Can ChatGPT write my research proposal?

Disclaimer: This blog post was written at a time when ChatGPT was still very new and exciting, and the world was still getting to grips with its capabilities and their implications. There will come a time – probably quite soon – when everything written below will feel as dated as an article about the miracle of powered flight.

Unless you’ve been living in a cave for the last few months, you’re likely to have come across ChatGPT. Built on a fairly* ground-breaking conversational artificial intelligence (AI) system, ChatGPT uses a large language model (a probability-based deep-learning algorithm that determines how likely a given word is to follow – or to come before – another in a particular sequence), trained on a huge body of existing text, to produce written responses in an interactive dialogue format. This essentially makes it an AI chatbot, albeit a pretty smart and very capable one. It can write stories, put together a decent-ish undergraduate student essay, compose music and even write and debug computer code. Because it was trained on a large chunk of the internet, including articles, books and English-language Wikipedia, it has an impressive ‘general knowledge’ because it ‘memorised’ a huge volume of facts during its training. Fairly predictably, the sudden appearance of ChatGPT and the dawning inevitability that its offspring and successors will be even more impressive has been met with reactions that range from wonderment to dismay. And, while some people who work in content-creation roles are looking anxiously over their shoulders and wondering how long it will be until ChatGPT or something similar pinches their job, others are looking to the technology with a view to making their lives more productive and perhaps a bit easier. Leaving aside the blatant cheating exemplified by the hung-over social sciences student** who needs an essay on the political views of Karl Marx by tomorrow morning, is AI in general and ChatGPT in particular a tool that we can legitimately use in academia to increase our productivity and improve the quality of our output? More specifically, can we use ChatGPT to write research proposals?

I think the answer is yes. And no. To explain: Do I think that ChatGPT can write your research proposal for you? No, I don’t. But do I think ChatGPT is a useful tool that could be employed as part of the grant-writing process? I do.

So how does it perform?

There’s no doubt that ChatGPT has a basic 'understanding' of what some of the main constituent parts of a research proposal are. When I asked it to “write a research proposal about developing a general vaccine against cancer”, it gave me a title ("Development of a General Vaccine Against Cancer: A Proposal for Research" – okay, not particularly inspired); a neat little introduction that told me a bit about what cancer is, the limitations of current treatments, and the general benefits of vaccines; an objective (which, if I’m quibbling, I’d say looked more like an aim); some very high-level methodology; and a kind of impact summary (as in, "a vaccine for cancer would be great!" – it’s admittedly not too much of a challenge to describe the would-be benefits of a cancer vaccine). In other similar exercises, it came up with some research questions and a list of objectives that actually looked like research objectives. The proposal itself was very short, not specific to any funder or scheme, and wouldn’t have been funded in a million years. But what did I expect? After all, I gave it very little to go on, and I asked it to come up with a research project to achieve something that’s not currently possible.

Being realistic

An eighteenth-century cookbook is reputed to have included a recipe for a pie that began “first catch your hare”. This, I think, characterises the scale of the challenge that’s asked of the research-proposal writer, be they a human or an AI. I’ve always maintained, not very controversially I like to think, that there are two key parts to producing a research proposal: designing and planning the proposed project; and writing about it in such a way as to sell it to a funder. They should never be tackled concurrently, and it goes without saying that the latter is only possible once the former has been completed. And, actually, it’s the former – designing and planning the project – that’s the really tricky bit. It requires pre-existing expert subject knowledge, original and insightful thinking, the skill of ideation and no small measure of inspiration, the ability to look beyond the state of the art to develop novel approaches, an appreciation of what’s possible and of what’s available in terms of resources, the ability to marshal those resources and assemble and co-ordinate a project team, along with an in-depth knowledge of the specific funder’s remit and priorities and an understanding of how the particular funding scheme is positioned.

All of this represents the substantial hare-catching exercise that must be undertaken before the proposal can be written. And it’s quite a challenge, even for a seasoned professor let alone a language-model chatbot. The best proposals that I see tend to be the ones where a whole load of behind-the-scenes work in terms of designing and planning the project has been done before a single word of the proposal was ever written. They’re great proposals not necessarily because the applicants are great writers, but because they are strategic thinkers who are capable of developing great ideas, and are really good at assembling a project and team around those ideas. Once all of that’s done – and I mean really done, with everything down to the last detail properly planned out – then it shouldn’t actually be too hard a job to describe it all in a research proposal. Sure, there’s a selling job to be done, an angle to be found and communicated, but that’s actually quite formulaic. And that may be where ChatGPT comes in.

At this point, it’s important to remind ourselves of the old computing acronym, GiGo – garbage in, garbage out. If we just feed ChatGPT with garbage – be that unrealistic expectations of an anti-cancer vaccine, vagueness and hugely-insufficient detail, or a combination of both – then we’re not going to get very much of use back out of it. We can't realistically expect it to undertake a large and complex writing exercise on the basis of a single, information-poor prompt, and get it anywhere near right first time. Moreover, we can’t expect it to have great and original ideas on highly-complex topics on our behalf. (It may be possible to take an algorithmic approach to generating genuinely competitive new proposals based on previous grants and some very specific parameters fed in by a human operator, but I’m still not sure that could really be described as original thinking.) This point is important. When I asked ChatGPT to write a cancer-vaccine proposal, it proposed as part of the approach identifying tumour-associated antigens and using advanced genomic and proteomic techniques. I didn’t include any of these words in the short instruction that I provided, so on the face of it ChatGPT would seem to have a combination of a vast amount of specialist knowledge, and an ability to think and thereby apply that knowledge in an intelligent way. It has indeed assimilated a large amount of knowledge (although it doesn’t currently have real-time access to the internet), but we need to remember that it has simply learned how to string together words in a way that resembles how the people who wrote the words it trained on typically put those words together themselves. So the word ‘antigen’ commonly appears in texts about vaccines, and words like ‘genomic’ and ‘proteomic’ may often appear alongside phrases like ‘cancer research’. ChatGPT is putting them all together in a way that appears uncannily like it’s intelligently and creatively generating research ideas of its own, but it’s not engaging in original thinking.

Nevertheless, what we can do is provide ChatGPT with richer input, and work with it iteratively to develop and refine its output. We’re moving away here from full ‘sit-back-and-relax-while-ChatGPT-writes-it-for-you’ automation, but I think there’s still real potential for the technology to provide us with a useful tool. Your hedge trimmer won’t cut your hedge on its own, but it sure does come in handy when you need to tackle the privet topiary – so long as you know how to use it (and how to use it safely).

ChatGPT as a tool, in the hands of the (human) user  

Firstly, it’s fair to say that ChatGPT is a good editor. If you ever struggle to write
concise, clear and compelling text that’s easy to read and puts across your ideas succinctly, then consider running chunks of it through ChatGPT. I’m quite often faced with proposals that contain some very lengthy sentences, where the writer sometimes appears to have got a bit lost mid-way through and descended into confused and confusing narrative that’s really hard to follow. Such proposals are often riddled with inconsistencies, strange grammar that not only irritates the grammar-pedants but also compromises readability, and minor typos that smack of a job half done. It’s always difficult to be really critical of one’s own written output – after all, you know what you meant to say – and it takes a lot of time for someone else to tidy it up. But ChatGPT writes good, clear and well-ordered English, and it can tidy your text up almost instantly. This might be particularly useful if you’re not a first-language English speaker. I took a paragraph from a previous (unsuccessful) proposal written some years ago by just such an applicant, and asked ChatGPT to edit it for readability. The result really was much improved. So consider putting bits of your proposal (perhaps a paragraph, or a few paragraphs, at a time) through ChatGPT with the instruction to edit them for readability or simply to improve them. Bear in mind that the system is (by the designers’ own admission) by no means infallible and could make mistakes of omission, deletion, misinterpretation, misrepresentation and so on. So do check all of its output carefully to make sure you’re happy with it in every way. Just the same as how you’d check whatever came back from any co-applicant, peer reviewer or copy-editor before incorporating it into the latest draft of your proposal.

Secondly, many of us are susceptible to the tyranny of the blank sheet of paper. Getting started on writing anything can feel daunting, and it’s often important just to get something down, even if it bears little or no resemblance to what will eventually become your final draft. With the right instruction, ChatGPT could write you a decent introduction to the topic for your background section. You’ll want to check that any facts and figures included are authoritative and up to date, and you may need to develop the focus in a particular direction (or get ChatGPT to do this). But it will give you a starting point, and something to work with. And even if you hate it, seeing a concrete example of what you don’t want can sometimes help you to identify what you do want.

For example, I asked (told? I try to be polite…) ChatGPT to “write an introduction for a UK research proposal about bowel cancer”. It spat out a few paragraphs, the first of which gave some useful general background. I continued the conversation and asked it to “expand the first paragraph in the above” – and hey-presto it did. Nothing to set the world on fire, but a decently-structured introduction with some useful facts and figures. Don’t expect it to show much in the way of insights as to what a particular funder or funding scheme will be looking for. I asked it to optimise the text it had produced for the UK’s Medical Research Council and it had a go, but with no obvious improvement evident in that respect.

A good title, sometimes with an accompanying acronym, is a small but essential part of a compelling proposal. And here again ChatGPT can help out. When given a research-project summary and asked to come up with a title, it did so. The result was a bit wordy, but then you can always ask it to produce a shorter one. Asked for an acronym it again came up with the goods – a bit naff at its first attempt, but better when asked to try again. An iterative ‘conversation’ with ChatGPT quickly refined the quality and suitability of its outputs. At the very least, you could use ChatGPT to generate some ideas for a title and acronym.

As noted, ChatGPT doesn’t do original thinking. But summaries don’t require original thinking, just the ability to condense the key points of some text down into a shorter form. ChatGPT seems pretty good at this. Give it the text you want to summarise, and a word-count limit, and see what you get back. If necessary, ask it to edit to make the summary more accessible to a lay reader. If it has de-emphasised or ignored key points then, once again, use further instructions to iteratively refine and improve. As with pretty much everything you get ChatGPT to write for you, treat the output as no more than a working draft and, at the very least, read it carefully before pasting it into the grant-application form. As well as checking for errors and omissions, keep an eye out for things that look like they’re out of date, and also for generic ‘fluff’ and waffle that’s more or less devoid of any meaning (many humans are guilty of writing this stuff too!). You’ll almost certainly get the best results by working collaboratively with ChatGPT and complementing its output with your own additions, edits, revisions and enhancements.

A vital part of any research proposal is a list of specific research objectives. Reviewers will expect them to be relevant to the project’s aim, achievable within the time-frame of the project, and readily measurable. Defining the project objectives can sometimes seem quite challenging. I found that ChatGPT was able to produce a list of mostly fairly credible-seeming research objectives when I fed it some text about the proposed approach and methodology. This is backwards-engineering of course – methodology should be designed with reference to the research objectives – but it may serve as a useful sanity check to ensure that everything in the methodology and objectives is relevant, appropriate and shows clear and logical mapping between these two key aspects of the proposal.

Is it cheating?

And so on. There will be ways of using ChatGPT as a tool to improve your proposals that I haven’t even thought of yet. But is all of this in some way unethical and immoral – cheating? Thinking back to our hung-over student, the essay is a test of their ability to find and gather relevant facts and information, marshal their thoughts, and produce some structured, thematic text that holds everything together in a logical manner. They will be marked and graded on their ability to do this. If someone or something has substantially done this for them, then the grade awarded won’t reflect their own abilities and will thus have been fraudulently obtained. In contrast, undergraduate students are not normally being tested and assessed on their ability to spell, so no one will object to their having used a spell-checker tool.

A research proposal, on the other hand, is fundamentally a marketing document for a research project. It will be graded by reviewers and panel members, but not of course with a view to awarding the applicant (or not) with some sort of qualification. It is not the piece of writing that’s being graded, but rather the importance, novelty, feasibility and overall excellence of the proposed research. Sure, a badly-written proposal may result in a poor score – it’s well-nigh impossible to review something that’s very difficult to follow and understand, and the benefit of the doubt is seldom given. But panels will also give low scores to well-written proposals that are simply lacking in novelty, for example, or that seem to be methodologically unfeasible. And of course the grading exercise here is simply a means of prioritising the best, most important and most relevant research ideas for funding. Funders just want to commission the research that’s most likely to support their own remit, vision, and strategy, and I don’t think they’re really too fussed about exactly who wrote the proposal and how***. So, provided the funder hasn’t updated their guidance to include a ban on using AI tools, then it’s surely in no way cheating.  

Me … or ChatGPT?

You might legitimately be wondering whether I actually wrote this blog post myself, or whether I had ChatGPT write it for me. And the thought certainly did occur. I typed “Write a blog post about using ChatGPT as a tool to write research grant applications” into the text-input box, and it spat out around 450 fairly anodyne words setting the scene, describing itself, and carefully caveating the piece by reminding me that it’s just a tool, that needs to be used judiciously with proper human supervision and input. But it did describe some concrete ways in which it could be of use to the grant writer – specifically by helping to draft a compelling introduction, by producing a rough draft of sections to be refined, and by improving the style, grammar and clarity of something that’s already written while providing feedback on how it has done this. In the interests of openness, I’ve included the complete text of what ChatGPT produced at the end of this post. Take a look and see what you think.

* I say ‘fairly’, because ChatGPT didn’t appear out of nowhere – it has predecessors and ‘older cousins’ that are also very impressive. But ChatGPT is arguably the first AI of its kind and on its scale to be unleashed on the public at large as an interactive AI tool that is, transparently, just an AI tool to play with.

** Other examples of cheating in different academic disciplines are available.

*** So long as, at its core, it's essentially the intellectual work of the applicant team. The partial exception here is fellowship proposals. Here, the individual is being assessed as well as their research project – but once again they’re not being assessed on their ability to write but rather on their ability to formulate a research idea and plan a project around it. So while the project and the ideas behind it must demonstrably be their own, I don’t think many grants panels would object to some support being obtained – from whatever source – with the process of writing about them.

Here’s ChatGPT’s attempt at this blog post. I asked it to write a blog post based on the subject that I gave it, and then I gave it back the text of its blog post and asked it to come up with a title for it. To be fair, I didn't spend very much time on the iterative-refinement process that I advocate above. It all seems fine to me, if a bit vanilla. I don’t disagree with any of it.

Streamlining Research Grant Applications: How ChatGPT Can Help Researchers Write Stronger Proposals

As a researcher, one of the most important skills you will need to develop is the ability to write a strong research grant application. A research grant application is your opportunity to convince funders that your research is important and worthy of support. It is a challenging task that requires careful planning, writing, and editing. Fortunately, there are tools available that can help make this process easier and more efficient. One such tool is ChatGPT.

ChatGPT is a large language model developed by OpenAI. It has been trained on a vast amount of text data, making it capable of generating natural language responses to a wide range of prompts. This makes ChatGPT an ideal tool for researchers who need help with writing research grant applications.

One of the most useful features of ChatGPT is its ability to generate text based on a specific prompt or topic. For example, if you are struggling to come up with a compelling introduction for your grant application, you can input a brief summary of your research topic into ChatGPT and ask it to generate an introduction for you. The resulting text can serve as a starting point for your own writing or as inspiration for further development of your ideas.

In addition to generating text, ChatGPT can also provide feedback on your writing. By inputting your grant application text into the tool, you can receive suggestions on how to improve your writing style, grammar, and clarity. This feedback can be invaluable in helping you craft a polished and effective grant application.

Another benefit of using ChatGPT is that it can help you save time. Writing a research grant application is a time-consuming process, and using ChatGPT to generate text and provide feedback can help streamline the process. By using the tool to generate a rough draft of your application, you can focus your time and energy on refining and improving your ideas.

Of course, it is important to note that ChatGPT is not a substitute for careful research and thoughtful writing. It is merely a tool that can assist you in the writing process. As a researcher, it is still your responsibility to conduct rigorous research, develop a clear and compelling argument, and present your ideas in a well-organized and coherent manner.

In conclusion, ChatGPT can be a valuable tool for researchers who are writing research grant applications. By generating text, providing feedback, and saving time, ChatGPT can help researchers produce high-quality grant applications that have a better chance of receiving funding. However, it is important to remember that ChatGPT is only a tool and should be used in conjunction with careful research and thoughtful writing. 

The views and opinions expressed in this blog are mine alone and are in no way endorsed by my employer. Factual information and guidance are provided on a 'best-endeavour' basis and may become out of date over time. Web-links were correct at time of writing but commonly go out of date. ChatGPT/OpenAI logos and branding remain the intellectual property of their owners. No responsibility can be taken for any action or inaction taken or not in respect of the content of this blog.