Six useful tests that ask different questions
Work can be allowed but misleading. It can be disclosed but false, original but harmful, or ethical but not protected by copyright. Apply each test on its own.
- Ethical
- The use respects people, their reasonable expectations, your duties, and the likely effects in this situation.
- Permitted
- The use follows the rule, instruction, contract, or law that applies to this work.
- Original
- The wording and ideas are not improperly taken from another source. In law, “original” can have a narrower meaning.
- Accurate
- The claims and what they suggest match reliable evidence. An honest process can still produce a false sentence.
- Disclosed
- The people who need to know are told what AI did. Telling them does not by itself give permission or excuse a problem.
- Copyrightable
- The work qualifies for copyright protection under the law that applies. This says nothing about its quality or ethics.
One tool, five different situations
Use these as starting points, not as legal categories. A private draft can turn into high-stakes work. A workplace task can also be covered by industry rules.
Personal / private
Usually gives you the most room to experiment
Is the material yours, and is it free of sensitive information? Could even a private chat with a tool expose someone else's words or data?
You usually do not need to disclose AI use to yourself. But another person's permission and your duty to keep information private still limit what you may upload.
Public / professional
What readers expect starts to matter
Did AI shape the reporting, evidence, quotations, argument, advice, or a personal account? What does the publication require?
Disclose important AI-generated work when a policy requires it or when staying silent would mislead people about how the work was made or whose experience it describes.
Academic / assessed
Follow the assessment instructions first
Do the rules clearly allow this use for this task? Must you declare the use, cite it, or describe your method in a set format?
Do not assume that disclosing banned help makes it allowed. If the rule is unclear, ask the instructor before using AI.
Client / employer
Permission, contracts, and data controls matter
Does the client or employer allow this tool and this use? Is the material approved for that system? Does a contract promise work created by a person?
Get approval when required. Use approved systems, have a person responsible for review, and tell the client or employer when they are entitled to know.
High-stakes / regulated
Use less AI and more expert oversight by default
Could the text affect health, safety, rights, money, freedom, eligibility, reputation, or access to essential services? Which rules apply in that field?
Do not let AI make a decision with serious effects. Use expert review and approved safeguards, or do not use AI.
Would the reader reasonably care?
“Materiality” asks whether the AI use would matter to the reader. Use that test for disclosure only after checking the rules and getting any needed consent. Follow these questions in order.
- 01
Rule
Does a current law, assessment instruction, contract, workplace policy, client brief, publication rule, or professional duty allow this use?
If not, do not use it. If the rule is unclear, ask the person or group that sets it. - 02
Authority and consent
Are you changing another person's draft, interview, private message, unpublished work, personal data, or recognizable voice? Are you allowed to do so? Does that person understand this tool, this use, and the meaningful risks, and have they agreed?
If not, remove the material or get permission before you upload it or imitate the voice. - 03
Materiality
Would a reasonable reader, evaluator, client, or collaborator care that AI had an important role in the claims, argument, reporting, evaluation, personal voice, or wording presented as someone's own?
If so, disclose it where they will see it and give enough detail to be useful. - 04
Consequence
Could a mistake, stereotype, privacy breach, or misleading impression affect someone's rights, safety, health, money, opportunity, reputation, or trust?
If so, add expert oversight, give AI a smaller role, or do not use it.
You ask AI to tidy a friend's personal statement. The reader may care because the statement is supposed to use your friend's own voice. The draft may also be private. Ask your friend before uploading it. Check the application's rules. Disclose the help if the rules require it or if the rewrite makes important changes to work presented as your friend's own.
Consent applies to what you put in, not only what comes out
Changing someone else's work
Permission to read, edit, publish, or store a document does not automatically include permission to send it to an outside AI provider. Before asking for consent, explain the tool, your purpose, the material you will use, and any meaningful risks.
Copying a recognizable voice
Asking AI to “write exactly like” a living writer, colleague, client, or private person can falsely suggest that they created or approved the work. It can also misuse your relationship with them, even if no sentence is copied. Instead, name features such as “short sentences and concrete detail.” Get permission if the work is meant to pass as that specific person's voice.
Keeping information private
Do not put confidential, legally privileged, sensitive, unpublished, or identifying material into a tool unless you are allowed to share it. You must also use an approved system whose terms, storage period, access rules, and data-use controls meet your duty to protect the material. Removing names helps only when the details left behind cannot identify or expose someone.
Bias and representation
Models can repeat stereotypes, erase differences within minority groups, and favor familiar cultural views. Check who the work describes, who it leaves out, and which labels it uses. Ask whether affected people or experts should help shape it.
UNESCO's Recommendation links AI use to human dignity, personal choice, privacy, diversity, freedom from discrimination, openness suited to the situation, and clear human responsibility. NIST lists privacy failures, harmful bias, invented information, damaged information quality, and poor human-AI setup among the risks of generative AI. Neither document gives one rule for every kind of writing.
Similar text, plagiarism, copyright, authorship, and credit
| Concept | The question it asks | What it does not settle |
|---|---|---|
| Similarity | Does the wording or structure resemble material in the set being compared? | A match is not automatically plagiarism or copyright infringement. No match does not prove the work is original. |
| Plagiarism | Did someone present another person's words or ideas without giving the credit expected in this setting? | Plagiarism is about honesty and proper credit. It is not another name for copyright infringement. |
| Copyright | In the country whose law applies, is the expression protected? Who owns it? Does a license or legal exception allow its use, or does the use infringe copyright? | A legal use can still be dishonest, violate academic standards or consent, or be poor-quality work. |
| Authorship | Who contributed the ideas and wording, and who can take responsibility for the work? | A name on the byline does not prove that every passage is original, true, or protected by law. |
| Attribution | Does the work clearly name the source or contributor in the way this setting requires? | A citation does not give permission, prove a claim, or fix too much copying. |
Trace each fact and idea to the source that actually supports it. Then cite that source in the form your setting requires. Give honest credit for people's quotations, research, translation, editing, and other work. Naming the AI tool does not replace that credit.
To check claims against their sources, use the fact-checking lesson. To make clear choices that make the work your own, use the authorship and voice lesson.
Say what the tool did, not just that you used AI
A useful disclosure explains AI's important role. It also names the tool, date or version, review process, and responsible person when people need those details to repeat the work, check policy compliance, or decide whether to trust it.
Light assistance
AI helped with spelling and sentence-level editing. The author reviewed and approved every change.
Substantial drafting
The author used [tool] to suggest an outline and draft parts of the text. The author then checked the sources and thoroughly revised the final version.
Research publication
On [date], we used [tool/version] to [specific purpose]. The authors checked the output against the cited sources and remain responsible for the paper.
Work for another party
Prepared with the client's approval using generative AI for the structure and first draft. [Responsible role] reviewed the final claims and wording.
Keep some writing out of AI tools
- The rules ban it.Disclosing AI use cannot make an unapproved assessment, assignment, submission, or client deliverable acceptable.
- You cannot get meaningful consent.Do not upload or change someone's sensitive material, or copy their identity, just because the tool lets you.
- The input must stay confidential.Use an approved, controlled system only when your duty and the person responsible for the material allow it. Otherwise, keep the material out.
- The task requires your own learning, testimony, judgment, or relationship.An AI-generated apology, evaluation, personal account, or professional conclusion may defeat the purpose of the writing.
- No qualified person can review a high-stakes result.AI should not create health, legal, financial, safety, eligibility, or rights-related guidance on its own for other people to rely on.
- You cannot check the sources or defend the result.If you cannot trace a confident answer, stop. A disclaimer does not solve the problem.
Ethics goes beyond law, and laws differ by place
The U.S. Copyright Office's January 2025 report says that using generative AI does not prevent copyright protection for expression created by a person. But material made only by AI is not protected. Prompts alone generally do not give a person enough control over how the result is expressed. Protection depends on what the person contributed. This report does not settle plagiarism, contracts, privacy, publicity rights, or every other legal issue.
The EU AI Act's Article 50 transparency rules have applied since August 2, 2026. The 2026 Digital Omnibus, which delayed the Act's high-risk rules, did not postpone them, apart from a short grace period for the machine-readable marking duty on systems already on the market. The rules cover specific uses, including some AI-generated or altered text published to inform the public about matters of public interest. An exception applies when a person reviews the text or controls its editing and a person or legal entity, such as a company or organization, takes editorial responsibility. Article 50 does not require a label on every sentence written with AI help.
This page gives general educational information, not legal advice. Laws, contracts, organization rules, professional duties, and platform terms differ by country or region and change over time. Read the current rules and get advice from a qualified professional for a real legal or regulated decision.
Current through October 2026
The decision model and materiality test are Vibe Writing Hub's editorial synthesis. In other words, its editors created them by bringing the listed guidance together. Each linked policy applies only to its own field and may change. Together, they show why the situation matters more than one universal rule for AI writing.
- UNESCO: Recommendation on the Ethics of Artificial Intelligence
- NIST AI 600-1: Generative Artificial Intelligence Profile
- U.S. Copyright Office: Copyrightability report (January 2025)
- European Commission: AI Act framework and application timeline
- ICMJE: Use of Artificial Intelligence in Publishing
- Nature Portfolio: AI editorial policy
- University of Oxford: safe and responsible GenAI guidance (2026)
- UK NCSC: public LLM information-security risks
- Associated Press: standards around generative AI
- Authors Guild: AI Best Practices for Authors (May 2026)