Quick answer
Plagiarism, copyright infringement, ownership and Google's rules are 4 separate tests. OpenAI's terms say you own the output as between you and OpenAI, but also that output "may not be unique". The US Copyright Office says AI output can be protected only where a human author determined sufficient expressive elements. Check before you publish. Not legal advice.
Key takeaways
- 4 tests apply: plagiarism rules, copyright infringement, ownership and Google's rules. Plagiarism is judged under the rules of whoever you answer to (Northwestern University's policy is one school's rule). Copyright infringement is a legal question. A text can pass one test and fail another.
- OpenAI's terms (effective January 1, 2026) say that, as between you and OpenAI, you own the output, that output may not be unique, and that the assignment does not extend to other users' output or any Third Party Output.
- The US Copyright Office says generative AI output can be protected by copyright only where a human author has determined sufficient expressive elements. The release says this can include human-authored work in the output or creative arrangements or modifications, but not the mere provision of prompts.
- OpenAI withdrew its own AI classifier as of July 20, 2023, citing its low rate of accuracy. This page gives no accuracy figure for any current detector. Add a plagiarism check, your own facts and a fact-check.
- This site's checklist below is a suggestion, not legal advice. Ask a lawyer about anything beyond ordinary website copy.
Is ChatGPT plagiarized, and can I use its text on my website?
Plagiarism is judged under the rules of whoever you answer to (Northwestern's policy is one school's rule), and 4 separate tests apply: plagiarism rules, copyright infringement, ownership and Google's rules. OpenAI's terms also say output "may not be unique".
OpenAI's terms say that, as between you and OpenAI and to the extent the law permits, you own the output, and the checklist at the end of this page is what this site suggests you check before publishing. This page reports what each named source states, not what a court would decide.
What are the 4 tests: plagiarism, infringement, ownership and Google's rules?
Each of the 4 tests asks a different question, and passing one does not pass the others. The table gives each source's wording as read on October 7, 2026.
| Question | Who decides | What the source says | Applies to |
|---|---|---|---|
| Is it plagiarism? | The school, employer or publisher whose rules apply | Northwestern's policy lists "Submitting material that in part or whole is not entirely one's own work without attributing those same portions to their correct source" and says plagiarism includes "the unauthorized use of generative artificial intelligence" to create content submitted as one's own | Northwestern students; an example of one institution's rule |
| Is it copyright infringement? | US law, ultimately a court | 17 U.S.C. 501 says "Anyone who violates any of the exclusive rights of the copyright owner as provided by sections 106 through 122" (and other acts the section lists) "is an infringer of the copyright" | Anyone, in the US |
| Who owns the output? | OpenAI's contract, and separately copyright law | OpenAI's terms: "you (a) retain your ownership rights in Input and (b) own the Output" as between you and OpenAI; the Copyright Office: protection "only where a human author has determined sufficient expressive elements" | You and OpenAI (contract); US copyright (Office view) |
| Will Google treat it as spam? | Google Search's spam policies | Using generative AI tools to "generate many pages without adding value for users may violate" the spam policy on "scaled content abuse" | Sites that want to appear in Google Search |
Can ChatGPT repeat text it learned from?
Researchers (Nasr et al., arXiv 2311.17035, November 2023) reported an adversarial attack that extracted training data from ChatGPT; the abstract gives a relative figure for the attack, a rate 150x higher than when the model is behaving properly, and no absolute rate for ordinary use.
A paper first posted to arXiv on 28 Nov 2023 (Nasr et al.) reports that an adversary can extract "gigabytes of training data" from language models, ChatGPT among them. The abstract says its divergence attack makes the model "emit training data at a rate 150x higher than when behaving properly"; that compares the attack with the model behaving properly, as the abstract words it, and nothing more. It also says current alignment techniques "do not eliminate memorization." The attack was tested in 2023; the abstract does not say how often ordinary prompts return training data, so this page gives no absolute rate and makes no claim about current versions.
Who owns what ChatGPT writes for you?
As between you and OpenAI, the terms say you do, with conditions.
These are the Terms of Use effective January 1, 2026, for ChatGPT, DALL·E and OpenAI's other services for individuals. The sentence reads: "As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output." The terms also say OpenAI assigns "all our right, title, and interest, if any, in and to Output".
The words "if any" are in the sentence itself. The next paragraph, headed "Similarity of content", says "Our assignment above does not extend to other users’ output or any Third Party Output."
The terms also say "You are responsible for Content, including ensuring that it does not violate any applicable law or these Terms." There, "Content" covers your input and the output.
Can you copyright text that AI helped write?
According to the US Copyright Office, only where a human author has determined sufficient expressive elements. Its January 29, 2025 release on Part 2 of its AI report says it "concludes that the outputs of generative AI can be protected by copyright only where a human author has determined sufficient expressive elements." The release adds: "This can include situations where a human-authored work is perceptible in an AI output, or a human makes creative arrangements or modifications of the output, but not the mere provision of prompts."
The same release says that using AI "to assist in the process of creation or the inclusion of AI-generated material in a larger human-generated work does not bar copyrightability." That is the Office's analysis, not a ruling on your page. A pre-publication version of its separate Part 3, on training, was released on May 9, 2025; we draw nothing from it.
This is not legal advice. For contracts, logos, slogans or anything you plan to enforce against a copier, ask a lawyer.
Can an AI detector tell that ChatGPT wrote a text?
This page cannot say. OpenAI's own classifier, by its figures, labelled 26% of AI-written text "likely AI-written" and incorrectly labelled 9% of human-written text on its challenge set of English texts, and OpenAI says that as of July 20, 2023 it "is no longer available due to its low rate of accuracy"; this page has no figure for tools sold today.
OpenAI's page states that "As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy." Before that, it reported that on a "challenge set" of English texts the classifier correctly identified 26% of AI-written text as "likely AI-written" while labeling human-written text as AI-written 9% of the time. Those figures describe one withdrawn tool on one test set in early 2023.
What does Google say about publishing AI-generated pages?
Google's guidance singles out many pages without added value and unchecked facts. The page says "using generative AI tools or other similar tools to generate many pages without adding value for users may violate Google's spam policy on scaled content abuse." The spam policy defines that as "many pages are generated for the primary purpose of manipulating search rankings and not helping users."
For anything you publish, the guidance says: "It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing." It also says that sharing how content was created "can help give your readers more context." The guidance page shows a last-updated date of 2026-10-01; the spam policies page shows 2026-08-28.
What should you check before publishing AI-assisted text?
Run through this list; it is this site's suggestion, built from the sources above, not a legal or regulatory requirement.
- Find the rules that apply to you. A client contract, a publisher, a school or an industry body may have its own attribution rules. The plagiarism test is theirs, not ours.
- Fact-check every claim. Google's guidance calls this critical.
- Run a text-matching check. Use a check that compares your draft with published text, not an AI-authorship detector. We did not test any tool, so this page names none and claims no accuracy.
- Add what only you know, and keep a record of what you wrote and changed. Add your own writing and edits, and save the prompt, raw output and your edits to show what a person changed. The Copyright Office ties protection "only where a human author has determined sufficient expressive elements"; this site's suggestion is a record of your own contribution, and we do not say which elements qualify.
- Do not mass-produce pages. Google's guidance names many pages without added value for users.
- Say how it was made when it helps the reader. Google says it can give readers more context.
- Ask a lawyer about anything you may need to enforce.
For posts, see our AI social media content generator guide; for where review steps fit, the AI automation guide.
How was this page made?
Sources were opened on October 7, 2026: OpenAI's terms and classifier page, the US Copyright Office's release and AI page, Cornell LII's text of 17 U.S.C. 501, Northwestern University's academic integrity principles, two Google Search Central pages and one arXiv abstract. Terms and agency positions can change, and a university policy is one school's rule, not a standard. This page was drafted with Claude, an AI model made by Anthropic, and checked against each vendor's published material on October 7, 2026.
FAQ
Can ChatGPT output text it was trained on?
Researchers (Nasr et al., arXiv 2311.17035, November 2023) reported an adversarial attack that extracted training data from ChatGPT. The abstract gives a relative figure for the attack (a rate 150x higher than when the model is behaving properly) and no absolute rate for ordinary use, and the sources read here do not measure how often ordinary prompts return copied text. Check the text before publishing.
Do I own the text ChatGPT writes for me?
OpenAI's terms say that, as between you and OpenAI, and to the extent permitted by applicable law, you own the Output. That is a contract between you and OpenAI. Whether copyright law protects the text is a separate question, and the US Copyright Office ties protection to a human author's sufficient expressive elements. This is not legal advice.
Is it plagiarism to publish ChatGPT text without saying so?
It depends on whose rules apply. The one policy we read, Northwestern University's, covers students and lists unauthorized use of generative AI to create content submitted as one's own as plagiarism. We did not find a comparable rule for business websites, so check client contracts, platform terms and industry rules.
What does Google say about AI-written pages?
Google's guidance says using generative AI tools to generate many pages without adding value for users may violate its spam policy on scaled content abuse. The same page says to manually fact-check and review all AI-generated content before publishing.
Can an AI detector prove a text was written by ChatGPT?
This page cannot say. As of July 20, 2023, OpenAI's own AI classifier was no longer available, which OpenAI attributed to its low rate of accuracy. We found no tool maker's page we could cite for a detector accuracy figure, so this page gives none.