Check your department's referencing guide first — most UK universities added AI wording between 2024 and 2026, and that guidance overrides any general rule you find online.
Where there's no local rule, APA's approach treats a generative model as software: company as author, model name and version. Many UK courses also want the prompts you used in an appendix.
Why local guidance comes first here
Referencing conventions usually change slowly. This one changed fast, and institutions moved at different speeds and in different directions. Some added a citation format; some added a mandatory declaration form instead; some prohibit AI use in ways that make the citation question moot. A general answer from the internet can be simultaneously correct and useless for your module.
A reference isn't a licence. If your course prohibits using AI to generate text, citing the generated text is still a breach — you've just documented it. Settle whether the use is allowed before worrying about the format.
The shape of an AI citation
Where a format is required, the elements are consistent across styles even if the punctuation isn't:
- Author — the organisation that makes the model, not the model itself.
- Year — of the version you used.
- Model name and version — treated as the title, usually italicised, with the version identifier included because output isn't reproducible across versions.
- Descriptor — a bracketed note such as [Large language model].
- URL — of the tool.
In-text it behaves like any author-date citation. Apply your own guide's punctuation, not the shape above, wherever the two disagree.
The appendix, which is the part people miss
A generative conversation isn't a retrievable source. Nobody can look up what you saw — the same prompt produces different output tomorrow, and on a different version it produces something else again. That breaks the entire purpose of a citation, which is letting a reader check your source.
Which is why a growing number of UK courses ask for the prompts and relevant output in an appendix, and treat that as the actual evidence. If your department asks for this, it isn't bureaucracy — it's the only thing that makes the reference verifiable. Keep the transcript as you go rather than reconstructing it afterwards.
Declaration versus citation
These are separate requirements and courses increasingly want both. A citation credits specific content you used. A declaration states which tools you used and how, usually on a cover sheet or submission form, covering uses that never appear in the text at all — planning, proofreading, explanation. Undeclared but permitted use is itself a breach on many courses, which catches people out. See do I need to cite AI if I only used it to brainstorm?
One caution worth repeating: never cite a source a model gave you without finding the real record. Models fabricate citations that look entirely convincing.
SafeGrade checks Harvard and APA formatting across your list, and Sage can tell you what your specific course requires. Private, never stored, never used for training.
Run one free scan →No card. No catch. First scan free.