AI tax software for tax professionals

What counts as AI tax software for professionals?

Three different products get called the same thing, none of them the AI built into consumer filing software, and buying the wrong one is the most common mistake.

Research is where the market is most mature, because the output is a memo a human reads. Preparation and review touch the return itself, which raises the stakes on accuracy.

How widely are firms actually using this?

Wider than the scepticism in the profession suggests. Thomson Reuters Institute's 2026 report on AI in professional services found 34% of tax firms already using generative AI in their work, with a further 47% either planning to use or considering a tool.

That leaves under a fifth of firms with no position at all, which is worth knowing if you are deciding whether this is early or late.

Which tools come up most

Naming products in a moving market dates quickly, so treat this as the shape of the field rather than a ranking.

• TaxGPT is an assistant aimed at preparers, covering cited research, document analysis, return review and client management. It reports more than 80,000 users and over 650,000 questions asked.
• Blue J and Thomson Reuters CoCounsel lead on research, and are the names that come up when practitioners specify that they need answers with citations.
• Keeper sits on the bookkeeping-to-tax workflow for smaller firms and for freelancers.
• Intuit's own AI is built into the consumer products, and Intuit has extended AI into its professional line as well.

None of these is a filing platform. Your return still goes out through Lacerte, ProSeries, UltraTax, Drake or whichever package your firm uses, and the AI tool sits beside it.

Why citations are the whole test

A general model can produce a confident answer with an invented citation, and invented tax authority is the failure mode that ends careers rather than afternoons.

A research tool built on a publisher's own content can point at the actual paragraph. That is the difference worth paying for, and it is the question to ask in a demonstration: not whether the tool answers well, but whether you can open what it cites and find the answer there.

Where AI is weakest for preparers

The same places a general chatbot gets tax questions wrong for everyone else, magnified because your clients' returns are more complicated than the average consumer's.

• Recent law. Tax year 2026 brings the charitable deduction's 0.5% of AGI floor, the limitation replacing the repealed Pease limitation, and a second year of the Schedule 1-A deductions.
• State conformity. Models routinely answer state questions as though the state followed federal treatment.
• Schedule K-1 and partner-level detail, where reviewers repeatedly found errors during the 2026 season.
• Digital assets, where Form 1099-DA reporting only began carrying cost basis with 2026 transactions.
• Anything requiring facts held only in the client file.

The rule that keeps this safe

Nothing an AI tool produces reaches a client or a return without a preparer reading it first. That sounds obvious and is the control that most often slips when a tool starts working well.

1. Use AI for the first draft of research, never the final position.
2. Open every citation before relying on a conclusion.
3. Keep a record of where a position came from, the same as for any other research.
4. Do not paste client-identifying data into a general consumer chatbot.
5. Check your professional liability position and your engagement letters before changing how work is produced.

The IRS is running its own models on the other side of this, using AI-assisted systems to score returns for non-compliance and recommend cases for examination. An AI-assisted error on your side is not treated more gently because a machine made it.