Can AI prepare tax returns? What to automate—and still review

The short answer: AI drafts, professionals remain responsible

Yes—AI can prepare tax returns, in the sense that matters most to a working firm. Modern tax-preparation AI can ingest a client's W-2s, 1099s, 1098s, K-1s, and IDs, extract the figures, populate the correct fields in Drake, ProSeries, or Lacerte, and hand a preparer a draft return that is ready to review. Document handling and repetitive data entry are genuine, defensible uses of automation when the firm keeps professional review in the workflow.

What AI cannot do is become the preparer. It cannot exercise professional judgment, it cannot take legal responsibility for a return, and it cannot sign one. Those are not limitations of today's models that will disappear next year; they are how the U.S. tax system is designed. The IRS is explicit that a paid preparer is "primarily responsible for the overall substantive accuracy" of a return and is "required by law" to sign it and include a Preparer Tax Identification Number (PTIN). No software—however capable—changes who holds that responsibility.

So the useful question is not "can AI replace the preparer?" It is: how much of the return can AI get right and organized before the preparer opens it, and how do I keep review firmly in human hands? The rest of this guide answers that, grounded in the actual rules that govern paid preparation.

What the law actually makes you responsible for

Before adopting any AI tool, it helps to be precise about what the signing professional is on the hook for—because that is exactly what automation cannot offload.

Signature, PTIN, and "primary responsibility"

A paid preparer must sign the returns they prepare and enter a valid PTIN. These are not formalities. The IRS backs them with penalties under Internal Revenue Code §6695: a penalty for failure to sign a return and a separate penalty for failure to furnish a PTIN (each $60 per failure for 2025, indexed annually). The IRS specifically targets "ghost preparers"—people who prepare returns for pay but omit their identity. An AI tool is not the signing preparer; the paid return preparer must sign the return, include their PTIN, and accept responsibility for the work.

Accuracy and the understatement penalties

Responsibility for accuracy is enforced through IRC §6694. A preparer can face a penalty for an understatement due to an unreasonable position, and a substantially larger penalty for willful or reckless conduct—$5,000 or 75% of the income derived from preparing the return, whichever is greater. The takeaway for AI adoption is blunt: if an AI-populated figure is wrong and the return understates tax, the exposure lands on the signing professional, not the software vendor. That is precisely why "review" is not a nicety in an automated workflow—it is the control that protects the person whose name is on the return.

Circular 230 and the duty of due diligence

Treasury Department Circular 230 governs practice before the IRS and requires a practitioner to "exercise due diligence" in preparing and filing returns. Importantly, Circular 230 also describes when reliance is reasonable: a practitioner is presumed to have exercised due diligence if they rely on the work product of another person and used reasonable care in engaging, supervising, and evaluating that person. Applied to AI, the analogy is instructive—reliance can be reasonable, but only when the professional supervises and checks the tool's output rather than accepting it blindly.

How AI-assisted tax preparation actually works

A well-built workflow mirrors the steps a preparer already follows and simply removes the manual data handling, keeping the professional at every decision point. In practice it runs as a defined pipeline:

1. Secure document intake. The client uploads documents through a portal, or the firm scans them. Files are encrypted and organized by client and return so nothing is loose or mislabeled at the start.
2. Classification. AI identifies each document by type—W-2, 1099-NEC, 1099-DIV, 1099-B, 1098, K-1, driver's license—so each one routes to the right place in the return.
3. Extraction and data entry. The system reads each field and maps the value to the correct line in your tax software, replacing the keystroke-by-keystroke transcription that consumes most preparation time.
4. Completeness and exception checks. The workflow compares this year against the prior year, detects missing forms, and flags figures that look inconsistent—so gaps surface as questions instead of silently entered errors.
5. Professional review and sign-off. The preparer opens the drafted return, verifies flagged items and every material figure against source documents, resolves exceptions, applies judgment, and approves the return for signature and e-file. This step is never automated away.

The output is a return that arrives at review already populated and pre-checked, so the professional spends their scarce time on judgment and verification rather than typing. Note that even filing has a human-choice dimension: the IRS e-file mandate requires preparers who file 11 or more covered returns to e-file, but clients may still elect to paper-file—another reminder that automation operates inside a framework of professional and client decisions.

The due-diligence standard AI cannot meet for you

Nowhere is the human-in-the-loop requirement more concrete than in the paid-preparer due-diligence rules for the Earned Income Tax Credit, Child Tax Credit/Additional CTC, the American Opportunity Tax Credit, and head-of-household status. Under IRC §6695(g), a preparer who determines eligibility for or the amount of these benefits must meet four due-diligence requirements and file Form 8867, the Paid Preparer's Due Diligence Checklist, with the return.

The knowledge requirement is the part AI structurally cannot satisfy: the preparer must interview the taxpayer, ask adequate questions, contemporaneously document those questions and the client's answers, and review enough information to judge eligibility. AI can collect responses, organize the client's documents, and surface inconsistencies, but the preparer must assess those inconsistencies, ask and document appropriate follow-up questions, and satisfy the due-diligence requirements. Preparers must also retain their due-diligence records—including Form 8867, computation worksheets, and the documents relied upon—for three years. If your AI workflow touches credit-eligible returns, it must support this professional process rather than replace it.

The data-handling rules AI users overlook

Adopting AI is not only a question of accuracy—it is a question of how client data is used and protected, and here the rules are stricter than many firms realize.

IRC §7216: consent to use or disclose return information

Section 7216 imposes criminal penalties on preparers who knowingly or recklessly disclose or use a taxpayer's return information for purposes other than preparing the return, unless an exception or the taxpayer's consent applies. The Treasury regulations were deliberately written for the software era and cover the disclosure and use of return information by electronic and software-based preparation. The practical implication for AI: if a tool uses client tax data in ways that go beyond preparing that client's return—training external models, sharing with third parties—you may need specific, informed taxpayer consent. Before you route client data through any AI service, confirm exactly how that data is used, and whether §7216 consent is required.

The FTC Safeguards Rule and the WISP requirement

Paid tax preparers are treated as "financial institutions" under the Gramm-Leach-Bliley Act, which places them under the FTC Safeguards Rule. Every firm is required to maintain a Written Information Security Plan (WISP); the IRS reinforces the same expectation through Publication 4557, Safeguarding Taxpayer Data, and the Security Summit's Publication 5708 WISP template. There is no small-firm exception—solo and seasonal preparers are covered too. Any AI tool you adopt becomes part of the environment your WISP must account for, so encryption, access controls, and data retention are procurement questions, not afterthoughts.

Professional standards catch up to AI

The profession has already addressed reliance on tools directly. The AICPA's revised Statements on Standards for Tax Services, effective January 1, 2024, added a standard on reliance on tools (SSTS No. 1, §1.4) and a standard on data protection (§1.3). The reliance standard states the principle cleanly: a member may reasonably rely on tools used to provide tax services, but "use of the tool does not absolve the member of their professional obligations." That single sentence is the correct mental model for AI in tax preparation.

What tax firms should evaluate in an AI tool

Workflow fit

The best automation adapts to how your firm already works rather than forcing a new system on you. If your team prepares and reviews in Drake, AI should deposit a drafted return there—not in a separate portal that adds re-entry steps. Evaluate whether a tool slots into your existing intake, preparation, and review sequence or asks you to rebuild it.

Accuracy and a visible automation boundary

Draw a hard line between what the software produces and what the professional owns. Automation should handle extraction, data entry, and exception flagging; the professional owns interpretation, judgment, review, approval, and the signature. A trustworthy tool makes that line explicit and shows the preparer what the AI did and why, so verification is fast and evidence-based rather than blind trust.

Integration and data governance

"Works with your tax software" is not the same as a tested, supported connection to your specific program and version—confirm it. Ask how client data is encrypted in transit and at rest, who can access it, where it is stored, how long it is retained, and whether the vendor uses your data to train models (a §7216 question). These belong in the evaluation alongside accuracy.

Evaluation question	Why it matters	What to verify
What does the AI enter, and what does it leave for review?	Defines the boundary between automation and professional judgment	A documented list of automated fields and mandatory review points
Which tax programs and versions are actually supported?	An unconfirmed integration creates rework, not savings	The specific software, version, and how the drafted return is delivered
How are exceptions and missing documents surfaced?	Silent gaps are how incomplete returns reach a client	The missing-document and inconsistency flags in a live walkthrough
How is client data used, secured, and retained?	§7216, the FTC Safeguards Rule, and your WISP all apply	Encryption, access controls, retention, and whether data trains models
Can a preparer trace every figure to its source?	Due diligence and §6694 exposure require verifiable work	A source-to-field audit trail the reviewer can inspect

Where AI goes wrong—and how review catches it

Understanding the failure modes is what makes review effective rather than perfunctory. AI extraction tends to struggle in predictable places: poor-quality scans and photos; documents with unusual layouts; multi-account or multi-page brokerage statements where totals and wash-sale adjustments must reconcile; handwritten annotations; and anything requiring classification judgment, such as whether income belongs on a Schedule C or as other income. Large language models can also produce confident-sounding but incorrect output, so a figure that "looks right" is not the same as a figure that is right.

The mitigation is not to distrust automation—it is to design review around these known weak points. A strong workflow flags low-confidence extractions, surfaces year-over-year anomalies, and forces a reviewer's eyes onto exactly the fields most likely to be wrong. Reserve professional attention for judgment calls and reconciliations; let automation clear the routine. That division is how firms get the speed of AI without inheriting its errors.

A practical example, start to finish

Consider a firm preparing an individual return for a client with two W-2s, a mortgage interest statement, and a brokerage account. The following figures are illustrative, not statistical claims.

Traditionally, a preparer might spend 45–60 minutes keying documents into the tax software before any real analysis begins. In an AI-assisted workflow, the client uploads the documents through a secure portal; the system classifies and extracts them, populates the fields in the firm's tax program, and flags two things: a 1099-DIV referenced on the brokerage summary that has not yet been provided, and dividend income materially higher than the prior year. The preparer opens the drafted return, sees both flags, requests the missing form, verifies the entered figures against the source documents, reconciles the brokerage activity, applies judgment to the investment income, confirms there are no credit-eligibility questions requiring a Form 8867 interview, and approves the return for the client's signature and e-file.

The repetitive work compressed from most of an hour to a few minutes; the professional's time shifted to the review, reconciliation, and advice that clients actually pay for—and their name goes on a return they genuinely checked. That is the honest promise of AI in tax preparation: not a preparer replacement, but a preparation assistant that lets tax professionals do more of the work only they can do.

Frequently asked questions

Can AI file a tax return by itself?

No. AI can assemble a review-ready draft, but a paid preparer is required by law to review the return for accuracy, sign it, and include their PTIN before it is filed. Failure to sign or furnish a PTIN carries penalties under IRC §6695, and the signing professional remains responsible for the return.

Is it legal to use AI to prepare tax returns?

Yes, when the signing tax professional reviews and signs the return and the tool is used consistent with the rules. Key obligations include IRC §7216 (consent to use or disclose return information), the FTC Safeguards Rule and a Written Information Security Plan, and the AICPA standard that reliance on a tool does not remove your professional obligations.

Who is responsible if an AI-assisted return is wrong?

The signing tax professional is responsible for the substantive accuracy of the return under IRS rules, regardless of whether software or AI assisted. IRC §6694 penalties for understatements attach to the preparer, which is why professional review and verification remain mandatory control points.

Can AI handle EITC or Child Tax Credit returns?

It can help organize documents and prepare for the client interview, but it cannot satisfy the paid-preparer due-diligence knowledge requirement under IRC §6695(g). The professional must interview the client, document the responses, complete Form 8867, and retain the records for three years.

Will AI train on my clients' tax data?

That depends entirely on the vendor, and it is a question you must ask. If a tool uses client return information beyond preparing that client's return, IRC §7216 may require specific, informed taxpayer consent. Confirm data use, retention, and model-training practices before adopting any tool.

Sources and methodology

This article is based on published IRS guidance, the Internal Revenue Code preparer provisions, FTC and AICPA standards, and Tax Automate product documentation. Illustrative figures are labeled as such and are not statistical claims. Rules and penalty amounts are current as of publication and should be verified for the applicable tax year.

• IRS — PTIN Requirements for Tax Return Preparers
• IRS — Tax Preparer Penalties (IRC §6694 and §6695)
• IRS — Circular 230, Regulations Governing Practice before the IRS
• IRS — Section 7216 Information Center
• IRS — Due Diligence Law, Regulations and Requirements (Form 8867)
• IRS — E-file Requirements for Specified Tax Return Preparers
• IRS — Publication 4557, Safeguarding Taxpayer Data (PDF)
• FTC — Safeguards Rule: What Your Business Needs to Know
• AICPA — Statements on Standards for Tax Services (SSTS)