R&D tax credit claims – should you do it yourself?

The R&D Corporation Tax Credit is valuable. For accounting periods commencing on or after 1 January 2026, the credit has increased from 30% to 35% of qualifying expenditure.

So it is understandable that some businesses choose to prepare their R&D tax credit claim themselves. And, increasingly, there is another option: ask AI to help. AI can explain the rules, suggest how a project might be described and turn a few notes into an impressively polished technical narrative in seconds – but there is an important distinction:

Producing an R&D tax credit claim is relatively easy. Producing a claim that accurately reflects the work undertaken, captures the right expenditure and can withstand scrutiny is considerably harder.

→  Here are 10 things to think about…

1 | Do you really know what qualifies?

Perhaps the biggest risk with a self-prepared claim is starting with the wrong definition of R&D.

Revenue’s criteria are quite specific. Qualifying activities must be systematic, investigative or experimental, take place in a field of science or technology, seek scientific or technological advancement and involve the resolution of scientific or technological uncertainty.

Something can be innovative, technically difficult, expensive or completely new to your business without necessarily satisfying those tests. The opposite can also happen. Work that seems relatively routine to the people undertaking it may contain qualifying R&D.

Knowing where that boundary lies is one of the most important parts of preparing a robust claim.

2 | You may claim too little

The most obvious perceived risk of preparing your own claim is claiming too much. But underclaiming can be just as significant.

Businesses often identify the obvious flagship R&D project while overlooking qualifying work taking place elsewhere. A manufacturer might focus on developing a new product but overlook experimental work around production processes, materials, automation or testing.

A software business might identify a major new platform but overlook qualifying technical challenges encountered in integration, architecture, scalability, security or performance.

Without experience of looking for qualifying R&D across different activities, it is easy to leave legitimate expenditure unclaimed – and with a 35% credit rate, that can be costly.

3 | Or you may claim too much

The opposite problem is equally important. When you are close to a project, it can be difficult to assess objectively whether the work genuinely satisfies Revenue’s criteria.

X   A challenging project isn’t automatically R&D. Nor is developing a new product, introducing new software or improving an existing process.

✓  A robust claim needs to distinguish between the qualifying R&D and the commercial, routine or implementation activities surrounding it.

Getting that boundary wrong can inflate a claim and increase the risk if Revenue examines it.

4 | AI can write a convincing story. But is it your story?

This is becoming an increasingly important issue. Give an AI tool a description of a project and ask it to explain why the work qualifies for the R&D tax credit, and it can produce something that sounds remarkably convincing.

It may talk about technological uncertainty, advancement, experimentation and systematic investigation. The problem is that the language sounding right doesn’t mean the claim is right.

AI doesn’t know what actually happened within your project unless you provide it with accurate and sufficiently detailed information. It can make assumptions. It can overstate the technological challenge. It can introduce activities that didn’t happen. And it can turn fairly ordinary development work into language that sounds much more like qualifying R&D.

That creates a dangerous situation: A technically impressive narrative that the people who actually carried out the work may struggle to substantiate.

5 | AI-generated evidence isn’t contemporaneous evidence

This distinction is particularly important. Revenue expects companies to maintain appropriate records supporting their R&D claims. Its guidance emphasises the importance of contemporaneous evidence demonstrating both the qualifying activities and the associated expenditure.

Creating a polished project description with AI at the end of the accounting period doesn’t retrospectively create evidence that the R&D took place.

The stronger evidence is often already sitting within the business: project plans, design documents, test results, development records, technical discussions, version histories, laboratory records, meeting notes and other material created while the project was happening.

The challenge is identifying and connecting that evidence to the claim.

6 | Would your technical team recognise the claim?

There is a simple test worth applying to any self-prepared or AI-assisted claim. Give the final technical narrative to the engineer, scientist, developer or other technical lead who actually carried out the work.

Ask: “Is this an accurate description of what happened?”

→  Could they explain the technological advancement being sought? Could they describe the uncertainty that existed at the outset?

→  Could they explain why the answer wasn’t readily available to a competent professional?

→  Could they talk through the experimentation undertaken and the results?

If the narrative contains language that the technical team themselves wouldn’t naturally use or struggle to explain, that should be a warning sign.

7 | Have you captured the costs correctly?

Identifying qualifying R&D is only half the exercise. Revenue also expects businesses to be able to  demonstrate that the expenditure included in the claim relates to the qualifying activities.

That means establishing appropriate methodologies for areas such as employee costs, subcontracted R&D and other qualifying expenditure. There is a risk in simply identifying an “R&D team” and assuming all of its costs qualify.

Equally, businesses can overlook employees outside the obvious technical team who spent part of their time directly carrying out qualifying R&D. The calculation needs to follow the activity, not simply the job title or department.

8 | Are you keeping the right evidence?

Revenue describes two essential requirements for an R&D claim: the Science Test and the Accounting Test. Records need to support both.

X   The important question therefore isn’t simply: “Could we explain why this was R&D?” 

✓   It is: “Could we demonstrate it?”

That distinction becomes particularly important if Revenue examines the claim months or years after the project took place.

9 | Don’t overlook the administrative requirements

There are procedural requirements too. Claims generally need to be made within 12 months of the end of the accounting period in which the expenditure was incurred.

And for accounting periods beginning on or after 1 January 2024, companies claiming for the first time, or which haven’t claimed in the previous three years, generally need to submit a pre-filing notification at least 90 days before making the claim.

Missing an administrative requirement can have consequences regardless of the strength of the underlying R&D.

10 | What happens if Revenue examines the claim?

This is perhaps the question every self-claimant should ask before submitting:

“If Revenue asked us to substantiate this tomorrow, could we?”

✅   Could you explain why each project qualifies?

✅   Could your technical people defend the scientific or technological arguments?

✅   Could your finance team demonstrate how the expenditure was calculated?

✅   Could you produce contemporaneous records supporting what the claim says happened?

And, crucially, would all of those elements tell the same story? If the answer is yes, you may have a strong claim. If the answer is “we’d just have to work that out if Revenue asked”, there may be more work to do.

AI can help. It shouldn’t replace judgement. 

There is nothing inherently wrong with using AI when preparing an R&D tax credit claim. Used appropriately, it can help organise information, improve clarity, summarise technical material or identify questions that need answering. But AI cannot independently determine whether your activities qualify. It cannot verify that the technical narrative accurately reflects what happened. It cannot create contemporaneous evidence that doesn’t exist. And it cannot take responsibility for the claim.

The same principle applies to preparing a claim yourself. The question isn’t whether you can do it. The question is whether you have the tax, financial and scientific or technological expertise needed to identify the full opportunity, draw the qualifying boundaries correctly, calculate the expenditure and build a claim you would be comfortable defending.

With the R&D Tax Credit now worth 35% of qualifying expenditure, getting that right matters.