Language editing in the artificial intelligence era: achieving epistemic integrity

Large language models (LLMs) are increasingly used to write and edit scientific manuscripts; however, their fluency can obscure mismatches between linguistic claim strength and evidential support. This tutorial examines such mismatches through the lens of epistemic integrity, defined as accurate alignment of scientific statements with the quality, strength, and scope of the underlying evidence. It presents sentence-level epistemic markers that require careful human judgment, including reporting verbs, hedges, tense choices, evaluative adjectives and adverbs, and local scope markers. Inappropriate modification of these features can overstate or understate certainty, strengthen or weaken a claim beyond what the evidence warrants, or distort an author’s intended emphasis. The tutorial also addresses discourse-level issues, including the need to clearly distinguish between results and interpretation, shifts in certainty across manuscript sections, consistency in scope limitations, and cumulative changes in evaluative tone. Because LLMs cannot independently assess study design, methodological limitations, disciplinary conventions, or evidential sufficiency, polished revisions may introduce epistemic miscalibration even when they are grammatically correct. Editors should therefore evaluate not only local wording but also the consistency of claims across the manuscript. Human editorial expertise remains essential for preserving rhetorical coherence, evidential accuracy, and scientific integrity in artificial intelligence–assisted academic writing, alongside attention to grammar, style, clarity, and readability.