How to Cite AI: Disclose Use in School and Work

Knowing how to cite AI has become a basic skill for students, researchers, writers and anyone who delivers work to a client. The rules are still settling, and they differ between schools, publishers and employers, but a few principles hold almost everywhere: be honest about what the assistant did, keep a record you can show, and never present a machine’s output as your own thinking or as a verified source.

This guide separates citing from disclosing, summarises how the main style guides handle generated text, gives you statements you can adapt and lists the mistakes that cause real trouble.

Cite, disclose or both?

These two actions answer different questions, and mixing them up leads to awkward citations.

Citing tells a reader where a specific piece of text, idea or data came from, so they can find it. It is about sources.
Disclosing tells a reader how a tool was used in producing the work: for brainstorming, editing, translation, coding or drafting. It is about process.

A generated answer is a poor source for facts. It is produced from patterns, it can be wrong, and the same prompt may give a different answer tomorrow, so a reader cannot reliably retrieve it. For a factual claim, find the original document, check it and cite that. Cite the assistant only when its own output is the subject, for example when you quote a response to show how it behaves, or when your style guide asks you to credit the tool for text you reproduce. In every other case, a disclosure statement is the more honest tool.

If you use an assistant to find sources, remember that models can invent references that look real. Our explainer on AI hallucinations covers why, and the rule is simple: never cite a reference you have not opened and read.

How to cite AI in the major styles

Style guides have published guidance on generative AI, and it keeps changing, so always check the current edition or your institution’s page. The table summarises the general approach of the main styles at the time of writing.

Style	Treatment of the tool	What to include	Where it appears
APA	Software output; the tool’s developer is the author	Developer, year, tool name and version, description, URL; describe the prompt in the text	In-text citation and reference list
MLA	Not an author; cite the tool as the source of generated content	Prompt in quotation marks, tool name, version, developer, date used, URL	Works cited entry
Chicago	Treated like personal communication, not a published source	Tool, prompt, date and developer	Footnote or endnote, usually not in the bibliography
Harvard (varies)	Depends on the institution	Often tool, developer, date, prompt and URL	Reference list, following local rules
Journals and publishers	Tools cannot be authors	A statement of what was used and for what	Methods, acknowledgments or a disclosure section

For the details, go to the primary sources: the APA Style guidance, the MLA Style Center page on citing generative AI and the Chicago Manual of Style Q&A. If your course or journal names its own format, that format overrides all of these.

Keep a log so you can prove what you did

A citation is only as good as the record behind it. Because assistants do not return identical answers, you cannot recreate a response from memory later. Keep a simple log from the first prompt.

Tool and version: the name of the service and, if shown, the model or version.
Date and time of use: answers and tools change, so the date matters.
The exact prompt: copy it, do not paraphrase.
The output: save the full response, or the part you used.
What you changed: a note on how you edited, checked or rejected it.
Sources you verified: links to the original documents you actually read.

Check how long your assistant keeps your chats. On Ask Mio the Free plan keeps a 30-day history and paid plans keep it without a time limit, and chats can be exported at any time, so export anything you need as evidence before it ages out.

Write a disclosure statement that fits the situation

A good statement is short, specific and true. It says which tool, for which task and what a human did afterwards. Avoid vague wording such as “AI was used in the preparation of this document”, which tells the reader nothing.

For a school or university assignment

“I used an AI assistant to brainstorm three possible outlines and to check my grammar in the final draft. I wrote all arguments myself, and I checked every source against the original text.”

For a report to a client or manager

“The first draft of sections 2 and 3 was produced with an AI assistant from my notes. I rewrote the text, verified all figures against the source data and take responsibility for the content.”

For a research paper or article

“An AI assistant was used to improve the language of the manuscript and to suggest code for the figures. It was not used to generate data, analysis or conclusions. The authors reviewed and edited all output and are responsible for the work.”

For published content aimed at the public

“Images in this article were generated with an AI image tool and edited by our team.” Short and plain is enough, and it supports the transparency expectations discussed below.

A worked example: one essay, three uses of AI

Suppose a student writes a 2,000-word essay on urban transport and uses an assistant three times. First, they ask for ten possible angles and pick one. Second, they paste their own paragraph about bus lanes and ask for a clearer version, then rewrite it again in their own words. Third, they ask for a list of studies on congestion pricing.

The record looks like this. The brainstorming and the clarity edit go into the disclosure statement, because they shaped the work but produced no source. The list of studies is treated differently: the student treats it only as a lead, searches for each title in a library database and finds that two of the six do not exist. They cite the four real ones from the original papers, with page numbers, and mention nothing of the others. The log holds the prompts, the dates and the original paragraph, so the student can show the process if the instructor asks.

The same logic works at work. A consultant who has an assistant tidy a draft discloses the editing to the client if the contract asks for it, and cites the real market report behind each number. The assistant is the tool; the report is the source. Keeping the two apart makes every statement you write shorter and every claim easier to defend.

Rules by context

Schools and universities

Policies vary from “no AI at all” to “any use, if disclosed”. Read your course rules first, and ask when they are unclear, because the safest approach is the one the instructor has approved in writing. Using a tool in a way that is not allowed can count as misconduct even if you disclose it afterwards. The background on academic integrity is useful context, and our guide to AI for students covers honest study habits.

Research and publishing

Many publishers say that an AI tool cannot be named as an author, because authors take responsibility and a tool cannot. They usually ask for a statement about what was used. Images and data generated by a tool raise extra questions, so read the journal’s instructions before you submit.

Work and clients

Contracts and company policies increasingly say whether AI may be used and whether the client must be told. If your company has no policy yet, our article on writing an AI use policy is a starting point. A short note to a client about how a deliverable was produced builds trust, and it protects you if questions come up later.

Public content and EU transparency rules

Some AI-generated or manipulated content must be labelled under the EU AI Act, for example realistic synthetic images, audio and video. Our explainers on the EU AI Act and on AI watermarks and content provenance describe what is expected. The exact obligations depend on the type of content and your role, so check the current text when you publish at scale.

Wording in the text itself

Beyond the reference list, say in the body how you used the tool whenever the reader could otherwise be misled. If you quote an answer, put it in quotation marks and name the prompt, for example: “When asked to summarise the main objections to congestion pricing, the assistant listed cost, fairness and enforcement.” If you only used the tool to polish wording, one line in a note is enough.

Avoid language that gives the tool human credit or blame, such as “the AI decided” or “the AI believes”. It produced text from a prompt, and you chose what to keep. Clear, plain phrasing helps readers, markers and editors judge the work fairly, and it keeps responsibility where it belongs: with you.

Mistakes that cause real trouble

Listing the assistant as an author. Most style guides and publishers reject this.
Citing fabricated references. Open every source before it goes in your bibliography.
Hiding the use. Concealment is what turns a permitted tool into a misconduct case.
Disclosing too vaguely. A reader should be able to tell which parts a person wrote.
Trusting detectors. A detector score is not proof either way; see how reliable AI content detectors are.
Relying on memory. Without a log, you cannot support a claim about what you did.

Using Ask Mio with citations in mind

If your work needs sources, use Research mode, which searches the web and cites its sources; it is included from the Chat plan at 5 euro a month. You then follow each link, read the original and cite that source, not the assistant. In Chat and Write mode on the free plan, you can still draft and edit, then add your disclosure statement. Mio picks a suitable model for each request, and it does not name which one, so record the date and the mode you used, along with the product name.

Chats can be exported or deleted by you at any time, and Ask Mio states that data is not used to train models. Our guide on researching with cited sources shows how to check each link.

Frequently Asked Questions

Do I always have to cite AI?

Not always, but you should disclose meaningful use wherever rules or honesty require it. Check your school, publisher or employer policy. Light editing such as spell-checking is often exempt, while generating text, images or analysis usually needs a statement. When in doubt, disclose briefly and specifically.

Can I cite an AI assistant as a source for facts?

You can, but you should not rely on it. Generated answers can be wrong, and readers cannot retrieve them. Find the original source, verify the fact and cite that. Cite the assistant only where a style guide asks for it, or when the output itself is what you discuss.

What should a disclosure statement include?

Name the tool, say what you used it for and say what you did yourself afterwards, such as checking sources and rewriting. Keep it to one or two sentences. Include the date of use if your style requires it, and avoid vague phrasing that hides which parts are yours.

Can an AI assistant be listed as an author?

Generally no. Most journals and publishers say that authorship requires accountability, which a tool cannot take. Authors remain responsible for everything in the work. The tool is usually acknowledged in a methods or disclosure section instead, with a statement of what it was used for.

Do AI detectors replace disclosure?

No. Detectors give probabilities, produce false positives and can be fooled, so neither a high nor a low score proves anything. Honest disclosure, plus a log of prompts and drafts, is stronger evidence of what you actually did than any detector result.

The Bottom Line

How to cite AI comes down to three habits: cite the real sources you verified, disclose specifically how a tool helped, and keep a log you can show on request. Rules differ by context, so check the current guidance from your school, publisher or employer before you submit. For research you can verify link by link, see Ask Mio’s plans and start on the free one.

How to Cite AI: Disclose AI Use in School and Work