There are two separate questions about generative AI and your dissertation, and students routinely collapse them into one. The first is whether you are allowed to use it, which your university’s assessment rules answer. The second is how to reference the output if you do, which your referencing style answers. This guide deals with the second. Answer the first one, from your own academic integrity policy, before you read any further — a perfectly formatted reference to a use that was not permitted does not help you.
The short version: in Harvard you reference an AI tool much like a source with a corporate author, adding a label that marks it as AI-generated and, where the tool gives you one, a shareable link to the specific output. The details differ by which Harvard your department uses, and that variation is the part of this that actually costs students marks.
First, the thing that makes this harder than it should be
Harvard is not one style. It is a family of author–date conventions with no governing body, and UK universities resolve the AI question differently because they resolved everything else differently first. Some departments follow Cite Them Right, whose 13th edition added guidance on AI-generated material and on working with AI tools as part of an assignment. Others maintain a named house variant with its own AI template.
So the first step is not formatting. It is finding out which guide your markers use, and whether that guide has published an AI example yet. If it has not, your department will usually tell you to follow the acknowledgement route described below rather than invent a reference format.
The Harvard reference, with a real published template
Leeds Harvard — one of the named UK variants — publishes an explicit generative AI template and example. The University of Leeds Library gives the pattern as:
Template: Company. Year. Name of Generative AI (Version number if known). [Generative AI]. [Date accessed]. Available from: URL of specific output if available.
Example: Microsoft. 2024. Copilot. [Generative AI]. [Accessed 12 March 2024].
Source: University of Leeds Library, Leeds Harvard referencing examples, “Generative artificial intelligence”.
Read the shape rather than memorising it. Five elements do the work, and every Harvard variant’s AI reference is built from some arrangement of them:
| Element | What goes there | Worked value |
|---|---|---|
| Author | The company that made the tool, not the tool | OpenAI, Microsoft, Google, Anthropic |
| Year | The year of the version you used | 2026 |
| Title | The tool’s name, italicised, with the version where known | ChatGPT (version as shown in the tool) |
| Descriptor | A bracketed label marking it as AI output | [Generative AI] |
| Access | Date accessed, plus a share link to the specific output if the tool provides one | [Accessed 4 September 2026] |
Two rules travel with the format and are more important than the punctuation. Leeds states that if you directly copy and paste words from the AI output, you must put them in quotation marks to show they are a direct quote. And where the tool provides a shareable link to the conversation or output, you should include it — because unlike a book, an AI response is not retrievable by anyone else without that link.
In-text citation
The in-text citation follows the ordinary author–date pattern for whichever Harvard you are on, treating the company as the author: (Microsoft, 2024), or Microsoft (2024) produced…. If you are quoting the output directly, quotation marks go round the quoted words exactly as they would round any other direct quotation.
The acknowledgement, which is a separate requirement
This is the part most students miss entirely, because it is not referencing at all. Leeds tells students that if you have used generative AI to assist you in creating your work — even where none of its output appears in the text — you should include an acknowledgement stating which AI tool you used and how you used it, giving idea generation, summarising notes and generating an image as examples.
The distinction is worth being precise about, because the two obligations catch different things:
- Cite and reference when AI output appears in your work — a quoted sentence, a generated image, a summary you reproduce.
- Acknowledge when AI helped you produce work that is otherwise entirely yours — brainstorming, reorganising your own notes, explaining a concept you then wrote up in your own words.
An acknowledgement usually sits with the front matter or in a short declaration, and a single honest sentence does the job: “I used [tool] to generate initial topic ideas in October 2026. All analysis, writing and referencing in this dissertation is my own.”
Why the references AI produces are a separate problem
Formatting a reference to the tool is straightforward. The real risk is the references the tool hands you. Cite Them Right’s own guidance on emerging AI tools notes that there have been instances where AI services produced references to non-existent journal articles, and that AI tools have produced factually incorrect information when prompted for academic-standard text. It concludes that these errors and the lack of traceability raise serious questions about the practicality of using AI output in an academic setting.
Leeds puts the consequence in the plainest terms available: if you use references that are not genuine, that is a breach of the University’s academic integrity rules, and you should not include any reference to information you have not seen and used yourself.
The practical rule that follows: every reference in your bibliography must be one you opened. A fabricated citation is not a formatting error that a marker corrects; it is an integrity question, and it is one of the easiest things in a dissertation to check. Our guide to writing a literature review with AI, honestly sets out where the line between reasonable and never-delegated assistance sits.
A note for psychology students
If your department uses APA rather than Harvard, the elements are broadly the same — company as author, tool and version as title, a bracketed descriptor, and a URL — but the punctuation and ordering differ, and APA has published its own specific guidance on citing generative AI. Take the template from your own department’s APA guide or from APA’s published guidance rather than adapting a Harvard example, because the two styles treat the version string and the URL differently.
Five mistakes that cost marks
- Referencing the tool but not acknowledging the assistance. These are two different obligations and most UK guidance now expects both.
- Naming the tool as the author. The author is the company. “ChatGPT (2026)” is the wrong shape; “OpenAI (2026)” is the right one.
- Copying output without quotation marks. Reproducing AI text unquoted is presenting someone else’s words as your own, whatever the source.
- Passing on AI-generated references unchecked. The fastest route to a misconduct meeting. Open every source.
- Using a template from another university. Harvard variants differ, and an AI reference formatted to a guide your markers do not use is simply wrong for your submission.
How to settle this in fifteen minutes
- Open your assessment brief and find the sentence about generative AI. If there is not one, email your supervisor and get the answer in writing.
- Open your library’s referencing guide and search it for “AI” or “generative”. Note whether an example exists.
- If an example exists, follow it exactly. If not, use the five elements above and stay consistent, then confirm with your supervisor.
- Write your acknowledgement sentence now, while you still remember what you used and when.
- Check every reference in your list resolves to something you actually opened.
The wider rules for building the list — which Harvard you are on, what the 13th edition changed and how the four common source types are formatted — are in our complete UK Harvard referencing guide. Law students should start from OSCOLA instead, which handles AI in footnotes rather than author–date.
If the bibliography is where your time is going, that is clerical work no marking criterion rewards. Tesify builds your bibliography from the sources your text actually cites, in your chosen style, so nothing is listed that you did not add. Everything stays 100% written by you, and it is free to start.
Frequently asked questions
How do you reference ChatGPT in Harvard?
Treat the company as the author and the tool as the title, with a bracketed descriptor and an access date: Company. Year. Tool name (version). [Generative AI]. [Accessed date]. Available from: URL of the specific output if the tool provides one. Follow your own library’s variant where it publishes one.
Is the author OpenAI or ChatGPT?
OpenAI. The company occupies the author position, exactly as an organisation does for a corporate report. The tool name goes in the title position.
Do I need to include the prompt I used?
Some guidance asks for the prompt, often in an appendix, and some does not. Check your own library guide. Where a shareable link to the output exists, including it is more useful than the prompt alone because it shows the whole exchange.
What if my library guide has no AI example?
Use the five standard elements, apply them consistently, and confirm with your supervisor. Consistency is the criterion every Harvard variant shares, and a documented decision is defensible in a way that silent improvisation is not.
Do I have to declare AI use if none of its output is in my dissertation?
Usually yes. UK guidance increasingly asks for an acknowledgement of assistance — idea generation, summarising, restructuring — separately from any citation of output. Check your assessment brief, which is the document that binds you.
Is using AI to help with my dissertation cheating?
It depends entirely on your university’s rules and on what you delegate. Formatting citations and checking grammar sit at one end; having a chapter written for you sits at the other. Our guides to the best AI tools for a dissertation and using AI on a literature review honestly work through the line.
Can I cite an AI-generated image?
Yes, and you should, using the same elements with the image described. Leeds points readers to its separate image guidance for the additional detail.
What happens if I cite a source the AI invented?
It is treated as an academic integrity matter rather than a formatting slip, because you have claimed to have used a source that does not exist. Our guide to being accused of academic misconduct explains the process if it arises.