AI systems can produce a plausible title, author or link that does not exist. This is especially risky for schoolwork, health, finance and legal research, where a confident answer is not proof.
Short answer: Ask for primary sources with working links, open every source yourself, and remove any claim you cannot verify independently.
Check the likely cause first
- Look up the quoted title or DOI in the publisher, journal or official database rather than trusting the chatbot link.
- Compare the claimed date, author and conclusion with the actual page.
- Treat an answer without verifiable sources as a starting hypothesis, not a finished citation.
Fix it in the safest order
- Ask the model to separate confirmed facts, assumptions and unknowns.
- Request primary documentation and specify that it should say “not found” instead of inventing a citation.
- Open each source and search within it for the claim being attributed to it.
- Save the verified title, author, date and URL in your own notes before writing.
Change one thing at a time and retest. That makes it easier to identify the cause and undo a setting if the result is worse.
How to confirm the fix
A usable citation should resolve to a real source whose text actually supports the nearby claim. Two unrelated sources are better than two pages repeating the same press release.
What not to do
- Do not cite a search-result snippet without opening the page.
- Do not ask the AI to “sound certain”; uncertainty is useful information.
When to contact support
For high-stakes decisions, ask a qualified professional or librarian to review the primary evidence and the way it is being interpreted.
For more step-by-step help, browse the PHWhy topic hubs or use the site search for the exact error message.
