Generative AI Guidance: Using GenAI Responsibly

Referencing and citation

Below are links to Cite Them Right's generative AI guidance.

• Harvard
• APA
• MLA
• MHRA - 4th edition
• IEEE - refer to chicago
• Chicago
• OSCOLA

Academic misconduct

Just like failing to declare sources and existing research in your work constitutes academic misconduct, the same goes for failing to declare your use of generative AI.

Goldsmiths requires that all staff and students abide by its values of honesty and trust within their work. For students, it is necessary to abide by the college's academic misconduct policy and procedure regarding AI tool usage:

Use of Artificial Intelligence (AI) tools to produce work that is presented as the student’s own. The use of AI tools may be legitimate within some assessments, where this is the case, this will be set out in the module specification or assessment brief. The use of AI tools in this instance will be appropriately cited within the student’s submitted assessment. (Goldsmiths, University of London, 2023)

You can reference an AI tool if usage is allowed. It is also useful to note the date and version of the tool, as technology in this area develops and changes very rapidly. However, please remember that it is not considered an academic resource and usually lacks references. Do not copy directly from an AI as this is a form of plagiarism.

References

Goldsmiths, University of London. (2023). Academic Misconduct Policy and Procedures. https://www.gold.ac.uk/media/docs/gam/Academic-Misconduct-Policy-and-Procedures.pdf 

Generative AI in assessment

Goldsmiths VLE Statements

Your course convenor will have selected from 3 options on your module VLE page to let you know what how you can use generative AI tools in your assessment submissions:

1. Generative AI tools may be used for any assessments in this module, including for elements that will be included in the final submission.
2. The use of generative AI tools is permitted during the development process of assessments in this module, but students are not permitted to submit AI-generated work in the final submission.
3. Generative AI tools must not be used in any way for any assessments in this module.

Plagiarism Checkers

Turnitin checks for similarity between a students submitted work and published work. Turnitin has also introduced an AI detection tool in April 2023. However, Goldsmiths and other universities have not enabled it mostly due to accuracy concerns and unfairness towards those who have used AI-powered translation tools in their submissions.

Academic misconduct

Just like failing to declare sources and existing research in your work constitutes academic misconduct, the same goes for failing to declare your use of generative AI.

Goldsmiths requires that all staff and students abide by its values of honesty and trust within their work. For students, it is necessary to abide by the college's academic misconduct policy and procedure regarding AI tool usage:

Use of Artificial Intelligence (AI) tools to produce work that is presented as the student’s own. The use of AI tools may be legitimate within some assessments, where this is the case, this will be set out in the module specification or assessment brief. The use of AI tools in this instance will be appropriately cited within the student’s submitted assessment. (Goldsmiths, University of London, 2023)

You can reference an AI tool if usage is allowed. It is also useful to note the date and version of the tool, as technology in this area develops and changes very rapidly. However, please remember that it is not considered an academic resource and usually lacks references. Do not copy directly from an AI as this is a form of plagiarism.

References

Goldsmiths, University of London. (2023). Academic Misconduct Policy and Procedures. https://www.gold.ac.uk/media/docs/gam/Academic-Misconduct-Policy-and-Procedures.pdf 

Accessibility and Inclusion

Generative AI tools offer some exciting possibilities around accessibility and inclusivity – particularly from an assistive technology and neurodiversity perspective.  However there are still concerns around representation of minoritised communities in the outputs of genAI tools due to the western-focussed, Eurocentric data used to train generative AI models.  There is also inequality of access to digital and online knowledge, skills and tools across the world – and this extends to access to generative AI tools – this is called the digital divide.  As such, can we label genAI tools as authentically inclusive and accessible when so many people worldwide simply do not have a means to access them.

That said, here are some potential applications for genAI as an assistive technology:

• The language processing power of LLMs make them adept speech-to-text and text-to-speech tools.  Remember: do not input any personal data (yours or other people’s) into LLMs.  Read more about this in the Transcription section. GenAI tools have also improved speech synthesis in text-to-speech tools: Google NotebookLM’s podcast feature is a good example of this.
• Computer vision technology allows computers to ‘see’ images and videos as a whole (rather than bits and streams of data).  This means alternative text and audio description can be quickly and easily generated for those with visual impairments.  Microsoft applications have this built-in for images – but beware of hallucinations!
• LLMs and other tools can be used to summarise lengthy and complex text, making reading easier and more accessible.  These tools can also be used to change the tone of text (into something sterner or more polite) which can be useful for those of us who are neurodiverse.  Similarly, we can use genAI tools to break down big complex tasks into smaller, easily digestible ones making it easier to complete projects.

Bibliography

Henneborn, L. and Burden, A. (2024) Inclusivity in Generative AI Should Be an Attribute, Not an Add-On, Stanford Social Innovation Review. Available at: https://ssir.org/articles/entry/inclusive-generative-artificial-intelligence (Accessed: 26 September 2025).

Mainstreaming accessibility and inclusivity in AI and digital (2024) UNESCO. Available at: https://www.unesco.org/en/articles/mainstreaming-accessibility-and-inclusivity-ai-and-digital-technologies (Accessed: 26 September 2025).

Syed, W. (2025) ‘Advancing Accessibility and Social Inclusion through Generative AI and Machine Learning-Powered Multi-Modality on Mobile Platforms’, International Journal of Scientific Research in Computer Science, Engineering and Information Technology, 11(1), pp. 475–483. Available at: https://doi.org/10.32628/CSEIT25111254.

Welker, Y. (2023) Generative AI holds great potential for those with disabilities - but it needs policy to shape it, World Economic Forum. Available at: https://www.weforum.org/stories/2023/11/generative-ai-holds-potential-disabilities/ (Accessed: 26 September 2025).

Thinking critically and evaluating outputs

Critical use of any resource is necessary within your academic work. When using any generative AI tool (from AI writing tools to AI image generators), we must assess the suitability for our work, management and ownership of the tools, and the wider impact of using such tools.  You can find more information on some of the areas to consider on the Ethical and Legal Implications page of this LibGuide.

Recent research published by researchers at Microsoft and Carnegie Mellon University (Lee et al., 2025) identified a worrying link between overuse of genAI tools and a decline in critical thinking.  They found that those with higher confidence in genAI capability and outputs were less likely to critically evaluate and interrogate those outputs, while those who had more self-confidence in their own skills, abilities and knowledge were more likely to engage in critical thinking when using genAI tools.

ROBOT Test

Below is a set of questions to support you in critically evaluating the AI tools you might use within your work. LibrAIry's original test has been modified slightly to include additional considerations. It is recommended you ask these questions before and throughout your research project.

In addition to the above questions, you want to consider the potential environmental impacts of using AI. Development and usage of AI requires a substantial amount of energy and resources which have a significant environmental impact.

Hallucinations

Hallucinations occur when an AI tool produces incorrect or non-existent information.  Often this information can look real or factually correct.

It is important to always double check the information returned by an AI tool. There are various cautionary tales in academia and industry about its usage including a lawyer who cited fake articles recommended by ChatGPT who may now be facing sanctions for presenting false information to the court.

Here at Goldsmiths, multiple staff have reported LLMs producing false sources and references - presented in a list that looks correct, professional and adheres to referencing styles but further exploration identifies non-existent sources such as journal articles and book chapters. So the moral of this story: always fact check your outputs!

References

Lee, H.-P. et al. (2025) ‘The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers’, in Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. CHI 2025: CHI Conference on Human Factors in Computing Systems, Yokohama Japan: ACM, pp. 1–22. Available at: https://doi.org/10.1145/3706598.3713778.

Transcription

It can be very tempting to use generative AI tools for transcription – especially with research interviews or focus groups and online meetings or lectures!  If you must use a transcription tool, you should always request consent from all interviewees, attendees and/or the presenter before activating tools – both as a courtesy and from a data protection and privacy perspective!  It is important to note that research into generative AI transcription tools has raised significant concerns regarding their accuracy, reliability and hallucinations and that human transcription mitigates these issues in most cases.

Remember there are built-in dictation and transcription tools in Microsoft products.  As part of Goldsmith’s contract with Microsoft, certain safety measures and guardrails are in place to minimise the transfer of personal and sensitive data as defined under the UK GDPR.

As with all generative AI tools you should be mindful of hallucinations – where generative models make up information and present it as fact.  In research conducted by Koenecke et al. (2024) identified that 40% of hallucinations fabricated by OpenAI’s Whisper tool could be considered harmful or of concern due to the speakers being misinterpreted or misrepresented.  You should try retain recordings of what you are transcribing in order to ensure that transcriptions are accurate.

Generative AI transcription tools can also struggle with understanding accents or specialist vocabulary.  You should be mindful of this if you plan on transcribing speech using genAI.

Bibliography

Burke, G. and Schellmann, H. (2024) Researchers say an AI-powered transcription tool used in hospitals invents things no one ever said, AP News. Available at: https://apnews.com/article/ai-artificial-intelligence-health-business-90020cdf5fa16c79ca2e5b6c4c9bbb14 (Accessed: 23 September 2025).

Koenecke, A. et al. (2024) ‘Careless Whisper: Speech-to-Text Hallucination Harms’, in The 2024 ACM Conference on Fairness, Accountability, and Transparency. FAccT ’24: The 2024 ACM Conference on Fairness, Accountability, and Transparency, Rio de Janeiro Brazil: ACM, pp. 1672–1681. Available at: https://doi.org/10.1145/3630106.3658996.

Patel, R. (2025) ‘GenAI vs NMT: Understanding the Future of Language Translation’, AiThority, 23 January. Available at: https://aithority.com/machine-learning/genai-vs-nmt-understanding-the-future-of-language-translation/ (Accessed: 23 September 2025).

Siegel, R. et al. (2023) ‘Poster: From Hashes to Ashes - A Comparison of Transcription Services’, in Proceedings of the 2023 ACM SIGSAC Conference on Computer and Communications Security. CCS ’23: ACM SIGSAC Conference on Computer and Communications Security, Copenhagen Denmark: ACM, pp. 3534–3536. Available at: https://doi.org/10.1145/3576915.3624380.