Making the most of Generative AI

The University’s Code of Academic Conduct states that 'attempting to pass off work created by artificial intelligence as your own' is a type of cheating.

See the University's guidance on AI in learning and assessment to check what you are allowed to do, and do check the guidance again each year, as the guidelines may still be evolving.

For each assignment, you need to be clear about the guidance from your course or department on using GenAI. Check module documents or speak to your Academic Advisor or Course Leader to find out if the use of GenAI is permitted.

See the University’s guidance on Artificial Intelligence in Research.

Generative AI can produce information or media that is inaccurate, biased, or outdated. Find out more about these problems:

• Limited reasoning - GenAI tools use predictive statistics to create content (rather than using facts to drive logical reasoning and creative synthesis, or to constrain creativity). This can limit its capacity to generate new insights, but also means it may introduce inaccuracies and false information.
• False information – GenAI tools may unintentionally fabricate hallucinations (false information, including false quotations and citations or media), or intentionally be used by people to create deep fakes (convincing but misleading images or audio-visual recordings). Check for the reliability of output.
• Inaccurate inputs – Some GenAI systems search the entire internet, and may therefore include input data from human-generated myths as well as hallucinations and deep fakes from AI tools. If reliability is important, check that the AI tool is limited to only use academic sources for input data.
• Lack of currency - Even if the AI system is now searching recent material, because the language model has been trained on old data, old information and biases may be reproduced. Check for recency of the sources and perspectives.
• Training biases – Most AI systems have incorporated socio-political biases from older training data or the humans involved in the training process. Some AI tools have therefore provided output containing sexist, racist, or other information considered offensive today. Check AI output for biased content.
• Commercial or malicious distortions – Optimisation of AI results for commercial gain may lead to biased search results, whilst malicious actors may ‘poison’ AI results by publishing misleading content where they know it will be harvested by the AI tool! Be alert to the possibilities of these distortions.

Remember that biased input data produces biased output!
Evaluate all GenAI outputs for any possible biases and inaccuracies.

Generative AI is not a source of verifiable, genuine, original academic information!

Whereas a peer review process is used to assess an academic journal article or systematic review, and to consider whether conclusions are reliable and whether the methodology of the paper could be repeated by another researcher, output created by GenAI can rarely be replicated and have not been assessed for reliability.

Also think about transparency and replicability/reproducibility. If you are using an AI tool for research analysis, to what extent can you clarify how a GenAI tool has arrived at its conclusions? In academic publishing, it is possible to ask authors for clarification of their methods if there is doubt about their conclusions. This isn’t always possible with generative AI.

For both of these reasons, GenAI cannot be considered a reliable academic tool or source. You need to find reliable academic sources to cite as evidence or reliable tools to manipulate date.

Do not upload any personal and confidential data into public AI services. Only fully anonymised data can be shared with public AI systems. The ICO’s Code of Conduct on Anonymisation provides guidance on anonymisation techniques.

In 2025, Generative AI is estimated to use 60 times more energy than a conventional search engine (see JISC guidance). So could you undertake a simple search instead of asking GenAI to create something new? For example, to illustrate a report, could you search for existing images, and reference the sources accordingly?

If you use Generative AI to develop your work, you need to keep clear records about how you have used it, to safeguard your academic integrity. Keep records of the following, and either add them as appendices in submitted work, or simply store them in case you are asked for evidence later:

• Record the name of the AI tool, and the URL.
• Record the question prompts you entered in the AI system, with the date for each.
• Store a copy of the output created, and the date created.
• Briefly explain any changes you made to the GenAI outputs.

Remember that the wording of your question becomes input data for the GenAI process and will shape the integrity of the AI output. Even if you are only using AI to generate ideas, a question which uses outdated terms or biased wording will lead to outdated or biased output.

Because AI can generate inaccurate, biased, or outdated information or media, it is essential to question every aspect of its output, including the overall direction of its response, as well as the individual details:

• Consider whether the GenAI has interpreted the question or prompt in the most appropriate way.
• Fact-check all the information produced by GenAI, including verifying the source of all citations used to support its claims.
• If using generated media, also be alert to inaccuracies or unacceptable stereotypes in generated images or videos.
• See Is it important whether the GenAI tool produces reliable content? in the Planning Stage for further guidance about what to check.

How you reference AI-generated material depends on the source and how it is used in your work

To provide a comprehensive reference for an AI tool you have used, include the URL and date accessed.

(AI cannot be acknowledged as a co-author in academic work as it cannot be morally or legally responsible or accountable. Only you can take ownership over the use and outcomes of AI systems used in your academic work.)

If use of AI is approved for your work but you need to provide evidence of how you used it, then you may need to use one or more appendices for this.

Number each appendix so that you can refer to this in your work, e.g. (ChatGPT, 2024; see Appendix A for input prompts and outputs.)

For a research project for which you used AI, you may need to include in your methodology a description of how you used the AI. This helps to assure research integrity and assist the reproducibility or replicability of the research.