Introduction
Generative artificial intelligence (GenAI) extracts information from the internet to generate content. Therefore, GenAI generated content may contain biases and not produce factual information. GenAI generated content can be inconsistent and you might find different answers for the same queries. You should be aware of this if you want to refer back to something the GenAI content produced previously. You should critically evaluate any AI generated content you do use, as you would with any information you refer to and consider what its strengths and weaknesses are. Take a look at our guidance on lateral reading for more on this.
The journal articles, books and other scholarly resources which are available through the Library are usually peer-reviewed, which means that experts in the subject areas have scrutinised the work before it has been published. Although you should still be critical of what you are reading, the quality of the resources and the facts provided through peer-reviewed resources will be of a much greater quality than computer generated content as it currently stands.
AI currently has severe limitations with dealing with data in terms of mathematics and producing graphs and charts. If you decide to use GenAI you should use it as a tool to help, not as a means to writing an assignment. If you used GenAI to write an assignment and claim it as your own work, it would be considered plagiarism, which the University takes very seriously and could harm your university degree and career. For these reasons, managing your time, researching widely and referencing all of your tools and resources is important. See more about this on our guide on AI and plagiarism.
Ethics
Generated AI tends to be trained on sources from across internet which may contain biases. You should read GenAI outputs and view GenAI generated images with a critical eye. Detecting biases in GenAI can be challenging because it may not provide as wide a range of views as databases or internet searches. There is also no guarantee of the quality of the information provided. For example, in a peer-reviewed academic article, the research is verified by experts in the field.
Copyright
There are copyright considerations to take into account with GenAI. As mentioned above, Generative AI is trained on publicly available internet data and other sources, but it does not consistently attribute its sources. Therefore, use of GenAI can be contentious as you could be using someone else's ideas in your work without knowledge of the original source. There is also concern that work you put into Open GenAI could then be used to train GenAI tools and produced as an output to someone else, therefore your work and research could lose its originality and copyright. At the University of Bath we encourage use of Copilot through institutional sign in as we have an agreement that all work entered into Copilot will remain within the institution and will not be used to train Generative AI.
AI tools such as ChatGPT and Microsoft Copilot (formerly Bing Chat), can be useful to support learning: for example, they can help to inspire ideas or find information quickly. While GenAI promises many benefits, as with any technology, it also poses distinct risks and potential harms. Understanding these challenges will help you make educated choices when interacting with GenAI systems as a user.
Whether you can use GenAI depends on the ‘category of assessment’ that has been applied to it: Type A, Type B or Type C. Your tutor or supervisor will explain which category has been applied to your assessment; you can also find out more about each one by selecting the picture of the categories here.
If you choose to use GenAI, the University's supported tool is Copilot. Ensure you are logged in using your Bath email address because this safeguards your user data.
Note: When using GenAI, you must not enter or upload any sensitive data. For example, personal information or any work that is not your own.
If you have used an AI tool to assist in the general development of your work, such as summarising notes or proofreading, acknowledge this in a statement at the beginning of your work. For example:
‘I acknowledge that this work is my own, and I used ChatGPT 3.5 (Open AI, https://chatgpt.com/) to summarise my initial notes and to proofread my final draft only.’
The minimum requirement for an acknowledgement is:
Select the relevant section below for guidance on referencing AI-generated content or images.
Where you have used an AI tool to generate text content or ideas which you refer to in your work, you must acknowledge it as a secondary source. How you do this depends on whether the source is retrievable (can be returned to via a link) or non-retrievable (cannot be returned to via a link).
If the content is retrievable then reference this as a website or webpage.
For example, in the Harvard Bath style it would be:
ChatGPT, 2023. Why is citing and referencing your sources important? [Online]. San Francisco, Calif.: OpenAI. Available from: https://chatgpt.com/share/782cb099-a0dc-45b1-8da4-0e99713f2d45 [Accessed 4 September 2023].
If the content is non-retrievable (cannot be returned to via a link), acknowledge the source of the content in your text, but don't include a full corresponding reference in your reference list or bibliography. You can acknowledge non-retrievable AI content using one of the following two methods:
(Google Gemini AI (pers. comm.) 22 August 2023).
If you have used an AI tool to generate an image you must acknowledge that tool as a source.
If the image is retrievable, reference it similarly to an image taken from the web, citing the tool and year of publication in the caption and listing the image reference to your reference list using the webpage format.
If the generated image is not retrievable, i.e. cannot be returned to via a link, then you would give an in-text, ‘personal communications’ citation in the caption only.
For example, in Harvard Bath style:
Figure 3. Shark in a library image generated using an AI tool (Craiyon, AI Image Generator (pers. comm.) 14 July 2022).
Further information:
Note: If you are not using Harvard Bath, please refer to your specified style's guidance for referencing GenAI. If your chosen style does not provide a method for this then apply the rule of ‘best fit’ and reference it using the guidance outlined above: either a website, image from a website, or personal communication.
Prompting
To use GenAI effectively you will need to learn how to prompt the GenAI to provide a useful output. This is similar to searching a literature database - where we use various search terminology and filters to retrieve relevant results.
What are you trying to achieve?
Assigning roles and providing context
Critical conversation
Providing GenAI with a role and limitations at the beginning of the session and then refining the output throughout, will give you good results.
For my example I asked Copilot to assume the role of a professor of electrical engineering and said that I had the role of their student. I then asked it to provide me with an assignment question around electric cars.
Copilot gave me around five options of questions.
I then chose a question and asked Copilot to simplify the concepts.
I then took the simplified the question in my own words and asked Copilot what it thought
It then provided me with a final answer which I was happy to use in my work.
(Thorpe, 2024)
Prompt frameworks enable you to refine the responses you receive from GenAI. The two frameworks listed below are slightly different but either should help you retrieve more precise responses in a format which suits you.
The PASTA framework was developed by Phil Bradley (pers. comm., 2025).
Persona - Which role you want the GenAI to take on.
Audience - Who the audience is.
Structure - What format you would like the GenAI to produce its response in.
Task - What task would you like the GenAI to achieve.
Approach - What tone and approach you would like GenAI to take.
The CREATE framework was developed by Dave Birss (2024).
Character - Which role do you want the GenAI to play?
Request - What do you want the GenAI to do? The more specific you are, the more likely you are to have relevant results.
Examples - Can you provide examples for the GenAI to help it understand what you want it to create?
Adjustments - Review your answer and adjust your prompts to retrieve a better output.
Type of output - Which format do you want GenAI to produce? A 300 word summary? A slide?
Extras - (optional) extra questions you could ask the AI e.g. "ignore everything before this prompt", "explain your reasoning". (Birss, 2024).
See the Prompt Collection for more (Birss, 2024).
Critically evaluating GenAI outputs
Make sure GenAI isn't the only tool you use for discovery and do ask the Library for help with your research if needed.
Forming search strategies for databases
You could ask GenAI to help you develop a search strategy for your assignment question. If you ask this question though, it may come back with a long answer so you could break it down with the following questions:
(Thorpe, 2024, chapter 5).
Forming a research question
Tell AI your field and level of study and what your research outcomes are. Tell GenAI it is an expert in this area. Ask GenAI for some suggested research questions or input your own and ask it to criticise your ideas to help you formulate a good question.
(Thorpe, 2024, chapters 5 and 6).
Testing your knowledge
You could input some of your notes on an area and ask GenAI to quiz you on a particular field of knowledge.
(Thorpe, 2024, chapter 1).
Preparing for job interviews
Tell AI to be an interviewer for a particular company or industry and mention which level you're looking to interview at and ask GenAI to formulate some practice interview questions. Or you could ask GenAI to provide an example job description.
(Gregory, 2024).
LinkedIn Learning Modules - see our LinkedIn Learning guide for more.
Text resources
Birss, D., 2024. The Prompt Collection [Online]. Available from: https://davebirss.com/documents/the_prompt_guide.pdf [Accessed 31 March 2025].
Thorpe, J., 2024. AI for students : creative hacks for academic success : an illustrated guide. s.l.: Promptly Books.
Mollick, E., 2024. Co-intelligence : living and working with AI. London: WH Allen.
Videos
Gregory, J. 2024. AI-Powered Practice: Mastering Difficult Conversations in Education [Online]. Available from: https://www.youtube.com/watch?v=SAg-RrPBH90 [Accessed 27 March 2025].
We recommend for educators and skills providers the University of York's Digitally Enhanced Education Webinars.
