AI Tax Preparation: The Preparer Still Owns the Error

AI Tax Preparation: Your Name Is Still on the Return

AI tax preparation moved from experiment to routine in many firms over the last two filing seasons, and in June 2026 the IRS made its position explicit. Its Office of Professional Responsibility issued introductory guidelines explaining how Circular 230, the rules governing practice before the IRS, apply when practitioners use AI. The guidance does not ban anything. It says something more consequential: the duties of diligence, competence, and confidentiality are unchanged, and a practitioner who relies on a tool without checking its work has not met them. The software may have drafted the memo, but the name on it is still yours.

What Circular 230 Duties Mean With AI in the Workflow

The duties themselves have not changed. What they require in practice has.

What the IRS Guidance Actually Says

The Office of Professional Responsibility alert, released on 24 June 2026, offers an introductory set of principles for using AI responsibly in federal tax work, and it frames AI as something that should support professional judgement rather than replace it. Practitioners are expected to review AI-generated documents for accuracy before they reach a client or the IRS. Competence now includes understanding the technology used in client work, including its risks and its limitations.

The guidance also addresses confidentiality, including situations where client consent may be required before tax return information is shared with an AI service, fee arrangements where AI has replaced manual effort, and the responsibility of those who manage a practice to put procedures in place. Read the alert itself rather than relying on summaries. The details matter, and the guidance describes itself as introductory, which suggests more may follow.

The underlying principle is not uniquely American. Professional standards for accountants and tax advisers in most jurisdictions already place responsibility for advice on the adviser, and AI does not change that anywhere.

Where AI Tax Preparation Goes Wrong

Tax is close to a worst case for language models. Rules change every year, thresholds are indexed, treatment varies between federal and state systems and between countries, and a single missed condition can reverse an answer. Models also carry a knowledge cutoff, so the most recent changes are precisely the ones they are least likely to know.

Why AI Tax Preparation Errors Look Like Research

Ask a model about a deduction and it may cite a code section, a regulation, or a revenue ruling with complete confidence. Sometimes the citation is right. Sometimes the section exists but says something different. Sometimes the ruling does not exist at all. The formatting is identical in every case, and a busy reviewer skimming for reasonableness cannot tell them apart. It is the same pattern that has led to sanctions for fabricated legal citations in courtrooms, now arriving in tax workpapers.

Six AI Tax Errors to Check Every Time

A reviewer who checks these six categories will catch most of what AI gets wrong in tax work.

1. Outdated figures. Contribution limits, brackets, phase-outs, and thresholds carried over from a prior year.
2. Wrong jurisdiction. Federal rules applied to a state question, or one state's treatment applied to another.
3. Invented or misread authority. Citations that do not exist or do not actually support the stated position.
4. Arithmetic across long chains. Calculations that drift through multi-step computations while looking tidy.
5. Missed elections and deadlines. Procedural requirements left out of analysis that is otherwise sound.
6. Facts assumed rather than asked. Gaps in client information quietly filled with typical values.

Pros and Cons of AI in Tax Practice

Used well, AI makes a tax practice faster and frees time for advisory work. Used carelessly, it creates exactly the exposure the guidance warns about.

• Pro: faster document intake. Organising client documents and extracting figures takes a fraction of the time.
• Pro: quicker first-draft research. Initial issue spotting and memo structure arrive in minutes.
• Pro: more advisory capacity. Time saved on compliance work can move to planning conversations clients value.
• Con: citations need independent checking. Every authority has to be confirmed at the source.
• Con: data handling obligations. Sharing return information with an AI service can trigger consent and security requirements.
• Con: awkward fee questions. Clients increasingly ask how work was done and what that means for the bill.

Real Scenarios Worth Thinking Through

These scenarios are illustrative, showing how AI tax preparation risk plays out in practice rather than presented as verified case studies.

A small firm uses an AI assistant to draft a response to an IRS notice. The letter cites a revenue procedure that was superseded two years earlier. A partner spots the problem purely from memory, and the firm realises nothing in its process would have flagged it otherwise.

A preparer uploads a client's documents to a consumer AI chatbot to summarise a complicated return. The summary is genuinely helpful. The firm never obtained consent to share return information with that service, and its own data policy did not permit it.

A firm halves preparation time on simple returns using AI but keeps billing the same hourly estimates. A client asks directly how the work was done, receives an honest answer, and the conversation that follows is about fees rather than tax.

“After testing multiple AI models on coding, research, and business prompts, combined outputs produced more reliable results than any single model.” Internal multi-model evaluation, Talkory research team.

Firm Controls That Match the Guidance

A written AI policy is the practical response. It should name the approved tools, which should be enterprise services with appropriate data protections, and list prohibited uses such as pasting return information into consumer chatbots. It should set a review standard in which a named practitioner signs off on every AI-assisted position, describe how client consent is obtained where required, and require AI use to be noted in workpapers. A fee policy that reflects how AI changes the effort behind the work belongs in the same document.

Consulting firms have already learned how easily polished AI drafts pass review, as covered in who checks the deliverable in AI-assisted consulting. On the data side, the controls auditors expect are similar to those in our SOC 2 checklist for AI workflows.

Training matters as much as policy. Staff who understand how language models fail make far better reviewers than staff who are simply told to check the output. A short internal session walking through real examples of invented citations, outdated thresholds, and confident arithmetic mistakes does more for due diligence than any checkbox. Revisit the policy every filing season, because the tools are changing quickly and the IRS has signalled that its current guidance is only a starting point.

Why Talkory Wins

The most dangerous AI tax error is a confident citation. Talkory puts the same tax question to GPT, Claude, Gemini, Grok, Perplexity Sonar, and Kimi K3 together. When all six point to the same rule and authority, the reviewer has a strong starting point for verification. When they cite different sections or reach different conclusions, the disagreement shows exactly which position needs primary-source checking before it reaches a client or a return. It does not replace professional judgement. It makes the review the guidance requires faster and far more targeted. Firms should use it, like any AI service, within their data handling and consent rules.

Final Verdict

AI tax preparation now sits squarely inside the professional standards regime, and the message is simple: tools can draft, calculate, and research, but the practitioner owns the result. Review every AI-generated position, understand the tools well enough to know how they fail, control where client data goes and obtain consent where required, price work honestly, and put firm-wide procedures in writing. Practitioners who do that keep the speed without taking on the sanctions risk.