AI tools can now draft a convincing scientific paragraph in seconds. The problem is that a paragraph can be fluent, confident, and well structured — and still be wrong. Before you reuse AI-generated scientific content, run one question first: where's the evidence?
1. Pin down the claim
Separate checkable statements ("X reduces Y by 30%") from background and opinion. Vague claims are hard to verify — and that is a warning sign in itself.
2. Check that the source exists
Look up every reference on PubMed, via its DOI, or on the journal's website. Realistic-looking but non-existent references are a well-known AI failure mode.
3. Check that it actually says that
A real paper can still be misquoted. Compare the population, intervention, outcome, and conclusion with what the AI claims.
4. Weigh the evidence
A systematic review of randomised trials carries more weight than a single observational study, which carries more weight than an opinion piece.
5. Catch overstatements
Watch for "proven", "cures", "always", causal language built on correlations, and relative risks presented without absolute numbers.
Want to practise this in ten minutes? MODO's Research Check turns these five steps into ten one-minute lessons, with a checklist and templates you can reuse.
