Quick answer
Can ChatGPT make up citations and sources?
A reference can look academically complete and still be invented, mismatched, or impossible to verify.
Four ways an AI citation can fail
Entirely fabricated
The title, author, journal, case, report, quotation, or URL does not exist outside the generated answer.
Real pieces, false combination
The author and journal exist, but the title, year, volume, or conclusion belongs to something else.
Real source, wrong support
The linked page exists yet does not say what the answer attributes to it, or applies to another date or context.
Uncheckable reference
The answer omits enough metadata that the reader cannot identify the original work or relevant passage reliably.
Check a citation in five steps
- 01
Freeze the claim
Copy the exact sentence the citation is supposed to support. Do not verify only the general topic.
- 02
Search exact metadata
Search the quoted title, author, publication, year, identifier, case number, or URL independently of the chat.
- 03
Open the original
Use the publisher, journal, court, regulator, standards body, repository, or official documentation page where possible.
- 04
Match the passage
Confirm that the source supports the exact claim, population, version, jurisdiction, time period, and degree of certainty.
- 05
Record the result
Mark the citation verified, mismatched, fabricated, outdated, or unresolved before reusing it.
Why a working link is not enough
A link proves only that a page exists. The answer may have selected a weak source, misunderstood the relevant passage, omitted a qualification, or attached the source to a stronger claim than it supports.
For academic and professional work, verify the bibliographic details and the claim-to-passage relationship separately. A correct citation format cannot rescue incorrect attribution.
Use ChatGPT to expose the weak spots
This prompt can produce a better audit trail, especially when search or supplied documents are available. It still does not replace opening the original source yourself because the model can misquote or misread the page during the audit.
Where Makes Mistakes fits
The extension makes the provider warning harder to ignore and adds a fast re-check prompt after an answer. It does not inspect references or determine whether a source exists.
Use the reminder to start the audit, then use bibliographic databases, official repositories, primary documents, and the original passage to finish it.
Sources and further reading
Product details and guidance were checked against these first-party pages on July 29, 2026. Re-check current listings before making an install or high-stakes decision.
- Does ChatGPT tell the truth?
OpenAI Help Center
OpenAI's guidance on incorrect answers, fabricated citations, confidence, search tools, and verification.
FAQ
Questions people ask
Why does ChatGPT invent references?
It generates plausible language and metadata patterns. When the required source is missing from its available evidence, it may still complete the pattern with details that look credible.
Can ChatGPT cite a real paper incorrectly?
Yes. A paper may be real while the title details, quotation, page number, method, population, or conclusion attributed to it is wrong.
How do I verify a ChatGPT citation quickly?
Search the exact metadata, open the original publisher or repository page, and compare the relevant passage with the exact generated claim. Mark any mismatch explicitly.
Should I cite ChatGPT as the source?
For external factual claims, cite the original evidence rather than treating the generated answer as the authority. Follow the citation rules of your institution or publisher for documenting AI use itself.