Using AI can be very helpful, but it must be done correctly. Otherwise, one might plagiarise, reinforce bias in other humans/databases/algorithms, make an error with other consequences etc.Therefore, you should reference the AI as a tool/resource when using AI for an assignment and/or research.
The following uses of AI are considered valid:
Note: If your use case is not present on the list, please confer with your course lecturers on whether or not your use case is valid before submitting your work.
Forbidden use cases:
Rewriting the work of other authors without the appropriate attribution.
For more information on using AI correctly and ethically, check the BUas guidelines on referencing AI: Usage & referencing guide on Generative AI
When using generative AI, make sure the output delivered to you is contributing to the value of the content you are working on. Everything that AI-tools produce comes from a dataset. If this dataset is incomplete, inaccurate, or subject to specific influences, it has implications for the output of the generative AI.
Always thoroughly check the output before you consider yourself as an author of the content:
How to check GenAI-output?
Use the methods below to help you assess AI-output:
Quick Assessment: SIFT
For a quick check, you can use the SIFT Toolbox for ChatGPT, created by the developer of the SIFT method itself. You can find it here, and it is still in an experimental phase.
In-depth Assessment: CRAAP
To critically assess AI output, you can use a customized version of the CRAAP test.
During the CRAAP test, you evaluate both the quality of the AI tool itself and its output using the same criteria:
C → CURRENCY (how current is the tool’s output)
R → RELEVANCE (is the tool relevant to your topic)
A → AUTHORITY (is the creator an authority on this subject)
A → ACCURACY (is the output true, correct, and reliable)
P → PURPOSE (why does the AI tool exist)
Below, these criteria are explained in more detail, along with an easy checklist you can use.
Currency (how current is the output?)
The most recent sources may not be included in the output if the tool is working with an older dataset.
Even if the dataset is current, the tool may generate references that are outdated.
Relevance (is the tool relevant to your topic?)
Is the AI tool the most suitable tool for your task?
Results can be incomplete. Some scientific publications are behind paywalls, which the tool cannot access and can only rely on title data and abstracts.
GenAI tools cannot construct comprehensive and reproducible search strategies.
Authority (is the creator an authority on this subject?)
It is difficult to evaluate who the author is and what expertise or authority they have in the field.
Results from AI tools often represent fewer perspectives. Training data leans heavily toward the West and English-speaking countries.
Accuracy (is the output true, correct and reliable?)
The accuracy of summaries and references varies greatly.
Hallucinations occur regularly. For example, fabricated references.
Actual citations do not always correspond with the arguments within a summary.
Double-check everything that is generated.
Purpose (what is the reason for the existence of the AI tool?)
BUas strives for the ethical use of AI, including considerations of academic integrity, diversity and inclusivity, data privacy, and environmental impact.
The company behind an AI tool is not always transparent about commercial interests, training methods, algorithm operations, and biases.
Bias in semantic searching: this technique is not always transparent and can result in loss of control over our search strategy, with a risk of cherry picking.
The original CRAAP Test is an evaluation method developed by Sarah Blakeslee of the Meriam Library, California State University, Chico. The University of Bristol has adapted this so that the method is suitable for AI tools.
Sources:
When you use a GenAI tool, you can write your prompt in a way that asks for references and a reference list. Be aware that some of the references it gives you might not be real. This is because many GenAI tools do not actually search the internet or access databases. Instead, they predict what the answer might look like, including what the references might be.
This means the tool can sometimes make up references. It is your responsibility to check every reference yourself and make sure that it is real and correct. Always think critically about any information you receive from a GenAI tool.
For general AI use, the in-text reference should be placed at the beginning, within or at the end of every paragraph which was written/aided by GenAI. It contains the following elements:
1. Name of GenAI tool
2. Version of the GenAI tool
3. Generation Date (& time, only if relevant)
Within your sentence you mention the verb which summarizes your written acknowledgement of the use of GenAI and its extent.
The in-text reference should be structured as follows:
(Name AI tool, Version, Generation date, Generation time)
Note: If it is a quote of the generated text, you add “…” in front and behind the text. For a paraphrase this is not necessary.
If there is no version available or if the version is part of the name (e.g. GPT4o), you can leave it out. It is mentioned on the reference list. At Buas we use Claude for Education. The name of the version is already included.
Examples of in-text references
Paraphrase:
The verb in the paraphrase is Explain. Time is not relevant in this example.
Quote:
The verb in the paraphrase is Explain. Time is not relevant in this example. Since this is a quote of the text, you add “…” in front and behind the text.
Also when you are using GenAI-tools as a writing aid, you will need to acknowledge this in your writing / assignments.
Examples of use cases of AI as a writing aid:
You can make a general remark of this in the Preface or at the Methods section of your document. If there is no Preface or Methodology section you can mention it in the Acknowledgements. It is not necessary to mention this for every occasion. The in-text reference contains the following elements:
The name of the Tool is mentioned in the text. When mentioned for the first time, the name of the tool is presented in italics. The Developer’s name and Version can be referred to between brackets at the end of the paragraph. If a tool has different modes, also mention the specific mode in your text. In case you use a specific version of a tool, you need to mention that as well. If this is not the case, you can leave the version out.
The in-text reference should be structured as follows: (Name of the developer, Version). Name of the tool is mentioned in the text itself.
Note: It is not part of the reference list
Example:
As part of the assignment multiple LinkedIn messages were created with Copilot (Microsoft, Enterprise version). I have used the creative mode. You can find the prompts and created LinkedIn messages in Appendix A.
This essay was written with the help of Claude (Anthropic, Claude for Education). Claude helped with checking grammar and rephrasing sentences. The model that was used is Claude Sonnet 4.1. The chat history is accessible via Appendix B.
Source and more examples: Usage & referencing guide on Generative AI
If you use GenAI tools as part of your research or creative process, it is important to acknowledge their use. Explain your application of these GenAI-tools in your Methodology section. If requested by your lecturers, you must be able to mention in your Methodology-section how you planned to use it, how you used it, the number of times you used it and during which period you used it.
If it is possible to make changes in the settings of the GenAI-tool, you have to explain your choices as well in your text. For instance, in Microsoft CoPilot there are different modes, such as creative mode. In your Methodology-section you can explain which mode you use and why you choose that mode. In Elicit you can add extra tables to your matrix.
Some GenAI tools have different accounttypes. Such as a free version, basic version, plus version or a subscription based version. In your Methodology-section you explain which acounttype you use.
Some GenAI tools have different versions over time which are based on their language models. For instance ChatGPT has ChatGPT-3.5 (November 2022), ChatGPT-4 (March 2023) and ChatGPT-4o or ChatGPT-4omni (May 2024).
At BUas we use Claude for Education, with access to several language models. When using Claude for Education, also mention which model you use.
Example of use cases:
Text as input and as output
If you have text as input and output, the description/prompt (= the input) and the text that was generated by the GenAI-tool (= output) need to be added in the body of your text. You indicate in the body of your text which part was generated. You can do this as a quotation or a paraphrase.
If the input or output is longer than 25 words or you have multiple prompts or generated texts or images: create an appendix or supplementary file instead to log it in its entirety. In some cases the GenAI-tool offers the possibility to create a public link to your interaction with the tool itself. This is the case for tools such as Gemini, ChatGPT, Meta and Claude. You can use this tool: https://aiarchives.org/ to make your own URL. This URL is also allowed provided that you add it to the reference list.
The in-text reference contains the following elements:
The name of the Tool is mentioned in the text. When mentioned for the first time, the name of the Tool is presented in italics. The Developer’s name and Version can be referred to between brackets within or at the end of the paragraph.
Note: If a tool has different modes/styles, such as Copilot and Claude have, also mention the specific mode in your text. In case you use a specific version of a tool, you need to mention that as well. If this is not the case, you can leave the version out. The in-text reference should be structured as follows: (Name of the developer, Version)
| Example 1: paraphrase As part of the assignment a literature review matrix on my topic was made with the help of Elicit (Elicit Research, Elicit Basic). I have used this prompt: “What is the influence of generative AI on the Hospitality industry?”. Based on the presented matrix, I choose to add two extra columns “methodology” and “limitations”. After that I choose the second article to work from… Example 2: multiple prompts As part of the assignment I was asked to review Claude (Anthropic, Claude for Education). Multiple prompts were developed to see the difference between the writing styles in Claude. You can find the prompts and the output in Appendix B. Based on the output my conclusions are as followed: …. Example 3: multiple versions The text for the leaflet on Ethical AI use is processed with Quillbot (Learneo). The first draft is presented in Appendix A. Suggestions for revisions from Quillbot (Appendix B) are processed in a second draft (Appendix C). |
Input is not a text
When the input is something other than text, you should mention the type of media you used. This applies to images, clip art, speech, music, code, etc. Always check the copyright of your input. Maybe it is necessary to ask for permission.
There are two situations:
| Example 1: The input is a picture that you took of a beach. Use the APA Guideline for pictures. Mention your own name as the author. You refer to it in the body of your text and on the reference list. The picture is placed in a folder on Brightspace. Use the URL for the reference. Example 2: The input is a video on sustainability that you made yourself. Use the APA Guideline for video’s. Mention your own name as the author. You refer to it in the body of your text and on the reference list. The video is placed on YouTube. Use the URL for the reference. Example 3: The input is a painting by a famous artist from the renaissance and you want to generate a new picture in the same style. You refer to the original work in the body of your text and on the reference list. You put the generated picture in your portfolio and mention the prompt used. |
Output is not a text
In this case your output is not a text, but something else. You need to mention the type of media that was created (i.e. video, speech, image, etc.). Place (the relevant) part of the output in an appendix or make it available online with an URL. You can use the regular APA guidelines belonging to the type of media that you have created. You are the author of the output.
|
Example 1: create concept art with Dall-E 2, images in an appendix Note. Images generated with the prompt: “Make several pictures that feature a patriotic couple” by OpenAI (Dall-E, Version 2). Reference: OpenAI. (2023). DALL-E (Version 2) [Image generator]. https://labs.openai.com |
| Example 2: create concept art with Dall-E 2, image is in the body of your text, prompt was more than 25 words Body of your text: An inspirational image was generated with Dall-E (Open AI, Version 2). This image can be found in Figure 4. The game will feature a patriotic couple which repairs dishwashers by day and combats crime by night. Figure 4 (In bold) A patriotic couple (In italics)  Note. Image generated by OpenAI, Dall-E (Version 2), see Appendix C for the prompt. Reference: OpenAI. (2023). DALL-E (Version 2) [Image generator]. https://labs.openai.co |
|
Example 3: the input is a picture, the output is a video Body of your text: |
The following information should be included in a reference on your reference list:
Written acknowledgement of the use of the tool and its extent are mentioned in the body of the text.
The reference list should be structured as:
Developer GenAI tool. (Year). Name of GenAI tool (Version) [Type of AI]. URL
If there is no version available or if the version is part of the name (e.g. GPT4o), you can leave it out.
In regards to AI types. There are several sub classifications possible for generative AI.Based on what it generates you can distinguish tools for text, audio, code, image, etc. In the table below you can see what you can use in your reference as a type. If you are not sure, you can use “generative AI”. Since this is an adjustment made during this academic year, it is allowed to use both options: either “generative AI” as a type or a sub classification. This adjustment was made in November 2024.
| Output | Use in you reference | Example |
| Text | Large Language Model or text generator | Microsoft Copilot, ChatGPT, Claude |
| Audio or music | Audio or music generator | Suno, Udio |
| Code | Code generator | GitHub, Copilot, Claude |
| Image | Image generator | Dall-e or Stable Diffusion |
| Reference, articles, literature review | Large Language Model or text generator | Elicit, Perplexity |
Reference list entries should have a hanging indent of 0.5 in. Start the first line of each reference at the left margin; indent each subsequent line five spaces.
|
Anthropic. (2025). Claude for Education [Large Language Model]. https://claude.ai/chat/3601a5bf-0ef2-4f00-8211-51546c OpenAI. (2024). ChatGPT4o [Generative AI]. https://chatgpt.com/?model=gpt-4o |
