Generative AI tools like ChatGPT, Claude, Copilot, Gemini, DALL-E, and Midjourney produce text, images, and code that you may want to cite in academic work. Because Harvard referencing has no single governing manual, there is no universal "official" format for AI-generated content — but a clear consensus is emerging across UK, Australian, and international universities. Most institutions treat AI output as a software-generated source, requiring you to name the tool, the version, the date, and the prompt you used.
The Current Landscape
Harvard referencing is a broad convention, not a single publication manual. This means there is no equivalent of APA's official guidance on citing AI — instead, each university publishes its own rules. However, most institutional guides have converged on a similar approach:
- Name the AI tool and its developer as the "author"
- Specify the version you used (GPT-4o, Claude 3.5, Gemini 1.5, etc.)
- Include the date you generated the output
- Provide or describe the prompt — either in your text or in an appendix
The formats below follow the conventions most commonly adopted by UK and Australian universities. Always cross-check with your institution's specific guidance.
Reference Format for AI-Generated Text
When you use text generated by an AI tool — a paragraph drafted by ChatGPT, a summary created by Claude, a code snippet from Copilot — here's how to cite it:
Reference List Entry
OpenAI (2025) ChatGPT (GPT-4o) [Large language model]. Available at: https://chat.openai.com (Accessed: 15 January 2025).
Breaking it down:
| Element | What to Include |
|---|---|
| Author | The company that developed the tool (OpenAI, Anthropic, Google, Microsoft) |
| Year | The year you generated the output |
| Title | The tool's name and version in parentheses |
| Format descriptor | [Large language model], [AI chatbot], or similar — in square brackets |
| URL | The tool's main URL |
| Access date | The date you generated the specific output |
Variations by Tool
Anthropic (2025) Claude (Claude 3.5 Sonnet) [Large language model]. Available at: https://claude.ai (Accessed: 20 February 2025).
Google (2025) Gemini (Gemini 1.5 Pro) [Large language model]. Available at: https://gemini.google.com (Accessed: 8 March 2025).
Microsoft (2025) Copilot (GPT-4 Turbo) [AI assistant]. Available at: https://copilot.microsoft.com (Accessed: 12 January 2025).
In-Text Citations
Parenthetical:(OpenAI, 2025)
Narrative:OpenAI's ChatGPT (2025) generated the following summary…
When quoting AI output directly, include the prompt as context and use quotation marks:
When prompted to 'summarise the main arguments for universal basic income in 100 words,' ChatGPT produced the following response: 'Universal basic income (UBI) is most commonly defended on three grounds…' (OpenAI, 2025).
Reference Format for AI-Generated Images
AI image generators like DALL-E, Midjourney, and Stable Diffusion require a slightly different format because the prompt functions as the "title" of the work:
Reference List Entry
OpenAI (2025) 'A watercolour painting of a library at sunset' [AI-generated image]. Created using DALL-E 3. Available at: https://openai.com/dall-e-3 (Accessed: 10 February 2025).
Midjourney (2025) 'An isometric view of a university campus in spring' [AI-generated image]. Created using Midjourney v6. Available at: https://www.midjourney.com (Accessed: 22 January 2025).
In-Text Citations
Parenthetical:(OpenAI, 2025)
Narrative:The image was generated using DALL-E 3 (OpenAI, 2025) with the prompt 'A watercolour painting of a library at sunset.'
When used as a figure:Label the figure normally (Figure 1, Figure 2, etc.) and include the citation in the figure caption:
Figure 1.AI-generated watercolour of a library at sunset (OpenAI, 2025). Created using DALL-E 3 with the prompt: 'A watercolour painting of a library at sunset.'
Including the Prompt
Transparency about the prompt you used is a core requirement in almost every university's AI citation policy. There are three common approaches:
1. In the Text
Describe or quote the prompt directly in your sentence:
When asked to 'explain the water cycle for a Year 8 science student,' ChatGPT produced a three-paragraph explanation (OpenAI, 2025).
2. In an Appendix
For longer prompts or multi-turn conversations, include the full exchange in an appendix and reference it in the text:
The full prompt and ChatGPT's response are reproduced in Appendix A (OpenAI, 2025).
Your appendix should include:
- The tool name and version
- The date you generated the output
- The exact prompt(s) you entered
- The AI's complete response(s)
3. In a Footnote
Some institutions accept footnotes for prompt details. This is less common in Harvard style (which avoids footnotes), but check your guide:
The summary was generated by ChatGPT.¹
¹ Prompt: 'Summarise the three main theories of motivation in organisational psychology.' ChatGPT (GPT-4o), 15 January 2025.
Most Harvard guides prefer options 1 or 2 — footnotes are more associated with OSCOLA and Chicago styles.
When AI Is a Tool vs. a Source
Not all AI use requires a formal citation. The distinction is:
AI as a Source — Cite It
You must cite the AI tool when you include its output in your paper:
- A paragraph generated by ChatGPT that you quote or paraphrase
- A code snippet produced by Copilot that you embed in your work
- An image generated by DALL-E that you include as a figure
- Data or analysis produced by an AI tool
AI as a Tool — Acknowledge It
When you use AI to assist your process but don't include its output directly, a methodology note or acknowledgement is usually sufficient:
AI tools were used during the preparation of this essay. ChatGPT (OpenAI, GPT-4o) was used to brainstorm initial ideas and check grammar. All final content was written by the author.
This kind of acknowledgement might go in:
- A methodology section
- An author's note or declaration
- A footnote (if permitted by your institution)
Bottom line:If AI-generated content appears in your paper, cite it. If AI helped you think or edit but its words don't appear, acknowledge it transparently.
Institutional Variation — Always Check Your University's Policy
This is the most important point on this page. Universities vary widely in three areas:
1. Whether AI Use Is Allowed at All
Some institutions prohibit AI-generated content in assessments entirely. Others allow it with conditions. A correct citation does not override a prohibition — if your department says "no AI," then no amount of proper referencing makes it acceptable.
2. The Specific Citation Format
While the formats on this page reflect the emerging consensus, your institution may specify a slightly different layout — for example:
- Some require the prompt in the reference list entry itself
- Some treat the AI tool name as the title rather than the developer as the author
- Some require you to deposit a transcript of the AI conversation with your submission
3. What Counts as "Use"
Does using AI to check your grammar count as use? What about asking it to suggest synonyms? Institutions define the threshold differently. Read your policy carefully and ask your tutor if you're unsure.
AI-Generated Content and Academic Integrity
A few important principles:
- AI can fabricate information. Large language models generate plausible text — not necessarily accurate text. Never cite AI-generated claims as fact without verifying them against primary sources.
- AI output is not peer-reviewed. Treat it as you would a personal communication or informal source — useful for ideas, but not a substitute for scholarly evidence.
- Transparency is non-negotiable. Even if your institution allows AI use, concealing it is an academic integrity violation. Always disclose AI involvement.
- Your analysis must be your own. AI can help you gather or draft, but your argument, evaluation, and critical thinking must be genuinely yours.
Example Reference List Entries
Here's a reference list showing several AI tools cited together, as they might appear at the end of a paper:
Anthropic (2025) Claude (Claude 3.5 Sonnet) [Large language model]. Available at: https://claude.ai (Accessed: 20 February 2025).
Google (2025) Gemini (Gemini 1.5 Pro) [Large language model]. Available at: https://gemini.google.com (Accessed: 8 March 2025).
Midjourney (2025) 'An isometric view of a university campus in spring' [AI-generated image]. Created using Midjourney v6. Available at: https://www.midjourney.com (Accessed: 22 January 2025).
OpenAI (2025) ChatGPT (GPT-4o) [Large language model]. Available at: https://chat.openai.com (Accessed: 15 January 2025).
OpenAI (2025) 'A watercolour painting of a library at sunset' [AI-generated image]. Created using DALL-E 3. Available at: https://openai.com/dall-e-3 (Accessed: 10 February 2025).
If you cite both ChatGPT and DALL-E from OpenAI in the same year, distinguish them with lowercase letters:
OpenAI (2025a) ChatGPT (GPT-4o) [Large language model]. Available at: https://chat.openai.com (Accessed: 15 January 2025).
OpenAI (2025b) 'A watercolour painting of a library at sunset' [AI-generated image]. Created using DALL-E 3. Available at: https://openai.com/dall-e-3 (Accessed: 10 February 2025).
Summary: AI Citation Formats at a Glance
| Source Type | Reference List Format |
|---|---|
| AI text (ChatGPT, Claude, etc.) | Developer (Year) Tool Name (Version) [Large language model]. Available at: URL (Accessed: Date). |
| AI image (DALL-E, Midjourney) | Developer (Year) 'Prompt text' [AI-generated image]. Created using Tool Version. Available at: URL (Accessed: Date). |
| AI code (Copilot, etc.) | Developer (Year) Tool Name (Version) [AI assistant]. Available at: URL (Accessed: Date). |
What's Next?
Now that you know how to cite AI-generated content, explore these related guides:
- Harvard Reference List — the complete guide to formatting your reference list
- How to Reference a Website in Harvard Style — closely related formatting conventions for online sources
- Harvard Referencing Style — the full overview of Harvard referencing, including institutional variation
