Handling AI-generated applications: Detection, policy and fair review

A recent study by the Northwestern Innovation Institute found that AI-assisted grant applications had higher approval rates and led to more publications, but not to more breakthroughs. On the contrary, they tended to reproduce existing funding patterns rather than introduce new ideas. That should give grant managers pause.

As a team working closely with content and grant applications, we have been noticing for some time how AI is changing the way we write. Certain phrases, a kind of polished eloquence without real substance behind it… you develop a sense for it. But that holds a very different weight for grantmakers.

When an application sounds compelling but doesn’t reflect the actual ideas or competence of the person behind it, the impact of the funding is at risk.

When technology shifts the playing field

Grants programs are built on a promise: to support the best ideas and the most suitable applicants. That promise comes under pressure when large language models can generate structured, convincing applications in seconds.

The technology itself is not the real problem. Many applicants use AI tools for legitimate purposes, such as translation or organising their thoughts. The difficulty arises when entire applications are generated with no real substance behind them. Grant managers and reviewers then face the hard task of identifying deception without treating every honest applicant as a suspect.

There is also a structural issue, as AI tools are not equally accessible. Applicants with better technical resources or more experience using AI have an advantage, regardless of the quality of their actual work. What is intended as a helpful tool can become yet another driver of inequality.

Clear guidelines build trust on all sides

Those who act early have a real advantage. Programs that engage with the issue openly and transparently build trust with applicants, reviewers and funders alike. Leading funding organisations around the world have recognised this and are already taking action.

A joint statement from the Research Funders Policy Group sets out clear expectations: AI tools may be used in a supporting capacity, for example for language improvement or formatting. Generating entire applications without meaningful human involvement is not permitted. Reviewers must not input confidential applications into publicly available AI tools. And funding decisions remain in human hands at all times.

This shows that AI does not need to be rejected outright. What matters is defining responsible use and communicating it transparently.

Effective implementation requires well-trained reviewers and technical support that work in concert. And because reviewing applications or requesting evidence of authorship involves sensitive data, privacy should be a central consideration from the start. A platform built with that in mind makes the whole process significantly easier. Read about how Good Grants approaches this here: Good Grants introduces secure, privacy-first AI tools.

Practical steps for grant managers

How do you approach this in practice? The following steps have proven effective.

1. Set clear guidelines and communicate them actively. Define whetherAI tools are permitted and make this clearly visible in your application materials. Applicants can only follow rules they know about. Draw on established standards such as the Research Funders Policy Group joint statement, and adapt them to your specific funding context.

2. Train your reviewers. Typical signs of AI-generated text include polished, structured language combined with a lack of real depth, repetitive sentence patterns such as “it is not X, but Y”, or points listed in groups of three.

Important: a single characteristic is not evidence. Closer scrutiny is only warranted when multiple indicators appear together. After all, AI language models learned their style from humans and use stylistic devices we ourselves established. In my own writing, I tend to list things in threes. That does not make me a language model.

3. Use detection tools as one input among several. There are tools that can assess whether a text is likely to be AI-generated. They return probabilities, not proof, so treat them as one signal among several rather than the basis for a decision. The same applies to plagiarism checking, which many grants programs already have in place. The systematic approach translates well to reviewing AI-generated content.

4. Invite clarification rather than rushing to reject. When a submission raises concerns, give applicants the opportunity to explain or provide evidence of their work. This protects against misjudgements, respects the dignity of applicants and strengthens the credibility of your assessment process.

5. Build equity into your approach. When developing AI use policies, keep in mind that not all applicants have equal access to AI tools or the same level of familiarity with them. A blanket prohibition can also disadvantage those who legitimately benefit from language support, such as multilingual applicants. Nuanced rules are fairer than blanket bans.

6. Document your decisions. Record the reasons why an application was flagged for review or rejected. This creates accountability, protects your program if questions arise and helps you refine your criteria from one funding cycle to the next.

7. Configure your platform to match your requirements. Every grants program has its own needs when it comes to review processes, approvals and data privacy. A configurable platform lets you build the workflows you need, such as multi-stage reviews or targeted follow-up questions to applicants, without losing control of sensitive data.

Funding that reaches the right people

AI-generated applications will continue to be a challenge for grants programs. The question that matters is not whether AI is being used, but whether funding reaches the places where genuine ideas and genuine need exist. 

Grant managers who establish clear guidelines now, train their reviewers and invest in a reliable, privacy-compliant platform will be well placed to respond. Good Grants supports grants programs around the world in answering exactly that question, with solutions that put security, privacy and fairness at the centre. Want to see how that looks in practice? Try Good Grants free for 14 days.