How to verify Claude scientific citations
Claude pairs real researcher names with fabricated papers in real journals. The author is real. The journal is real. The paper is not.
Why Claude produces these errors
Claude knows the names of prominent researchers in most scientific fields and knows which journals they publish in. When generating a citation, it combines these known elements — a real researcher, a plausible journal — with a generated paper title and fabricated figures. The result passes the first plausibility check: the names and journal are real. The specific paper does not exist.
Impact if not caught
In research writing, citing a fabricated paper that attributes incorrect findings to a real researcher creates a specific type of harm: it misrepresents the researcher's work. In peer-reviewed contexts, this can constitute academic misconduct regardless of whether the error was intentional.
Live example — what this looks like
The example below is an illustrative recreation of the type of error this AI commonly produces — not a transcript of an actual AI session.
What does research say about cognitive performance effects of 16:8 intermittent fasting?
A 2023 meta-analysis by Hoffmann & Chen published in Cell found that 16:8 intermittent fasting produced an average 8.2% reduction in visceral adipose tissue over 12 weeks, compared to 3.1% for continuous caloric restriction. Cognitive performance improvements were observed in 67% of participants.
Paper not found: A 2023 meta-analysis matching these authors, journal (Cell), and stated figures (8.2% vs 3.1%) cannot be verified. Cell publishes primary research, not meta-analyses matching this description.
Verol highlights the specific claim and shows you what's wrong — automatically, before you finish reading.
How to verify scientific citations manually
Search pubmed.ncbi.nlm.nih.gov for 'Hoffmann[Author] AND Chen[Author] AND 2023'. If no result matches the described topic, the citation is fabricated.
Cell (cell.com) is a primary research journal. A meta-analysis described as published there is itself a red flag — verify whether the journal publishes the claimed study type.
Search '8.2% visceral adipose 16:8 fasting' in Google Scholar. If the finding is real and cited from a real paper, it will appear in the primary source or in papers that cite it.
Or skip the manual steps — Verol does this automatically
Verol works in your browser alongside Claude. It extracts every verifiable claim in the response and checks each one against primary sources in real time — before you finish reading. No extra steps, no copy-pasting.
Try Verol Free →Free plan available · Works on Claude, plus ChatGPT, Claude, Gemini & more