When an AI assistant invents a source: how to find out

Short answer: if the AI cited a link, a news story or a study for an everyday question, search the title in a search engine, open the link and check that the source really says what the AI claims. If you cannot find it through independent channels, treat it as unverified. This guide is for everyday users, with no need for academic databases.

What helps before you start

• The AI's answer with the references in view.
• Access to an ordinary search engine.
• Special attention to numbers, literal quotes and study names.

Why it happens

Language models generate plausible text from patterns. When asked for a source, they may produce a reference with the right format that does not exist, or mix parts of real sources. This is known as hallucination. Even tools with web access can cite pages that do not support the claim.

Steps

1. List the cited sources. Separate title, authors, year, publication and link.
2. Search the title in quotation marks. If the source exists, it normally shows up in search engines, on the publication's own site or in other coverage of the topic.
3. Check the authors and the publication. Does the person exist? Does the journal or site exist and publish that?
4. Open the link. If you get an error, a strange redirect or unrelated content, be suspicious.
5. Check the content. Read the relevant passage and see whether the source really says what was attributed to it.
6. Watch for warning signs. Too many overly precise details, references that fit the argument perfectly and literal quotes without an exact location.
7. Ask the AI to point to the passage. It can say where it thinks the passage is, but it cannot confirm that it exists: check for yourself on the link or the publication's site. If it cannot say where it is, the claim deserves doubt.
8. Replace it with a checked source. If you cannot find it, use another source you verified or remove the claim.

An example

Fernanda asked the AI for data on a health topic, with references. The AI listed three articles. On searching, two did not exist and the third dealt with a different subject. She redid the search on a well-known outlet's site, found real sources and rewrote the text based on them. This is an illustrative example.

Common mistakes

• Trusting a reference because it has a professional format: format is easy to imitate.
• Checking only whether the link opens: the page may exist and not say that.
• Asking the AI itself to confirm: it may confirm wrongly.
• Copying quotes without checking: this spreads the error.

Next step

Next time you use AI for research, treat each source as a lead, not as proof. Verify at least the ones that support the most important claims.