AI and the R&D Tax Credit: What Actually Qualifies?

Artificial intelligence is redefining how companies operate—but not all AI-related work qualifies for the R&D tax credit. Just because a team is using machine learning or building with generative AI doesn’t mean it automatically meets the requirements under Section 41. For tax teams supporting AI investments, the key is understanding when that work crosses into qualifying R&D—and when it doesn’t.

Start with the Four-Part Test

Whether you’re building AI models or other types of software, the starting point for qualifying R&D remains the same: the four-part test under Section 41.

1. Permitted Purpose – The activity must be intended to develop or improve a business component’s functionality, performance, reliability, or quality.
2. Technological in Nature – The work must rely on hard sciences like computer science or engineering.
3. Elimination of Uncertainty – There must be uncertainty at the outset regarding capability, methodology, or design.
4. Process of Experimentation (POE) – The team must engage in a systematic trial-and-error process to resolve that uncertainty.

Most AI-related work will fall under the umbrella of software development. But if that software is intended primarily for your company’s internal use—such as AI tools for internal data analytics, HR automation, or supply chain optimization—the IRS applies an additional layer of scrutiny.

If It’s Internal-Use Software, You’ll Need to Pass Three More Tests

Internal-use software (IUS) must satisfy not only the standard four-part test but also the three-part “High Threshold of Innovation” test, which adds a higher burden of proof.

To qualify, the software must:

1. Be Innovative
   This is often misunderstood. “Innovative” doesn’t mean the software has to be cutting-edge in a market sense. It means the development must result in a substantial and measurable improvement in speed, cost, or capability compared to current solutions. The bar is technical, not just aspirational. And importantly, this test is not about business innovation or market success. The IRS is not looking for proof of a commercial breakthrough; they’re looking for clear technical advancement. That distinction matters, especially when talking to internal stakeholders who might associate “innovation” with financial or competitive risk.
2. Involve Significant Economic Risk
   The development must require substantial resources and face the risk of failure from a technical perspective. If the outcome is certain or if the project is unlikely to fail due to reliance on known techniques, it won’t qualify. Business uncertainty (e.g., will customers adopt it?) is not relevant here—only whether the technology might not work.
3. Not Be Commercially Available
   You must be developing the software because there’s no comparable product available for purchase or licensing. If you’re adapting something that’s already on the market and using it “as is,” it won’t pass this test. But if you’re modifying or extending an existing solution in ways that require significant technical effort, that may qualify.

What AI Activities Typically Qualify?

When AI projects meet the four-part test (and the three additional tests for IUS, if applicable), they’re often doing one or more of the following:

• Designing new AI models to solve complex business problems.
• Modifying existing algorithms to improve accuracy or performance.
• Developing proprietary machine learning infrastructure.
• Training large-scale models with custom datasets that require novel approaches.

These efforts usually involve technical uncertainty, and resolving it requires more than just implementation. There has to be trial-and-error, iteration, and meaningful engineering input.

What Usually Doesn’t Qualify

Simply using AI tools to do your job faster doesn’t count. Some common non-qualifying activities include:

• Applying out-of-the-box generative AI tools (e.g., prompting GPT to write code).
• Automating tasks with pre-trained models and no underlying development.
• Integrating third-party AI solutions without modification or customization.

The IRS is increasingly attuned to this distinction. They’ll be looking for documentation that demonstrates how your team went beyond using AI—and actually engineered something new.

Where the IRS May Scrutinize AI Projects

Our Director of Controversy puts it plainly:

“The IRS is going to look through the lens of ‘you’re just using existing AI to conduct research.’”

If your claim relies on adapting or extending third-party tools, you’ll need to document the technical challenges involved and how your team addressed them.

Documentation should clearly show:

• The uncertainty you faced at the outset.
• The experimental methods used to resolve it.
• Why existing tools weren’t sufficient.
• The expertise involved in overcoming those limitations.

Without this kind of detail, IRS reviewers may push back on the validity of the credit.

AI and the R&D Tax Credit: A Software Lens

AI innovation isn’t a special case—it’s still software development at its core. That means all the same rules apply. If your team is designing, developing, or substantially improving AI systems, and that work involves technical uncertainty and experimentation, it could qualify.

But if you’re simply applying tools others have built? Probably not.

Final Thoughts for Tax Teams

As more companies integrate AI into their operations, tax teams must get proactive about evaluating and documenting these activities. The opportunity to claim the credit is real—but so is the risk of overreach.

If you’re unsure where your AI-related work falls on the spectrum, a deeper technical and tax analysis is key. MASSIE can help you dig into the details and prepare documentation that stands up to scrutiny.

For more on how to build better documentation or train SMEs, check out:

• How to Train Your SMEs to Support R&D Tax Documentation
• Optimizing the R&D Tax Credit Process

If you have any questions, reach out. We’re here to help.

Jason Massie, CPA, Esq.

As the Founder and President of MASSIE R&D Tax Credits, Jason is responsible for the quality, consistency, and integrity of our services. His strategic vision and leadership drive innovation within the company, encouraging collaboration and ne...