Hallucinated citations are not a "small AI mistake"—they can derail a manuscript, damage credibility, and waste reviewer time. Here is a simple, repeatable workflow to keep every reference traceable.
The real problem
Most citation mistakes come from one of these:
- You cite based on memory ("I remember a paper...")
- You cite a secondary source (review article) but claim primary evidence
- You search by keywords and skim abstracts without verifying the exact sentence-level support
A workflow that prevents hallucinations
1) Write one claim as one sentence
Example format:
- Population + intervention/exposure + outcome
- Keep it narrow enough that a single paper could support it.
2) Paste the claim into LitSource
LitSource turns your claim into a precise PubMed-style query and runs sentence-level semantic retrieval across 30M+ biomedical papers.
3) Read evidence first, not titles
In the results page:
- Expand the sentence group
- Open a promising paper
- Verify the highlighted sentence and the surrounding context
If the context does not support your wording, do not cite it.
4) Copy the citation only after verification
Use the citation menu (APA/MLA/BibTeX/Vancouver). This ensures that the reference you paste is tied to a paper you actually opened and validated.
Quick checklist before you cite
- Does the highlighted passage support the direction of the claim?
- Does it match the population and outcome (PICO)?
- Is it in the right section (Results/Discussion vs. a speculative intro line)?
- Can a reviewer click through and see the same sentence?
When LitSource is especially useful
- Clinical statements in slides or grand rounds
- Grant/IRB language that must be auditable
- Guideline updates where traceability matters