Pulse · 2026-07-18 · Analysis
When should you ignore an AI citation?
Ignore an AI citation when you cannot open and verify the source yourself; fluent footnotes are not evidence.
AI assistance: Human-authored craft note; no generative filler.
Direct answer: Ignore an AI citation when you cannot open the source and confirm it says what the model claims. A confident footnote is a style choice until you verify it.
What this means at work
Models can invent papers, policies, URLs, and “standard industry rules.” The text looks finished. The risk sits with you the moment you forward it.
Evidence is something a skeptic can reopen: a ticket, a log, a doc, a page you actually load. If the citation only exists inside the chat, it is not a citation.
Do this next
- Highlight every claim that would hurt if wrong.
- For each citation-shaped claim, try to open the source.
- If you cannot open it, mark the claim unverified and remove or rewrite it.
- Ask a second prompt: “Which claims lack a source I could open? List them.”
Your reputation attaches to the send button, not to the model. Ignoring a fake citation is not rudeness. It is craft.
FAQ
What if the link looks real but 404s?
Treat it as unverified. Do not ship a dead or guessed URL as proof.
Can I ask the model to fix the citation?
You can ask. You still must open whatever it returns. The check stays human.
See also
- Field Guide: Verify
- Challenge: Why your AI sounds confident when it’s wrong
- Checklist: AI craft principles
Harder questions live in Challenges.