Generative AI and Research

Khalifa and Albadawy (2024) carried out a systematic review of where GenAI is being used in academic writing; ‘Using artificial intelligence in academic writing and research: An essential productivity too’. Their research identified six core domains where GenAI can make improvements to the research process:  

•  Facilitating idea generation and research design
• Improving content and structuring   
• Supporting literature review and synthesis
• Enhancing data management and analysis
• Supporting editing, review, and publishing
• Assisting in communication, outreach, and ethical compliance

Khalifa, M. and Albadawy, M. (2024) ‘Using artificial intelligence in academic writing and research: An essential productivity tool’, Computer Methods and Programs in Biomedicine Update, 5, p. 100145. Available at: https://doi.org/10.1016/j.cmpbup.2024.100145

When using a GenAI tool to process data you must abide by the existing University of Kent Data Protection Policy, and carry out a Data Protection Impact Assessment Screening Questionnaire. Completing the questionnaire will assess whether your use of GenAI is high risk and whether a full Data Protection Impact Assessment is required, for more information visit the Assurance and Data Protection pages.  

You will need to look at the small print of the GenAI tool that you use to check what the tool will do with your data. Most tools store and log the data that you upload. You’ll need to think about where the data will be stored, if your data will be used in answers to other users of the tool, as well as any data protection, copyright and privacy concerns.  

Bias can occur from the way that GenAI tools have been created and modelled by their developers, and through the data that they have been trained on. Any outputs from GenAI tools will need to checked and evaluated for accuracy. 

For further guidance on dealing with biases view the Kent AI Ethics page, section 3 covers Fairness, Bias & Cultural Sensitivity:

AI tools are trained on existing data, which can include biases or stereotypes. This means their responses might reflect limited perspectives. Ethical use involves questioning and challenging these patterns.

• Evaluate responses for bias and inclusivity to avoid reinforcing stereotypes.
• Be mindful of cultural and social implications when interpreting AI-generated content.
• Check multiple sources, not just AI outputs, for diverse viewpoints.

As the user of the GenAI tool and an author in the research process, it is your responsibility to check outputs generated by GenAI for accuracy and biases. 

Content produced by GenAI will need to be thoroughly checked for accuracy. GenAI tools cannot check for accuracy themselves, and do not fully comprehend the information that they output. When using AI researchers must ensure ownership of any content produced, meaning that they check for accuracy and take responsibility for what is produced. 

An ’AI hallucination’ is a response produced by a GenAI tool that sounds plausible but is factually incorrect, logically flawed, or completely invented. This could include inventing a fake academic article or citing a legal case that has never existed.

You will need to verify facts and figures from outputs against other sources and follow up references to check that they exist.

GenAI is trained on huge amounts of data. When using GenAI researchers will be adding to the data consumed by AI tools. Often this data will be a personal nature, so it is the responsibility of individuals to be aware of data protection and ethical research practices.  

When using GenAI we must adhere to the principles of the UK General Data Protection Regulation (UK GDPR) and the Data Protection Act 2018. The Information Commissioner’s Office which upholds information rights in the public interest have their own Guidance on AI and data protection. 

Under the University's Data Protection policy staff are required to carry out due diligence checks when working with a third-party service provider which will process, handle or store personal information. This applied to researchers planning to use GenAI in research.

The Assurance and Data Protection team at Kent have produced specific guidance for Kent staff using Artificial Intelligence. Under the guidance, when using GenAI the key considerations are:

• accountability
• the requirement to carry out a Data Protection Impact Assessment when using a third-party provider  
• referencing when you’ve used GenAI
• making your own informed decisions
• never uploading private or sensitive data
• to understand the default settings of GenAI tools ensuring that data cannot be shared unwillingly.

There is a separate Data Protection and Ethics process integration for research students to view.

Think of research data as any information that has been collected, observed, generated or created to support or validate original research findings. Data can take multiple forms including: survey results interview recordings and transcripts, observations, models, code, notebooks, lab books, videos and images, and sketchbooks.

Any planned usage of GenAI should be recorded in your Data Management Plan, including where the information will be stored and the reasons for use of GenAI.

GenAI use should be fully documented and recorded. Usage of GenAI will need to be declared when publishing in a journal. A full and accurate record of your use of Gen AI demonstrates academic integrity and transparency. Recording GenAI use and archiving finalised research data in a data repository means that you or others can revisit your research when needed.

You should record: the tools used; when the tools were used; the prompts or steps used to create the information; and the information itself. 

ChatGPT, the most popular GenAI tool, usually does not list its sources.  If you ask it for references, they are often hallucinated. Generally, GenAI synthesises multiple sources in its outputs, amalgamating them into a whole. This can make it difficult for the tool to then provide attribution, as it is piecing together new content from an array of existing material. 

Certain GenAI tools, such as CoPilot, will provide references and link back to the sources they have used for their outputs. Scopus AI which is provided by the library will only search and provide references to academic articles.  Consequently, some GenAI tools are preferable to others.

There is an increasing focus on how GenAI tools are being used to add content to their models. The Atlantic developed a searchable database of what Meta, the makers of Facebook, used to train their GenAI models. 

Researchers currently have little control over GenAI tools making deals with publishers. However, there are still methods that authors can use to safeguard the attribution of their research when they are sharing their research online. 

Before uploading part or all your writing to a GenAI tool you should look at any available terms and conditions. Even if you upload an article that you are working on for grammar checking it will most likely be added to the GenAI model as training data. This could also cause problems around the intellectual property and copyright of your research with your funders and publishers.  There has been coverage in academic and publishing news on Academia.edu podcasts, where articles uploaded by authors have been turned by an AI tool into podcasts and added to the article pages on Academis.edu. In this instance the authors were asked if they would like the podcast added to their page, but it is a cautionary tale of what can happen to your research.

Simply using copyright to stifle GenAI would create issues around innovation and access to knowledge, so Creative Commons are working to a future which is based on responsible reuse. Creative Commons are currently developing Signals, which is a framework for how an author’s work can be used in GenAI training. The point of Signals is to develop new norms in how machines read and use content. Content has been read by machines for search engines and to archive knowledge for many years prior to GenAI. However, GenAI tools are using content in a different way, a way that is without transparency or credit. 

The signals are to be applied by a Declaring Party, who declares how a content collection is to be used by machines. The categories of use they are applied to range from general to more specific categories, such as Text and Data Mining, AI Training, Generative AI Training, and AI Inference. The four proposed signal combinations are:

• Credit
• Credit + Direct Contribution
• Credit + Ecosystem Contribution
• Credit + Open

There is a detailed report on Signals from the Creative Commons ‘From Human Content to Machine Data Introducing CC Signals’, which provides further background on why they are being developed and how they plan to take them forward.

Under the Creative Commons proposed implementation the four signals are:

• Credit: GenAI tools must give appropriate credit based on the method, means, and context of their use of a source.
• Direct Contribution: GenAI tools must provide monetary or in-kind support to the Declaring Party for their development and maintenance of the assets, based on a good faith valuation taking into account their use of the assets and their financial means.
• Ecosystem Contribution: GenAI must provide monetary or in-kind support back to the ecosystem from which they are benefiting, based on a good faith valuation taking into account their use of the assets and their financial means.
• Open: The AI system used must be open. For example, AI systems must satisfy the Model Openness Framework (MOF) Class II, MOF Class I, or the Open Source AI Definition (OSAID).

The Open Access movement was developed to provide unrestricted access to research outputs. Over time Open Access material has also been used for text and data mining by researchers.

In 2014 an exception for Text and Data Mining was added under Section 29A of the Copyright, Designs and Patents Act 1988 (CDPA).

“An exception to copyright exists which allows researchers to make copies of any copyright material for the purpose of computational analysis if they already have the right to read the work (that is, they have ‘lawful access’ to the work). This exception only permits the making of copies for the purpose of text and data mining for non-commercial research.”

This allows text and data mining if you have lawful access to the work, and if the use is for non-commercial research. The intention behind this is to allow text and data mining for research purposes. When the changes were made to the Text and Data Mining section of the Act, it was long before GenAI became ubiquitous. 

Copyright has not protected authors, artists and creators from seeing their material trained on by GenAI tools.

In 2025 a case was brought by Getty Images against Stability AI in the UK. Stability AIs GenAI tool ‘Stable Diffusion’ is a model which turns textual prompts into images. Getty’s case was on Stable Diffusion being trained on their copyrighted images. The first claim that the training of the GenAI tool took place in the UK had to be dropped as there was no evidence that the training took place in the UK. The trial then focused on  secondary copyright infringement and trademark claims under UK law. The judge ruled that: “An AI model such as Stable Diffusion which does not store or reproduce any copyright works (and has never done so) is not an ‘infringing copy’.” The Judge did rule in Getty’s favour of some of Getty’s claims about trademark infringement related to watermarks, as Getty watermarks had appeared in images created by Stable Diffusion.

Court cases are prohibitive under the current legislation due to cost, as only well-funded organisations will take on legal cases which focus on arguments such as models retaining material and where training has taken place at this stage.  The current legislation is also out of date with current practices and technology.

The UK Government recently consulted on Copyright and Artificial Intelligence. This led to some proposals for how Artificial Intelligence can be used in the Data (Use and Access) Act 2025. However, the proposed changes were rejected, and it was decided that a separate investigation will take place into AI which will cover all areas, not just data. 

No, handing over your copyright to a publisher and publishing behind a subscription paywall will not prevent your work from being used by GenAI. GenAI will be trained on academic publishing, whether that is through open repositories, through agreements with publishers or by people uploading research themselves in their interactions with apps. In fact, if you publish your work in a 'traditional' subscription journal you assign copyright in your work over to the publisher who is then free to sell this onto GenAI tools.

The current relevant UK legislation, the Copyright, Designs and Patents Act 1988, allows text and data mining and recent court cases that have challenged this have not been successful. Current UK copyright legislation has not protected authors from seeing their material trained on by GenAI tools.  Choosing to publish non-Open Access goes against University, funder and REF policies, loses all of the benefits of Open Access and provides no more protection against GenAI use.

Across the HE sector there has been concern from academics about making their research available Open Access in repositories. These concerns are based in the fear that they will lose control over their work, which will likely become training material for GenAI and therefore used commercially.  There are suggestions that academics should opt out of Open Access. However, by choosing to opt out of making your research Open Access, you may risk non-compliance with funder policies and potentially make the publication ineligible for REF-submission. 

The University of Kent statement on Open Research states “We use technologies and scholarly practices to widen access to research, increase transparency in the research process, and improve reproducibility of results”.  GenAI can further the promotion of an Open Research culture by making research accessible to a greater number of people, which maximises its reach and benefit to the public.  

GenAI will be trained on academic publishing, whether that is through repositories, agreements with publishers or people uploading research themselves in their interactions with apps. Restricting access to research is unlikely to stop GenAI from accessing research. 

GenAI will benefit from being trained on academic publications. Ideally this would be in a transparent way; regulated through deals with publishers that researchers are fully aware of, and attribution ensured through GenAI outputs which are referenced with links to the original research. 

Libraries have been reporting issues with the performance of their repositories related to the activities of bots scraping data for GenAI models. A survey from the Confederation of Open Access Repositories (COAR) in 2025 found that over 90% of respondents reported issues, that led to outages and poor performance. GLAM-E Lab, who works with institutions from the Galleries, Libraries, Archives, and Museums sector, carried out a survey in 2025 with institutions who have digital collections. This survey indicated a large increase in traffic to their digital collections, which most respondents attributed to bots.  

Both surveys pointed to concerns that the increase in traffic was having on people being able to find and access material on repositories. Some bots identify themselves to the repositories, but many have been taking measures to hide where they are coming from. As it stands, we are unable to know how much of an impact bots are having on KAR, or if any research is being taken.

The Committee on Publication Ethics (COPE) which aims to provide guidance and leadership on publication ethics to all involved in scholarly publishing issued the Cope position in 2023. The key principle of their position is around authorship; AI tools cannot be listed as an author of a paper as they are unable to take responsibility for any submitted work. AI tools are non-legal entities, cannot manage copyright or licence agreements, or manage any conflict of interests. The position calls for the transparent disclosure of which tools were used and how they were used in any Materials and Methods (or similar) section. Authors are responsible for all their articles; this includes any material produced by GenAI and are responsible for breaches of publication ethics. 

If you are an author, editor or peer reviewer and have used GenAI tools in the publication process you will need to check the guidance from the publisher you are working with. Some publishers have specific guidance for AI use; others have incorporated AI into their ethics guidance. 

• American Chemical Society (ACS) 
• American Institute of Physics (AIP) 
• American Physical Society 
• American Physiological Society 
• Association of Computing Machinery (ACM) 
• Bristol University Press 
• Cambridge University Press (CUP) 
• Elsevier 
• Emerald 
• Institute of Electrical and Electronics Engineers (IEEE) 
• Institute of Physics (IoP) 
• Optica 
• Oxford University Press (OUP) - find the instructions for each specific journal title
• Royal Society of Chemistry (RSC) 
• Sage 
• Springer 
• Taylor & Francis 
• Wiley 

Certain journals will have their own author guidelines which can differ to those of the publisher, check on the journal page in case it differs.

Developers of GenAI tools need high-quality material to train their models. Publishers have vast collections of high-quality material, which they are keen to use to generate additional income. This has led to many academic publishers licensing their content to GenAI tools.  

The University of Kent has adopted an Author Rights Retention approach for the ownership of researchers’ publications, which started from the 1 August 2025, when our institutional Research Publications policy came into effect. Even with this approach publishers own the copyright for the material in their journals and can make deals. When research is published, authors transfer their copyright, often through copyright transfer agreements with publishers so that their articles can be published. The publishers then retain the right to reuse the research in the form that it was published in their journals. 

Wiley, Taylor & Francis and Sage have all licensed content to GenAI providers to generate income. Ithaka S+R has developed a Generative AI Licensing Agreement Tracker where information about these deals is kept, and where you can check to see which publishers have agreed deals and with which tools. 

Publishers have taken different approaches to how they engage with researchers over their deals with GenAI providers. Informa, the parent company of Taylor & Francis signed a deal with Microsoft in May 2024. It was the first high profile agreement to be reported in the mainstream media. Much of the coverage of the deal focused on how authors were not informed or consulted (Potter, 2024). Cambridge University Press made the news for asking authors to opt-in and sign a contract addendum that would allow the licencing of their books for AI use in exchange for a royalty payment from the deal made with a GenAI tool. 

You will need to check the policy of the journal that you are submitting your article to for their exact requirements. Generally, any AI use must be acknowledged in your article, within the methods or acknowledgement section.  The acknowledgement should state the tool including the version number, how it was used and why it was used.  

Some publishers have exceptions over the need to acknowledge GenAI when it has been used for help with translation, structure and grammar. Check the guidance for the publisher you are submitting to before leaving out an acknowledgment as it could become an issue that will delay the process. 

Putting an acknowledgment to GenAI use in the acknowledgements or methods section of the article will currently suffice but is likely to lead to inconsistencies in how this information is presented and recorded. To make GenAI use more transparent a group of researchers have developed the Generative AI Delegation Taxonomy (GAIDeT). Based on the CRediT (Contributor Roles Taxonomy), it has been developed to display the specific nature of how researchers are using AI in their research by mapping AI use to research tasks.  

The developers have created a GAIDeT Declaration Generator, which if completed will create a declaration that can be added to a publication’s disclosure, methods or acknowledgement section. Although there are no requirements for a standardised way of declaring GenAI use, the GAIDet taxonomy will provide transparency for your research until this, or another method reaches critical mass and become accepted. 

The growth of GenAI use has caused issues for publishers. Monitoring the use of AI in article submissions has proved difficult for publishers to manage. Publishers such as Wiley and Springer Nature are using AI detection services, or are developing their own in-house tools to spot problematic article submissions. AI detection tools bring up false positives and will highlight uses of translation and grammar checker software which have AI elements. 

Academic publishing is developing methods to combat undisclosed use of GenAI in submissions, including policies, acknowledgements and the GAIDeT taxonomy. Publishers are also relying on the peer review process to pick up any potential issues, however not every fact is checked by reviewers. The publishing process is built on trust, if authors are asked to disclose GenAI and then choose not to, there will inevitably be problematic papers. 

An example of the difficulties faced by publishers, editors and peer-reviewers can be seen in the controversy around the article ‘Cellular functions of spermatogonial stem cells in relation to JAK/STAT signaling pathway’ published in the journal Frontiers in Cell and Developmental Biology. 

Images within the article had been created using GenAI image creation tool Midjourney. This was permitted under the journal’s publication ethics guidelines as they had acknowledged the use of Midjourney, even if it was not declared in the acknowledgement or methods sections. The authors of the article did not properly check the images created by the tool; the images weren’t picked up by the peer-reviewers or the journal’s editors. 

The image in question was an out of proportion anatomical drawing of a rat. The paper was quickly retracted but is still available to view as the article contains original research even if the images are problematic. The article soon gained notoriety and received coverage in the mainstream media, initiating a debate around the quality and quantity of journal articles produced (Sample, 2025). The case highlights the issues around GenAI and the publishing process. GenAI use can be hard for publishers and peer reviewers to spot. The system relies on authorship and the transparent use of acknowledgements by those submitting to journals, without it other cases will slip through the net. 

Sample, I. (2025) ‘Quality of scientific papers questioned as academics “overwhelmed” by the millions published’, The Guardian, 13 July. Available at: https://www.theguardian.com/science/2025/jul/13/quality-of-scientific-papers-questioned-as-academics-overwhelmed-by-the-millions-published 

The Research Funders Policy Group is made up of the: Association of Medical Research Charities (AMRC), British Heart Foundation (BHF), Cancer Research UK (CRUK), National Institute for Health and Care Research (NIHR), Royal Academy of Engineering, Royal Society, UK Research and Innovation (UKRI). The group made a joint statement on the use of generative AI tools in funding applications and assessment.  

 The statement says that researchers: 

• must ensure that AI tools are used responsibly and in accordance with legal and ethical guidance
• acknowledge GenAI use in funding applications

In their policy Use of generative artificial intelligence in application preparation and assessment, UKRI define how GenAI can be used when producing applications. Funding applications are expected to uphold the values of integrity, honesty, rigour, transparency and open communication. Applicants and assessors must balance the benefits and risks that come with using GenAI. There are two areas of values and expectations that the applicants and assessors must adhere to: 

• Research integrity – there are several key values linked to the URKI’s concordat on research integrity. These cover honesty, rigour, accountability, transparency and open communication, care and respect, and confidentiality in the application and assessment process
• Trusted research and innovation principles – covers areas such as intellectual property, confidentiality, sensitive research

UKRI’s strategy Transforming our world with AI (2021) set out how AI can have a real world impact if cross-disciplinary and cross-sector research takes place. There is a need for investment in the use of AI to realise its potential, which can then have real world applications to impact on areas such as economic growth, climate, health, public services, transport and inequality. To enable this there is a need for technological developments, this needs to be accompanied by research from humanities and social sciences. The report outlines how UKRI will support research and innovation in AI to bring about changes to society. 

Wellcome have a policy on the Use of Generative Artificial Intelligence (AI) when applying for Wellcome grant funding. The policy provides guidance for those applying for funding from Wellcome. Researchers must disclose the use of AI in funding applications and are responsible for the use of AI in applications.