How to use AI responsibly in writing research proposals

Sara Diegoli
Feb 8
6 min read

Updated: Feb 15

This article explains how to use artificial intelligence (AI) in research proposal writing in a strategic and responsible way, focusing on practical methods, risks to avoid, and how researchers and organisations can integrate AI without sacrificing quality, judgement, or ethics. This is based on personal experience and conversations with peers in our academic community.

You don’t use AI for proposal writing. The simple thought offends you.

Until one day….you do.

The deadline looms. The guidance is dense. The cursor is blinking and almost mocking you. So, purely hypothetically, after checking over your shoulders in the open office, you open ChatGPT….

If this hasn’t happened to you yet, it probably will. And if it has, welcome friend, you’re not alone! This moment captures the reality of how many researchers are already using AI in proposal writing, often informally, under pressure, and without a clear strategy.

Let’s talk about how to use AI strategically in proposal writing — to lower cognitive load and speed up specific tasks, without outsourcing judgment, originality, or responsibility.

Not another prompt engineering piece, please…

There is no shortage of blogs and opinion pieces about “prompt engineering”. Call me when someone finds a prompt so well engineered that it will make ChatGPT actually intelligent!

Much of the prompt engineering advice focuses on the wrong problem: how to get AI tools to produce better text, faster. While that can be tempting, it risks missing the more important questions:

What kind of problem should we delegate to AI when writing proposals?

What should instead remain firmly human?

Because trust me, artificial intelligence, despite its name, is not that “intelligent” at all!

A strategy-based view of research proposal writing

So many of the researchers I work with forget that proposal writing is much more than “writing”. I know, it sounds counterintuitive, but stay with me.

Before we start writing, we need to understand the call, timelines, assessment criteria, etc. I sometimes can spend quite a long time making tables and thinking what structure I might want to use for the case for support, what evidence I might want to gather to show the excellence of my team or how I can best frame the problem my proposal is going to solve to match the call. These are cognitively demanding tasks: analysing information, structuring arguments, making strategic choices and…. eventually…. drafting text.

Rather than treating AI as a shortcut to writing, I would reframe AI as a tool to reduce cognitive load when wading through guidance, make implicit assessment logic clearer, help extract timelines, create checklists, suggest headings to organise my case for support to map onto the assessment criteria.

This approach is particularly helpful for researchers who tend to skip this initial analysis and rush in to produce lots of words (or copy and paste recycled text from a paper or previous proposal). You know who you are.

The result is often a proposal that starts twice as long as it should be and is then cut to the word limit. This approach is rife with issues, involves a lot of wasted time in subsequent rewrites, and often resorts to squeezing a sentence into the word count to address a criterion that probably deserved half a page. This is grant writing done backwards and is often a “survival” adaptation in response to funding schemes that are becoming more and more complex.

What are the best ways to use AI in proposal writing?

One of the strongest and safest uses of AI in proposal writing is analysing funder information.

Many calls provide detailed section-by-section guidance, alongside assessment criteria that apply to the proposal as a whole or cut across multiple sections.

One of the most useful applications of AI is extracting assessment criteria and mapping them to the different proposal sections. This “criteria-to-section mapping” can be really helpful to reduce the risk that important criteria are addressed weakly, implicitly, or not at all.

Using AI to structure your case for support

AI is also effective at helping structure your case for support, where funders might provide minimal guidance on which headings and subheadings to use. You can leverage AI to suggest a sub-structure that is aligned to assessment criteria to organise your content more strategically.

Deploying AI to extract timelines and checklists

AI can also extract timelines and even checklists to double-check your own understanding of the guidance and submission process and to help you stay organised. It is important to note that AI should not be trusted as a shortcut to avoid reading the guidance (as much as I wish that was possible) but rather as an extra pair of eyes to make sure you did not miss anything.

Using AI in “reviewer mode”… if only reviewers followed the rules

Another valuable use of AI is asking it to act as a “reviewer”, within reason! This should never be confused with peer review or expert assessment — it is a tool for presentation, not evaluation.

In this role, AI can help in surfacing weaknesses in the presentation and articulation of your idea. What AI cannot do is offer any insight whatsoever into the quality of your project. This distinction is really important. AI cannot judge novelty, scientific quality, or funding likelihood. It does not understand disciplinary politics, panel dynamics, or strategic priorities.

The real danger with AI is that it will very confidently express opinions on things it does not understand at all, and is equally happy to just make it up. That is why you need to always be very vigilant when using it in reviewer mode and make sure you remain in charge of the thinking.

The value lies in providing you with a reflection point to help you present your idea more clearly, spot any inconsistencies in your logic, and flag areas where you have not addressed the assessment criteria. AI will not make your project or idea intrinsically better.

Refining abstracts and titles

Abstracts and titles are another area where AI can be helpful — if used carefully — and surprisingly fun in all the wrong ways.

AI can be used effectively to write a first pass abstract, reducing word count hopefully by still preserving meaning. What you should always watch for is the introduction of new claims and impacts that are not supported elsewhere in the proposal and the introduction of hype.

I find that providing AI with a framework that I like for the abstract can guide the tool in the right direction. Try, for example, asking AI to structure your proposal abstract around our Grant Lean Canvas and see the results improve. Substitute that with your favourite structure until you find something that helps you do that first draft. That is sometimes half of the battle!

Brainstorming without outsourcing judgment

AI is also useful in structured brainstorming, particularly where breadth of perspective matters.

This includes:

identifying potential risks to delivery,
mapping stakeholders and engagement pathways,
and supporting Responsible Research & Innovation (RRI) by scanning for possible unintended consequences of technologies.

In these cases, AI works best as a horizon-scanning tool — surfacing categories of risk, affected groups, or downstream considerations that might otherwise be missed. If you have been in a brainstorming session, you will have heard the facilitator say: “In the first instance, we just want to generate as many ideas as we can”. There is nothing AI is better at than providing really long lists of stuff!

Leverage that time-saving power to co-generate lots of ideas, and then you decide what to keep. Crucially, AI should not be involved past the first stage of idea generation. Responsibility for evaluating the ideas remains firmly with the researcher.

My two pennies’ worth

My overall message is simple but important: AI is most powerful in proposal writing when it supports analysis, structure, mapping and in brainstorming, where quantity, not quality, is important — not when it replaces expertise or judgement.

Used well, AI can save time and become a useful tool, particularly for those who have not been socialised into the “hidden curriculum” of funding applications.

Used badly, it can flatten ideas, inflate word count, inflate claims, and obscure responsibility. The difference is not technical skill in prompt engineering, but professional judgement about what thinking can be shared, and what must remain human.

That judgement — and the accountability that comes with it — always stays with the applicant. In short, AI can support thinking in proposal writing, but it cannot replace responsibility for decisions.

If you have any questions, just reach out to me @ hello@anchoredin.uk and ask for Sara. I know you are just asking for a friend!