This article was originally written in 2023. It was updated in 2025 to reflect new developments.
ChatGPT has been taking the world by storm since its release in 2022. From coding assistance to book summaries, people have been using the chatbot to help access and understand information where a simple Google query might fall short. The technology seems to have endless potential, but how could it transform the life sciences field?
When we first published this feature in 2023, we looked at some current and future applications of ChatGPT that could revolutionise the way we view research, whilst also considering the caveats of the tech. Two years later, we’re revisting the topic to see what’s changed.
What is ChatGPT?
ChatGPT (generative pre-training transformer) is an online artificial intelligence (AI) chatbot. It is a large-language model (LLM) that was developed by OpenAI and released in November 2022. In just two months, it became the ‘fastest-growing consumer software application’ ever developed. Users can approach the bot with a question, and it will do its best to give a comprehensive answer that reflects a human conversation.
It was trained on vast amounts of textual data to learn the patterns and structure of human language. Using this training, it can predict what word should come next in a sequence. ChatGPT’s potential partly stems from its relative ease-of-access; unlike many other AI models, it is freely available to the public online. The more people use the tool, the more training it receives – meaning it is constantly evolving. But who better to tell you a bit about ChatGPT than ChatGPT itself?
The use of AI is becoming far more common in life sciences research, and ChatGPT is far from the only publicly available tool of this kind. However, given its ability to quickly perform tasks that would take a human hours, if not more, the potential is immense. In recent months, we’ve seen controversy as generative AI tools were used to create shocking images for publications, and the prevalence of suspected AI-assisted manuscripts increased dramatically. But what are some genuine up-and-coming applications of the technology, and what do you need to watch out for?
How has ChatGPT changed this year?
Between 2023 and 2024, there were several updates that have led to the improvement of ChatGPT. Namely, the launch of GPT-4o, which free users are given limited access to, which can accept text, audio, image and video input, providing audio, image or improved text output. Access to the internet has also transformed the free version of the tool. The capabilities of the model have also improved, notably in languages other than English. This summary from OpenAI themselves explores the changes, made public in May 2024.
In 2025, things have become even more sophisticated. OpenAI’s new Deep Research tool, powered by the o3 model, autonomously browses and analyses the web to deliver expert-level reports in minutes rather than hours. Available to Pro users at this time, its suggested uses are for documents such as literature reviews, policy briefs and project proposals.
A quick literature review
Perhaps the most obvious use of ChatGPT is to get it to trawl through vast amounts of literature on your behalf. With the ability to quickly scour the web, the bot might be able to find an obscure source that would have otherwise gone unnoticed and can quickly summarise whatever it finds. Alternatively, you can provide the chatbot with your own sources and have it pluck out key ideas. This has been one of the most common uses of the technology so far, with many treating the language model as a new and improved Google.
At the time of writing the original version of this article, the free version of ChatGPT only had access to the internet up until 2021, and as such it was unable to discuss more recent developments. However, the May 2024 launch of the latest version, GPT-4o, sees users on Open AI’s free plan given limited access to the tool’s web browsing capabilities. However, as a large language model,, ChatGPT still operates under the assumption that whatever it is saying should be true, and is generally unable to fact check its answers. As such, it’s highly recommended that users do not rely on the tool to carry out work on their behalf.
That said, the model is exceptionally good at writing summaries and picking out the relevant points in a piece of work. For that reason, some researchers have started to use ChatGPT as a starting point for literature reviews, prompting the chatbot to highlight key areas of interest and important papers. In fact, when we asked ChatGPT for its own opinion on its role in research it stated the following:
“AI can now conduct comprehensive literature reviews by scanning and summarizing vast amounts of scientific literature. This saves researchers significant time and ensures they are up-to-date with the latest findings in their field… AI-powered writing assistants have become more sophisticated, helping researchers draft, revise, and polish their manuscripts. These tools can suggest improvements in clarity, grammar, and style, making the publication process smoother.”
Just remember, as convincing as its answers may be, ChatGPT is not a human, and everything should be fact-checked and validated.
Hint: Ask ChatGPT to find you the top 10 research papers or to identify the top challenges it foresees in your field for inspiration in your literature review.
Enhancing your chat
Want to capture more of ChatGPT’s potential? Check out some of these Chrome extensions that can enhance the bot’s abilities.
WebChatGPT – addresses the model’s inability to access the internet post-2021. This extension adds web results to your prompts to allow for more up-to-date responses, but with the launch of GPT-4o, this tool may be of less use in 2024.
TweetGPT – want to up your online presence? TweetGPT will generate and reply to tweets on your behalf.
ChatGPT Writer – if you’re sick of writing emails, this could be the solution for you. ChatGPT writer is capable of writing emails and messages on your behalf, potentially cutting down on your admin time.
YouTube Summary powered by ChatGPT – perhaps the biggest time saver of all, get ChatGPT to summarise a YouTube video so you don’t need to watch it.
Your virtual lab assistant
As a busy scientist, you might find yourself wishing you had a loyal lab assistant, ready to help at any time. Could ChatGPT fill this role?
One potential solution is asking ChatGPT to find protocols or instructions for certain experiments. By tailoring your prompt to specify exactly what you need and the materials you have available, the chatbot could potentially provide you with a comprehensive, step-by-step procedure. And if you’re unsure of which experiment you should even be trying, ChatGPT might be able to figure it out for you if you give it enough information.
However, as with every application we’ve discussed, there are downsides, and the above applications have had mixed reviews. For example, one social media thread begins with praise for the AI model’s capabilities, whilst many commenters are wary of the answers. But as with many other uses, the answers can potentially be a nudge in the right direction, if nothing else.
Hint: Want to remove some of the less interesting aspects of your work? Get ChatGPT to do calculations for you. One study has already proven that ChatGPT has the ability to solve some biochemistry problems, and it is capable of performing simple mathematics.
An efficient bioinformatician
With scientists now generating terabytes of data in one experiment, bioinformatics has never been more important. But not only are many scientists not trained in the practice, creating an efficient pipeline can be time consuming even for those who have been coding for years.
The use of ChatGPT in bioinformatics is a hot topic online. By providing a comprehensive explanation of your experiment, data types and samples, ChatGPT can suggest the right tools and create a pipeline for analysis. It can even guide you through how to install the necessary packages. In this YouTube tutorial, the user demonstrates ChatGPT’s abilities, but also showcases how prompts can be tailored to combat the bot’s shortcomings.
More specific applications of ChatGPT’s computational power include the ability to annotate sequencing data. Providing the bot with a sequence, plus an explanation of what you are trying to find (ie. specific functional regions of proteins), means it may be able to use its pattern recognition abilities to locate these regions. This could potentially cut down the time it takes to annotate omics data using traditional methods.
It has been reported that GPT‑4o can interpret data charts, microscopy images and gel electrophoresis photos. In one case, it even generated synthetic neurosurgical datasets to augment small clinical cohorts for machine learning experimentation.
Hint: As always, it helps to give the bot something to work with. By telling it what software you have access to, the capabilities of your equipment and the limits of your data, ChatGPT can give you a more tailored answer to your question.
The stats don’t lie
Moreover, ChatGPT might be able to solve your statistical woes. By giving the bot information about your samples and what you wish to find out, it can suggest the correct statistical test to carry out.
“Hey ChatGPT, I have gene expression data for two independent groups with ten samples each. I want to compare them. What statistical test should I use?“
“With independent groups of ten samples each, you can use a statistical test such as the ‘independent samples t-test’ to compare the gene expression data between the two groups. The independent samples t-test is appropriate when you have two independent groups and want to assess whether there is a significant difference in the means of the gene expression between them.“
In the above example, the bot also provided a helpful rundown of how to perform the test!
Hint: To really harness the potential of this application, you should make sure to give ChatGPT as much information as possible in the first instance. Even though the bot will ask follow-up questions to determine which test is appropriate, providing this information at the start of the exchange can be a real time-saver. But if you’re truly lost and don’t know what information is relevant, the bot will guide you.
AI in drug discovery
Drug discovery can be a long and gruelling process. Libraries filled with thousands of small molecules must be assessed before scientists find ‘the one’. Various AI models have recently been used to assist in this process, and ChatGPT has found itself at the heart of many novel experiments.
When fed information about the properties of certain molecules, the model can make inferences about the patterns that are popping up. In doing so, it can foresee which molecules are ‘druggable’ and predict their properties. In a 2023 study, that is not yet peer reviewed, a team of researchers analysed this utility and determined that the chatbot could accurately answer the appropriate chemistry-based questions, and could also locate and produce the necessary files needed for further analysis. For example, it could find PDB and FASTA sequences from online databases, allowing the researchers to take the molecule forward to the next stages of development. It also assisted in the process of molecular docking. Another study explored how drug discovery researchers are customising ChatGPT to more effectively enhance their work.
And some individuals working for drug discovery companies have already lauded ChatGPT’s ease of use: “Instead of clicking and clicking and clicking, you just ask a question and it composes this text that you read and you understand,” said Petrina Kamya from Insilico Medicine, as quoted in a 2023 Nature Biotechnology news article.
Of course, the previous caveats still stand; ChatGPT does not actually know anything, and we shouldn’t risk anthropomorphising the technology. However, while the output currently needs to be validated by a human, many hope that the model can be customised to provide tailored and accurate scientific answers in the not-too-distant future.
Hint: Make sure you ask ChatGPT to point you to the right resources i.e., PDB files. This can cut down on the time it takes for you to reach the next stage in your experiment.
A communication tool
Often, scientific papers can be complex and tough to read, with some reports suggesting that journal articles have become more complicated in recent years. This is in spite of growing calls for accessibility, allowing patients and those without a scientific background to understand the work. For those scientists who can’t resist using jargon, ChatGPT might be able to help. The chatbot has been credited as an effective editing tool, and this extends to scientific papers. A recent study showed that the chatbot could be used as an effective language editor for journal articles. However, it was not successful in writing the article itself, providing false information and imaginary sources. But whilst the final results should always be read by a human, the speed at which ChatGPT can edit the work could transform the writing process.
Hint: Feed ChatGPT some of your own writing so it knows your style. It will then do its best to emulate your work, keeping things consistent
The perfect prompt
How can you engineer your prompt to get the best response out of the software?
- Be specific – ChatGPT has access to a mountain of knowledge, but a vague or generalised prompt can cause it to spit out a number of irrelevant lines of text. If you want something precise, make sure you specify.
- Give context, and your own sources if you can – the bot is great at figuring things out alone, but since it can’t access the internet post-2021, feeding it your own information can be the difference between a false answer and a brilliant one.
- Tell it how long you want the answer to be – often, ChatGPT showcases its extensive knowledge by providing lengthy answers to simple questions. Tell it to keep the answer short, and watch it effectively summarise the point.
- Don’t throw out a bad answer – if the bot doesn’t give you want you want first time round, don’t discount it. Home in on the good parts and ask again, highlighting what you liked and disliked about its previous attempt.
- Be direct! – ChatGPT may give convincing, human-like responses, seemingly engaging in a two-sided conversation, but it’s best to be direct with the bot so that it fully understands your question.
Other tools
As generative AI has become more sophisticated, more tools have been developed. Alongside alternatives to ChatGPT, such as Google’s Gemini, several research specific tools exist such as Elicit (an AI research assistant), Consensus (an AI search engine for research) and Scite (for smart citations).
How to use ChatGPT for research in 2025
Of course, we couldn’t complete this article without asking ChatGPT itself for an overview of how best to use AI in 2025. Here’s what the chatbot itself had to say:
Start with Deep Research — Use for a broad, curated foundation on your topic.
Deep dive with GPT‑4o — Upload charts, code snippets, or protocols for multimodal reasoning.
Validate sources — Use Scite or Consensus to confirm key citations.
Generate code & workflows — Use GPT‑4 Turbo/GPT‑4o for script writing, refinement, and debugging.
Draft and polish — Summarize findings, structure sections, refine language.
Human-check & disclose — Verify every AI-generated insight or reference. Include an AI usage statement in your acknowledgements.
It also suggested different tools to use:
| Tool | Strength | Use Case |
|---|---|---|
| Deep Research | Autonomous multi-step web search | Literature reviews & briefs |
| GPT‑4o | Multimodal reasoning | Data interpretation & coding |
| Elicit, Consensus | Verified summarization | Quick evidence synthesis |
| Scite.ai | Citation context & validation | Checking claims and references |
| Research Rabbit | Literature network mapping | Exploring related work |
| Iris.ai, ChatPDF | Content summarization | Rapid PDF digestion |
| Julius, Copilot | Code & data workflows | Analyses & visualizations |
The caveats
Even with this year’s improvements, the technology is still far from perfect. Misuse, especially in a scientific or medical context, could have grave consequences.
Perhaps the biggest problem with using ChatGPT is its habit of providing false information. But ChatGPT it is not consciously telling lies; it simply operates by carefully choosing the next word in a sequence, emulating the way a human would communicate. It essentially aims to tell you what you want to hear, which can be dangerous when combined with conformational bias.
The model also falls short when asked to provide sources. Numerous studies have suggested that the bot will generate academic citations that do not exist, and in April 2023 The Guardian reported that it was citing fake newspaper articles. Additionally, a study from earlier this year found that when asked for sources, ChatGPT was not able to provide a single legitimate citation.
But, of course, the model is always learning, and in an experiment conducted for the purposes of this feature in its original form in 2023, the bot seemed to have learned its lesson about lying. Access to GPT-4o, even for those using the tool for free, has also helped to mitigate the problem.
Moreover, in a strikingly human moment, the bot cited its own moral code for its reluctance to provide anything fake.
But its self-awareness does not solve all its problems. When talking to the bot, it’s easy to forget that it doesn’t actually know the answers. The developers have been keen to implement these warnings into the model’s behaviour, but this still doesn’t stop us expecting greatness. Maybe the biggest downfall of the technology is our own expectations about what it can do.
What does the future hold?
Although the ways of using ChatGPT we’ve suggested above may help with immediate research, in reality, ChatGPT’s true potential is likely in areas that we haven’t yet considered. So, what are some up-and-coming applications?
One of the most difficult parts of the clinical trials process, and the long-term analysis of a drug even after its approval, is the recording of side effects and drug interactions. Developing new technology to improve this process stems back further than ChatGPT’s existence, with a 2018 study discussing the use of data-mining to gather this information. Notably, this included social media mining, with the authors presenting the idea that trawling through social media content could uncover important health data that may not have been reported otherwise.
Given ChatGPT’s exceptional pattern recognition abilities, could it be used to perform this function? By picking out tweets or Facebook comments that discuss a drug, even in the vaguest terms, there is potential to gather a huge dataset could have previously gone unnoticed. And with current side effect reporting schemes, such as the UK’s Yellow Card system, being historically underused, a more comprehensive option is certainly an appealing prospect.
Elsewhere, researchers have tested ChatGPT’s ability to give medical diagnoses. When provided with descriptions of a patient’s condition, the chatbot was able, in some cases, to suggest an accurate diagnosis. And in an even more recent assessment of the bot’s diagnostic reasoning skills, it landed on the right diagnosis 40% of the time. While it may not seem like a lot, with more training it is thought that the tool could be an effective assistant in the clinic. Researchers have also suggested that the tool could be used for patient monitoring and education.
The rise of AI
Despite some people’s fears about AI, ChatGPT and other forms of generative AI have firmly cemented themselves as the next big thing in tech. Its ability to do things a human cannot, could have us wondering how we ever lived without it.
AI tools have transformed research – from discovery and synthesis to analysis and drafting. With Deep Research and GPT‑4o leading the charge, researchers can save time and explore new dimensions of inquiry. But ethical vigilance, careful prompting, and meticulous human oversight are essential.
Is AI making scientists hyper-productive? Check out this Nature News Explainer from February 2024, which asks whether access to ChatGPT and similar tools causes a negative productivity boost.
