Strictly, no: under the traditional definition of plagiarism — presenting someone else’s words or ideas as your own without attribution — text generated by ChatGPT or another large language model does not have a human author whose work is being uncredited, so it does not fit the definition cleanly. But that technical answer is not the practical one. COPE, ICMJE, and essentially every major publisher have converged on a related but distinct position: presenting AI-generated text as your own original, unaided writing without disclosing that a tool produced it is treated as a serious violation of authorship and originality norms, and depending on the venue’s policy it can be handled as academic dishonesty, a breach of publication ethics, or research misconduct — even though it is not “plagiarism” in the textbook sense of stealing from a specific, identifiable human source.
This page addresses that specific normative question — is undisclosed AI-generated text plagiarism, and if not, what is it? — which is distinct from two related questions CASRAI covers elsewhere: whether AI can be listed as an author, and why AI-detection tools sometimes flag genuinely human-written text.
Why AI-generated text doesn’t fit the traditional definition of plagiarism
CASRAI’s plagiarism entry describes a spectrum running from verbatim copying without quotation or citation, through paraphrasing without attribution (“mosaic” plagiarism), to misappropriating unpublished ideas encountered in peer or grant review. Every point on that spectrum shares one feature: a specific, identifiable person’s original expression or idea is being passed off as the plagiarist’s own. Generative AI systems like ChatGPT don’t work that way. They predict statistically likely word sequences from patterns learned across a large training corpus; the output is not a copy of any single source text, and there is no individual human author being deprived of credit for that specific passage. Editors, integrity offices, and standards bodies that have addressed this directly have generally been careful not to stretch the word “plagiarism” to cover it for exactly this reason — it describes a different failure than the one plagiarism policies were written to catch.
This is also why AI-generated text is a poor fit for traditional text-matching plagiarism detection. Tools like Turnitin Similarity, iThenticate, and Crossref Similarity Check — covered in CASRAI’s Anti-Plagiarism Software guide — work by matching a submitted text against a database of existing published sources. AI-generated prose is usually novel word-for-word, so it typically returns a low or zero similarity score even when it was never disclosed. That’s precisely why a separate detection category, AI-writing likelihood scoring (see CASRAI’s detection tool (AI-generated) entry), emerged as a distinct product category from similarity checking, and why it produces its own, differently distributed set of problems — including a well-documented false-positive pattern on genuinely human-written text, covered in full in Why Does My Paper Say “AI Detected”?.
Why undisclosed AI text is treated as a research-integrity problem anyway
Not fitting the definition of plagiarism doesn’t mean undisclosed AI-generated text is treated as fine. COPE’s position statement on Authorship and AI tools (published February 2023) established the framework most publishers still use: AI tools cannot be listed as authors, because authorship requires the ability to take responsibility for the work, consent to publication, and declare conflicts of interest, none of which a non-legal entity can do — but any generative AI used in producing the manuscript, its images or graphical elements, or in collecting or analyzing data must be disclosed, typically in the methods section or a dedicated statement. The ICMJE’s May 2023 recommendations update reached a compatible position and became, alongside COPE’s, the single most-cited reference point for journal AI policy; most major publishers adopted broadly consistent policies within months of both.
The integrity problem these bodies are actually naming is authenticity and transparency, not theft from a specific source. Presenting AI-generated text as though it reflects your own unaided drafting, reasoning, or expression — when it doesn’t — misrepresents how the work was actually produced, in much the same family of concern as undisclosed ghostwriting or an undisclosed editing service. That is why editors and institutions treat undisclosed AI use as an authorship-integrity or research-misconduct issue under their own policies even while declining to formally label it “plagiarism.” Some commentators and a growing body of literature on the topic have used the informal term “AIgiarism” to describe this practice — presenting AI output as one’s own original work — but this is descriptive shorthand used in commentary and emerging scholarship, not a formal category recognized in COPE’s, ICMJE’s, or ORI’s own definitions, and CASRAI does not treat it as an established term of art.
Disclosed vs. undisclosed use: what actually changes
Whether AI-generated text creates a problem at all turns almost entirely on disclosure, not on the mere fact that a tool was used:
- Disclosed, policy-compliant use. Most current journal and institutional policies permit generative AI for tasks like language polishing, translation assistance, brainstorming, or drafting support, provided the use is disclosed — typically in a dedicated statement or methods section, per CASRAI’s generative-AI disclosure statement and AI tool disclosure entries — and provided a responsible human author has verified the accuracy of anything the tool produced. Disclosed use, handled this way, is not treated as misconduct by any major publisher policy CASRAI has reviewed.
- Undisclosed use presented as entirely one’s own writing. This is the practice that draws scrutiny, regardless of whether the underlying text happens to be original (in the copying sense) or not. It misrepresents the authorship and verification process to editors and readers, and institutions vary in exactly how they classify and sanction it — as an authorship violation, a breach of a specific AI-use policy, or, where an institution’s own definition is broad enough, as a form of academic dishonesty or research misconduct.
For the closely related question of how a human contributor should record AI assistance within their own CRediT role attribution — as distinct from a standalone disclosure statement — see CASRAI’s AI Contribution Disclosure & CRediT Statements guide.
How this differs from two related questions
Three separate AI-and-integrity questions get conflated often enough that it’s worth distinguishing them explicitly:
- Is undisclosed AI-generated text plagiarism? (this page) — No, not under the traditional definition, but it is treated as an authenticity/disclosure violation by essentially every major publisher and editorial body position CASRAI has reviewed.
- Can AI itself be listed as an author? No, under every major publisher and editorial-body position checked for CASRAI’s AI as Author entry — AI systems cannot take responsibility for a work, consent to publication, or hold conflicts of interest.
- Why did an AI detector flag my (human-written) paper? A separate technical problem covered in Why Does My Paper Say “AI Detected”? — AI-writing detectors estimate machine-generation likelihood from statistical patterns and have a documented false-positive rate, particularly for formulaic academic writing and non-native English writers. A high AI-detection score is not, by itself, proof that undisclosed AI text was used.
Practical guidance
- Check the specific policy that applies to you. Journal, publisher, and institutional AI policies vary in what they permit and how they require disclosure. CASRAI’s guide to journal and publisher AI policy covers how editorial offices currently handle this; the specific policy of your target journal or institution is the authoritative source, not general guidance.
- Disclose generative AI use, even when it’s permitted. Disclosure is what separates acceptable use from a policy violation in nearly every framework reviewed here — it is rarely the mere fact of AI assistance that creates a problem.
- Verify everything an AI tool produces. Both COPE and ICMJE hold the human author fully responsible for the accuracy and integrity of the final text, regardless of what tool helped produce a draft — including checking for fabricated or fake citations, a well-documented failure mode of AI-generated text.
- Don’t confuse a similarity score with an AI-writing score. They measure different things and require different responses; see CASRAI’s Anti-Plagiarism Software guide for how the two compare.
- If your institution investigates undisclosed AI use, expect it to run through existing misconduct or integrity channels rather than a plagiarism-specific one. See CASRAI’s How a Research Misconduct Investigation Works guide and the research misconduct entry for how institutional processes generally proceed.
Frequently asked questions
Is using ChatGPT to write a paper considered plagiarism?
Not under the traditional definition, which requires a specific human source being uncredited. But presenting ChatGPT-generated text as your own unaided writing without disclosure is treated as an authorship-integrity and disclosure violation by COPE, ICMJE, and most major publishers, and institutions may classify it as academic dishonesty or misconduct under their own broader policies even though it isn’t textbook plagiarism.
Can I get in trouble for using AI-generated text even if I don’t copy anyone specifically?
Yes. The issue publishers and institutions act on is usually non-disclosure and misrepresentation of authorship, not copying from an identifiable source. Using AI-generated text without disclosing it, where disclosure is required, can trigger consequences even though no single person’s work was copied.
Does disclosing AI use protect me from an integrity finding?
Disclosure is generally what separates permitted from problematic use across the policies CASRAI has reviewed, but it doesn’t excuse leaving factual errors, fabricated citations, or unverified claims in a final text — the human author remains fully responsible for accuracy regardless of disclosure.
What’s the difference between AI plagiarism and an AI-detected false positive?
They’re opposite problems. Undisclosed AI-generated text presented as your own work is a real disclosure/authorship issue if it occurred. An AI-detected false positive is a detection tool incorrectly flagging genuinely human-written text as machine-generated — the writing wasn’t AI-generated at all. See Why Does My Paper Say “AI Detected”? for the false-positive case specifically.
Is there an official term for passing off AI text as your own writing?
Not a formally standardized one. Some commentary and emerging literature use the informal term “AIgiarism,” but COPE, ICMJE, and ORI address the practice through their existing authorship-disclosure and misconduct frameworks rather than a newly named category.
