RESEARCHWEAVE

READING GUIDE · By ResearchWeave ·

How to Check AI-Generated Paper Citations

A working link answers “does this source exist?” It does not, by itself, answer “does this source support this sentence?” Use this checklist before relying on an AI-generated research summary.

1. Separate the claims in the sentence

Copy the sentence you want to check. Underline each distinct assertion: what changed, in whom, compared with what, and over what period. A sentence can contain several assertions even when it ends with a single citation.

Illustrative example—not a study finding

“Intervention A improves outcome B, is safe over the long term, and works for all adults.”

That sentence requires separate checks for the outcome, safety follow-up and population. Evidence for one does not automatically establish the other two.

Start with the claim that matters to your next step, rather than treating a reference list as a pass-or-fail certificate for the whole answer.

2. Find the actual source

Open the citation and match the title, authors, publication year and identifier to the source being discussed. Check whether you are reading an original study, a review, a preprint or a commentary. Look for any correction, updated version or retraction notice on the publisher’s page.

If the citation leads to a review, identify whether the particular claim depends on an underlying study. Follow that study when you need to assess its methods or results. A search-result snippet is a pointer, not a substitute for the paper.

If you only have the abstract, record that limitation. Do not infer details about methods, subgroup results or safety that you cannot inspect.

3. Match the claim to the evidence

Use these questions as a reading worksheet:

  • Population or system: humans, animals, cells, or a computational dataset? Does it match the question?
  • Study design: what comparison was made, and what does that design allow the authors to conclude?
  • Outcome: what was actually measured? Is it the same outcome named in the AI answer?
  • Magnitude and uncertainty: what result and uncertainty did the authors report? Does the summary preserve both?
  • Time and scope: how long was follow-up, and does the summary extend beyond the studied setting?

Locate the relevant passage, table or figure. Check whether it is a measured result, an interpretation in the discussion, or a hypothesis for future work. Keep those categories distinct in your notes.

4. Record what remains uncertain

For your own reading notes, label each claim as supported within the study’s scope, partly supported, contradicted, or not yet verifiable. These are suggested note-taking labels—not an automated ResearchWeave scoring feature.

Write down the source location and the reason for your label. If necessary, narrow the sentence to what you could verify. Do not convert “I could not verify this” into “this is false.”

A reusable note

Claim: …
Source and passage/table: …
Population, design and outcome: …
What the source supports: …
What remains unverified: …
Next paper or section to read: …

This checklist is a starting point for reading, not a formal risk-of-bias assessment or systematic review. Higher-stakes decisions require appropriate expertise and fuller appraisal.

See the format, then inspect the sources

ResearchWeave is an early pilot for evidence-orientation briefs. AI output can be wrong, and a linked paper is not a guarantee of support.

Read the sample excerptAsk about a pilot trial

Access is confirmed individually. Do not send confidential or patient-identifiable information.

Related guide: How to turn a broad topic into a focused research question.