Skip to main content

Research

AI tax assistant research: what the evidence shows

Research on AI tax assistants points the same way: general chat tools give wrong tax answers, grounded tools still hallucinate, and firms want citations, security and human review. This page summarises seven findings from published studies and surveys, and explains how TaxOrch is designed to reduce each risk, without claiming to remove it.

Book a demoAsk a question

Last reviewed 18 official sourcesHow we check this page

Do consumer AI tax chatbots give correct answers?

According to The Washington Post's March 2024 test (opens in a new tab), TurboTax's Intuit Assist answered more than half of the reviewer's questions wrongly and H&R Block's assistant about 30%, including wash-sale rules wrongly applied to crypto. According to NerdWallet's March 2026 study (opens in a new tab), general chat models were near-perfect on exam-style facts but got the standard deduction wrong by about $3,000, said lapsed EV credits were still available and gave inconsistent estimates.

How often do grounded AI research tools hallucinate?

The report notes there is no published quantitative accuracy data for tax research tools, so the closest independent evidence comes from law. According to Stanford RegLab (opens in a new tab), Westlaw AI-Assisted Research hallucinated about 33% of the time, against 17% for Lexis+ AI and 43% for GPT-4: retrieval reduces errors but does not remove them. Accounting Today (opens in a new tab) has reported hallucinated citations in Big Four client reports.

Why do tax professionals hesitate to trust AI?

According to Thomson Reuters (opens in a new tab), 96% of professionals expect AI to protect client data, 94% want grounded outputs and 90% want explainable reasoning. Its AI trust gap research (opens in a new tab) found 42% of tax firms cite data security as a barrier, and that firms want isolated data environments, clear citations and audit trails. Consumers are wary too: according to Accounting Today (opens in a new tab), 63% of US consumers do not trust AI compared with tax professionals.

What do tax authorities say about relying on AI?

According to Thomson Reuters' report of the National Taxpayer Advocate's June 2024 remarks (opens in a new tab), "taxpayers should not solely rely on AI-generated tax advice". According to the Journal of Accountancy (opens in a new tab), in June 2026 the IRS Office of Professional Responsibility said Circular 230 duties of due diligence, competence and confidentiality apply fully to AI use. Australia's Tax Practitioners Board draft guidance (opens in a new tab) requires agents to verify AI output.

What data-security duties apply when AI handles tax data?

The report sets out three US obligations. Under Section 7216 (opens in a new tab), putting client return data into general AI tools is a disclosure, and consent must name the AI provider, the data and the purpose (opens in a new tab). Under the FTC Safeguards Rule (opens in a new tab), tax preparers count as financial institutions and need a risk assessment, encryption, multi-factor authentication and an incident response plan. IRS Publication 4557 (opens in a new tab) covers the written information security plan, which the report describes as legally required.

How quickly are tax firms adopting AI?

According to the Blue J and CPA.com survey (opens in a new tab), 60% of US tax firms use AI tax research weekly, up from 33% a year earlier, and 69% expect to move toward value-based, fixed or hybrid fees. According to Karbon's 2026 report (opens in a new tab), 98% of firms use AI daily but only 21% have a documented AI policy.

Do AI tax tools cover other countries and languages?

According to the report, research tools cover mainly the US, Canada and the UK, and international coverage depends on add-ons such as IBFD content (opens in a new tab) or the Big Four. Language matters as well: ClearTax reported (opens in a new tab) about 50,000 non-English filings within weeks of launching in six Indian regional languages.

Frequently asked questions

Is AI accurate enough to do my taxes?

Published tests found consumer AI chat tools gave many wrong tax answers, and the National Taxpayer Advocate has said taxpayers should not rely solely on AI-generated tax advice. Use AI for preparation and checking, and have a person review the result.

Does citing sources stop an AI from hallucinating?

No. Stanford RegLab found grounded legal research tools still hallucinated between about 17% and 33% of the time. Citations make answers checkable, which is why TaxOrch cites a specific section of an official document for every legal claim.

Does TaxOrch file my return?

No. TaxOrch prepares draft packages marked as not filed. You file through the tax authority, authorised software or an adviser.

Is my data used to train TaxOrch's models?

Not unless you explicitly opt in.

Sources

Related on TaxOrch

Next step

Ask TaxOrch about your own situation.

Bring your own questions about Research to a demo, or request a demo account and try them yourself. Every answer shows the official source it rests on, and firms can offer TaxOrch under their own brand.

Decision support, not tax advice. A professional reviews before anything is filed.