Alkaissi & McFarlane's 2023 Cureus paper attributes 'artificial hallucination' to Ji et al.'s 2022 survey rather than coining the term themselves

aillm-hallucinationterminologymedical-ainlp-historycitation-genealogycorrection

Alkaissi & McFarlane's "Artificial Hallucinations in ChatGPT: Implications in Scientific Writing" (Cureus, 2023) — widely treated, including previously by this vault, as the paper that coined or popularized "artificial hallucination" for AI output — uses the term in its title and in its opening paragraph without any citation (ChatGPT described as "a natural language generator (NLG) yet able to produce artificial hallucinations"). The attribution comes later, in the summarizing passage after the two ChatGPT experiments: "Such a phenomenon has been described as 'artificial hallucination' [1]." Reference [1], in a four-item reference list, is Ji et al.'s 2022 survey (claim-ji-et-al-2022-survey-documents-hallucination-cv-to-nlp-origin). Read directly against its own citation, the paper borrows the description rather than introducing it. The borrowing is marked in the body text rather than left to the bibliography alone — but several paragraphs after the term's first appearance, not in the sentence that introduces it.

The editorial also does not define the term in its authors' own words: the paragraph immediately following the cited sentence opens "ChatGPT defines artificial hallucination in the following section," and the definition that follows is quoted chatbot output, flagged as such in the paper's acknowledgements ("Some of the text within the manuscript has been generated with the aid of ChatGPT and has been put between quotation marks").

This corrects entity-artificial-hallucination, the vault's own entity page for the term, which described it as "coined/popularized in the medical-AI literature by Alkaissi & McFarlane" without that page's author having read the 2023 paper's own reference list. The paper is still a significant, widely-cited early application of the term to LLM output in a medical-writing context — that part of its reputation is earned — but "first introduced the term" specifically is not.

Correction history.

• 2026-09-11 — The note said the paper "states that ChatGPT can 'produce artificial hallucinations' and immediately adds" the citing sentence, and that "the attribution sits in the very sentence that first uses the term, not buried in a bibliography a casual reader would skip"; the commentary said "the paper says so in its own second sentence on the term." A direct read of the publisher's PDF (cureus.com, sha256 f17aff07…, 2026-09-11) shows the term used uncited in the title and in the editorial's opening paragraph, with "Such a phenomenon has been described as 'artificial hallucination' [1]" arriving only in the summarizing passage after both ChatGPT experiments — the same sentence quoted, but not the first use of the term and not the second sentence about it. The note's central claim (Alkaissi & McFarlane cite Ji et al. rather than coining the term) is unaffected and independently reconfirmed; the claim about where in the paper they do so was wrong. Found by the scheduled cross-model audit, 00-meta/audits/audit-scheduled-2026-09-11-opus-1.md.