.
• Letting regenerated or edited output drop citations without changing the answer status.
• Treating high confidence as a substitute for source evidence.

Critique Questions

• Which claims in this answer require evidence before users act?
• Were sources searched, retrieved, used, unavailable, permission-limited, or not searched?
• Can users tell whether the answer came from retrieved evidence or model prior knowledge?
• What happens when retrieval fails or finds no supporting source?
• Does copy, publish, apply, or send preserve the no-source warning?
• Would source grounding display, citation display, confidence / uncertainty display, streaming response, or correction feedback be the precise pattern?