Skip to content
AI Knowledge MapSearch / Text directory

Before sharing results, what should I disclose about AI's role, verification and limitations?

E035 · E Evaluating and correcting · Trust, use and disengagement

Key answer

Tell recipients what AI did, what was actually checked and which limitations affect use. Follow the setting's rules/agreements; one journal's policy is not a universal obligation.

Full explanation

ICMJE recommends medical-journal authors disclose whether/how AI was used and requires human review/responsibility. Apply this publishing guidance with the target journal's rules. ICMJE V.A For coursework, client work and internal documents, check course rules, contracts and organizational policy, especially independent-work and disclosure requirements.

Use three concise disclosures rather than “human verified”:

· Participation: source discovery, drafting, translation, code or analysis; add tools/versions/dates when rules require.

· Actual checks: what, evidence and coverage. Checking every amount against receipts differs from sampling five records.

· Limitations: unresolved data, untested environments, forecast assumptions and scope, limited to consequential matters.

A business analysis might say:

AI assisted source organization and drafting. The author recalculated every summary from original sales data and checked key citations. Interviews were sampled only; forecasts use supplied growth assumptions and have not been validated against later outcomes.

“Every” and “checked” require work records. Code deliveries should name tests and environments; review is not execution, and passing tests do not prove correctness in every setting.

AI can draft the disclosure, but the responsible person must compare it with records. Disclosure supports appropriate use; it does not automatically improve quality or replace checks.

References

1. ICMJE, Recommendations, V.A Use of AI by Authors, official authorship guidance; accessed October 13, 2026