Using AI for tax research without losing the guardrails

Artificial intelligence can speed up tax research, but it doesn't change what makes a conclusion defensible. The practitioner still has to understand the facts and verify the authority supporting the conclusion.

The questions below from NATP’s Practical Application of AI in Tax Research webinar focus on where AI can fit into that process, and where professional judgment still has to take over.

How can AI help with tax research?

AI can assist throughout the tax research process. Early on, it can help organize facts or identify missing documents. As the research develops, practitioners can use it to generate search terms and compare source material. It can also help draft a research memo or translate technical conclusions into client-facing language.

Its output, however, is not tax authority.

What can go wrong when using AI for tax research?

AI may hallucinate (make up facts or citations), rely on outdated law, cite the wrong jurisdiction, overstate a conclusion or leave the practitioner with inadequate documentation. General-purpose AI assistants can also point users toward secondary sources when primary authority is needed.

A polished response should not be mistaken for a verified one. Practitioners need to review the cited authorities and determine whether they support the conclusion before relying on the response.

Does AI change the professional standard for tax research?

No. The practitioner remains responsible for the tax conclusion, regardless of whether AI was part of the research process.

Practitioners must check AI-generated citations against the actual sources and read cited cases. Primary sources include the Internal Revenue Code, Treasury regulations, rulings and case law.

Practitioners must also protect taxpayer return information when using AI tools. Any disclosure or use must comply with applicable law, and the tool must have appropriate confidentiality safeguards. Sections 7216 and 6713, along with Circular 230, remain relevant when AI enters the research workflow.

How should tax professionals evaluate AI-located authority?

Start by determining whether the authority applies to the taxpayer’s facts and the relevant jurisdiction. Then assess its legal weight and whether it remains valid.

The Code and Treasury regulations generally carry more weight than lower-level guidance, but a source’s place in the hierarchy does not establish that it supports the taxpayer’s position. Practitioners still need to examine what the authority actually says. Use a citator to check whether a case has been overruled or otherwise affected by later decisions.

What should a tax research prompt tell the AI?

A strong prompt should define the research question and provide the relevant facts. It should also explain what sources the AI should use and what the finished response should look like.

For example, practitioners can instruct the AI to prioritize primary authority and distinguish controlling authority from persuasive material. The prompt should require current sources and prohibit invented citations. If facts are missing or authority cannot be verified, the tool should flag the gap rather than fill it with an assumption.

Clear instructions give the practitioner a more useful starting point for review.

Can AI help improve a tax research prompt?

Yes. Before starting the research, practitioners can ask AI to review the prompt itself for ambiguity or missing context.

AI’s review can also identify unstated assumptions and unclear source requirements. Resolving those gaps before requesting an answer can make the resulting research easier to evaluate.

Where does the tax professional fit into AI-assisted tax research?

The practitioner directs the research and decides whether the findings support the conclusion. AI can help develop the work product, but professional judgment remains essential whenever the output will inform advice or a return position.

For a deeper look at incorporating AI into a structured tax research process, NATP's 2026 Practical Application of AI in Tax Research webinar is available on demand.