Back to blog

The Ethics and Uses of AI in Writing R&D Tax Credit Reports

Against a backdrop of increased scrutiny of R&D tax credit claims, a fascinating development has emerged: HMRC has been ordered to disclose whether it uses artificial intelligence (AI) to review R&D tax claims, following Freedom of Information requests that have sparked considerable debate.

The irony is striking. While there's widespread concern about HMRC potentially using AI to assess claims, businesses are increasingly turning to the same technology to prepare their submissions. Some online tools now explicitly advertise AI-powered support for R&D report writing, and, let's be honest, few of us can claim we've never used ChatGPT to polish an email or generate ideas at work.

This raises crucial questions: Where do we draw the ethical line? How do we balance efficiency with accuracy? And what does responsible AI use look like in the context of R&D tax credits?

The Growing Role of AI in Technical Report Writing

Generative AI chatbots like ChatGPT, Gemini, Claude, and a whole host of industry-specific platforms have revolutionised how we approach technical writing. These tools fall under the large umbrella referred to as “AI” within this article.

For R&D tax credit claims, the appeal is obvious. These tools can quickly turn lengthy technical experimentation into coherent narratives and dramatically reduce the time spent on initial drafts.

Particularly for smaller businesses without dedicated tax teams, AI tools offer significant benefits when writing a technical narrative. This is a critical part of the claim now that HMRC requires claimants to submit an Additional Information Form, which must include key R&D project information. Large Language Models (LLMs) serve as a valuable starting point for what can otherwise feel like a daunting task.

The key phrase here is "starting point." When used appropriately, AI can speed up the initial drafting process, helping businesses present their scientific and technological advances, uncertainties and activities more clearly. It's the difference between staring at a blank page for hours and having a foundation to build upon.

Accuracy vs. Efficiency: The Trade-off

However, speed comes with significant risks. AI systems are notorious for "hallucinations", generating plausible but factually incorrect information. In the context of R&D tax credits, where technical accuracy is paramount, these errors can be catastrophic.

There is an obvious difference between a human-written narrative that explains failed experiments and breakthrough moments, and an AI-generated report filled with generic statements about "innovative solutions" and "cutting-edge technology." The former demonstrates genuine R&D activity; the latter reads like marketing copy.

More concerning is the temptation to use AI to "improve" claims by making them sound more qualifying. We've seen businesses use AI tools to enhance their reports to "make them more likely to be approved by HMRC." This fundamentally misses the point; your report should describe what you actually did, not what you think HMRC wants to hear.

Tell-tale signs of AI use in R&D tax credit claims include regurgitating HMRC’s guidelines without substantiating them. A poorly written argument for an eligible project includes lots of buzzwords from HMRC’s criteria without presenting evidence for it:

“Our company sought to develop a new engineering process that would be an advance in the field beyond the technological baseline that existed in the sector; this process would use cutting-edge technology to greatly improve the output of the process and overcome significant technological uncertainties to do so.”

Take the widely discussed Dundee United case, where the technical report circulated online consisted of basic, template-like sentences claiming eligibility without any substantive explanation of why the work qualified. Whether generated by AI or copied from templates, such lazy compliance approaches are easily spotted and inevitably rejected.

The Ethics of AI in R&D Claims

When businesses use AI to craft eligibility arguments without fully engaging with the underlying technical work or qualification criteria, are they any different from those who simply copy and paste from templates?

The R&D tax credit scheme requires "competent professionals" to confirm that qualifying R&D activities took place. This is a professional judgement about whether work meets specific legislative criteria, from someone who is an expert in the field of science or technology.

When AI becomes the primary of technical narratives, we risk undermining this principle.

Using AI tools to manufacture qualifying narratives is, at best, lazy and misleading and, at worst, constitutes a negligent or fraudulent claim.

Confidentiality and Data Security

Beyond accuracy concerns lies another critical issue: data security. R&D projects by nature often involve proprietary technologies, trade secrets, and commercially sensitive information. When businesses input detailed technical descriptions into third-party AI platforms, they may inadvertently expose intellectual property.

Most AI platforms retain and learn from user inputs, potentially making your confidential R&D information part of their training data. This raises serious questions about IP protection and GDPR compliance.

For businesses in highly competitive sectors, the efficiency gains from AI assistance might not justify the security risks.

This is one of the core arguments against HMRC potentially using AI to review claims; companies only agree for their data to be reviewed internally by HMRC, not for their technical know-how to be shared with unknown third parties with potentially liberal data protection policies.

The distinction is that companies can choose to share their own project details with third parties, but they do not agree to this when sharing with HMRC. However, with the variety of data retention policies across LLMs, sharing your proprietary information should be questioned internally, too.

HMRC's Position: A Question of Fairness

The public reaction to HMRC potentially using AI for claim reviews reveals an interesting double standard. Many of the same businesses criticising HMRC for assessment through algorithm are simultaneously embracing AI for claim preparation.

The truth is, if AI can help businesses prepare claims more effectively, can’t HMRC use similar technologies to process and review them more efficiently? The key issue is how it's used and whether human expertise and judgement remain central to the process.

AI could be a powerful tool for sorting thousands of R&D tax credit claims into risk categories, offering a “first draft” of a review. This is assuming that data governance and human oversight are embedded into the process.

But relying on the decisions of an algorithm (and potentially preparing communications with clients, as one R&D tax advisor has suspected) to classify whether a claim qualifies goes too far. It risks wasting the time of eligible claimants who need to defend their projects against a decision that was not made by a human.

Responsible Use of AI in R&D Tax Credit Claims

So, how can businesses use AI responsibly in R&D tax credit preparation? The answer lies in treating AI as a drafting aid, not a replacement for human expertise.

Here's our recommended approach:

• Use AI to structure and organise, not to make eligibility judgements. AI can help you outline your projects and identify key points to address, but it shouldn't determine whether work qualifies for relief.
• Always collaborate with your technical teams. The people who actually conducted the R&D work should be central to the claim preparation process. AI can help translate their technical expertise into accessible language, but it cannot replace their fundamental understanding of what was attempted and achieved.
• Cross-reference everything with the legislation. AI-generated content should always be checked against the specific R&D tax credit criteria. Generic statements about innovation need to be grounded in the actual legislative requirements.
• Anonymise your work where possible. Better safe than sorry – don’t include valuable IP in your chatbot prompts. You never know what could change with a third party’s data governance policy.

The key message here is that a longer preparation time is infinitely preferable to a investigated or rejected claim.

Moving Forward with AI and R&D Tax

The future of R&D tax credits will likely involve AI on both sides of the process; businesses will use it to prepare claims and HMRC will use it to review them. The key is ensuring that human judgement, technical expertise, and ethical responsibility remain at the centre of both processes.

The businesses that thrive in this new environment will be those that use AI to enhance their capabilities while never losing sight of the fundamental purpose of the R&D tax credit scheme – rewarding genuine innovation and technical advancement that benefits the UK economy.

If you're navigating the complexities of R&D tax credits in an AI-enhanced world, we're here to help. Our experienced team combines deep technical knowledge with the latest tools and best practices to ensure your claims are both efficient and compliant.

Get in touch to discuss how we can support your R&D tax credit strategy while maintaining the highest standards of accuracy.