ACCOUNTING

The Ethical Use of AI in Tax Preparation: A Practical Guide

July 27, 2026 • 11 MIN READ

TL;DR

Follow a five-step framework for ethical AI use in tax preparation that protects client data, maintains professional judgment, and complies with regulatory standards. This guide covers the four unacceptable use cases to avoid, how to build an AI review layer into your workflow, and specific disclosure language to use with clients.
AI can handle data extraction, classification, and draft preparation, but the final review and signature must always remain with a qualified human preparer.
Your biggest ethical risk is not the tool itself. It is treating AI output as finished work instead of a starting point.
The firms that win will use AI to reduce errors and increase speed while keeping the human in charge of every material decision.

I got a call last month from a CPA in Ohio. Mid-sized firm, twelve staff, solid reputation. He had been using ChatGPT to draft tax returns for a few weeks without telling anyone. His question was simple. “Am I going to get sued?”

The honest answer is that nobody knows yet. The IRS has not issued formal guidance on generative AI in tax preparation. The AICPA has published principles but not a specific standard. The regulatory environment is still catching up to the technology. What we do know is that a few clear rules separate ethical AI use from reckless AI use, and most firms are not following them.

I have spent the last year studying how AI actually behaves in tax and accounting workflows. I have tested tools, reviewed outputs, and talked to dozens of firms who are using AI right now. Some are doing it well. Most are making mistakes that could cost them their licenses.

The firms that get this right will have a real advantage. They will produce higher quality work with lower overhead and fewer errors. But that only happens if you treat AI as a junior associate who needs constant supervision, not as a replacement for your professional judgment.

The Four Use Cases That Need Ethical Guardrails

Most ethical questions in AI-powered tax preparation fall into four categories. If you understand these four buckets, you can evaluate any tool or workflow against them.

Data confidentiality. When you paste a client’s Social Security number, W-2 details, and prior year returns into a large language model, that data goes somewhere. With consumer tools like ChatGPT Free or Claude Free, your input may be used for model training. That means client data could appear in someone else’s output months later. This is a direct violation of your fiduciary duty and IRC Section 7216 which governs disclosure of tax return information. Always use enterprise-grade or business-tier accounts that contractually agree not to train on your data. Read the terms of service before you paste anything.

Accuracy and hallucination risk. AI models invent things. They do it confidently and they do it often. In tax preparation, a hallucinated deduction or a misapplied tax credit can trigger penalties, interest, and professional liability exposure. I tested four leading AI tools on a straightforward 2023 return scenario. Two of them invented a tax credit that did not exist. One misclassified a business expense category in a way that would have flagged an audit. The fourth got it right. The variability is too high to trust any single model without human verification.

Professional judgment delegation. The most common ethical error I see is preparers using AI to make judgment calls. Should this expense be capitalized or expensed? Does this contractor qualify as a 1099 worker or should they be a W-2 employee? Those questions require professional judgment based on facts and circumstances that the AI does not fully understand. The model can summarize the relevant code sections, but it cannot weigh the nuances of your specific client situation. That is your job.

Client awareness and consent. If you are using AI in your tax preparation workflow, your clients have a right to know. Many engagement letters do not address this at all. The ethical standard is disclosure. You do not need to list every tool you use, but your clients should understand that AI is involved in processing their data and that a human remains responsible for the final return.

Build an AI Review Layer Into Your Workflow

The single most important thing you can do to use AI ethically is to build a review layer that treats AI output as a draft, not a deliverable. This is not complicated but it requires discipline.

Here is the workflow I recommend to every firm I work with.

Step one is data extraction. Use AI to pull numbers from source documents. This is a low-risk use case because numbers either match the source or they do not. You can verify them quickly. The ethical risk here is small as long as you check the extraction against the original documents.

Step two is classification and organization. Use AI to sort expenses into categories, flag potential deductions, and identify missing information. This is where AI saves the most time. The ethical risk is moderate because errors in classification can change tax liability. You need to sample the classifications systematically, not randomly.

Step three is draft preparation. Use AI to produce a draft return based on the extracted and classified data. This is where the ethical risk becomes significant. The draft will contain errors. It will miss nuance. The question is whether you treat it as a starting point or as a finished product.

Step four is human review and override. This is non-negotiable. Every single number on the draft must be reviewed by a qualified preparer who understands the client’s situation. The AI is not the preparer. You are. If you sign a return that contains an AI-generated error you did not catch, you own that error.

The Five Ethical Questions Every Firm Should Answer

Before you adopt any AI tool for tax preparation, answer these five questions in writing. Keep the answers on file as part of your quality control documentation.

One: Does this tool have a business-tier data protection agreement that prevents training on my client data? If the answer is no, do not use it.

Two: Have I tested this tool on at least twenty historical returns to measure its error rate? If you have not benchmarked the tool against known-correct outputs, you do not know how it performs. Run the test. Document the results.

Three: Is there a human review step before any AI-generated output reaches a client or the IRS? If the workflow allows AI output to go directly to filing without human sign-off, you have an ethical gap.

Four: Have my clients been informed that AI is used in their return preparation? This can be a simple disclosure in your engagement letter. It does not need to be. But it needs to exist.

Five: Do I have a documented process for handling AI-generated errors when they occur? They will occur. The question is whether you catch them before filing or after. Have a remediation plan.

Firms that answer these five questions honestly will be in the top tier of ethical AI adoption. Firms that skip them are gambling with their clients’ trust and their own licenses.

What the Regulators Are Actually Saying

The IRS has not issued formal AI guidance yet, but the signals are clear. The IRS Large Business and International division has been using machine learning for audit selection for years. The agency understands AI and is watching how preparers use it.

The AICPA’s Code of Professional Conduct already applies. Principle 1.700.001 requires members to exercise professional competence, which includes understanding the technology they use. If you use AI without understanding its limitations, you may be violating this principle already.

State boards are starting to pay attention. California and New York have both raised the topic in professional ethics continuing education. I expect formal guidance from at least three state boards within the next twelve months.

The direction is clear. Regulators will not ban AI in tax preparation. They will require that human professionals remain responsible and accountable for the output. That is the ethical floor.

Practical Disclosure Language You Can Use Today

Your engagement letter needs to address AI use. Here is sample language you can adapt.

“We use artificial intelligence tools to assist with data extraction, classification, and draft preparation in our tax preparation workflow. These tools are used under enterprise-grade data protection agreements that prevent your data from being used for model training. All AI-assisted work is reviewed and verified by a qualified human preparer before filing. You retain the right to ask about any specific tool or process used in your return.”

This language is transparent without being alarmist. It addresses data privacy, human oversight, and client rights. You can add it to your standard engagement letter today.

If a client asks follow up questions, answer them honestly. Most clients will appreciate the transparency. Some may even see it as a sign that your firm is modern and competent. The risk of hiding AI use far outweighs the risk of disclosing it.

Is it legal to use AI for tax return preparation under IRS rules?

Yes, it is legal to use AI tools in tax preparation as long as a qualified human preparer reviews and takes responsibility for the final return. The IRS does not prohibit the use of technology or software. The key requirement is that the designated preparer verifies the accuracy of the return before signing. Your professional judgment and oversight are the legal requirement. The AI is simply a tool in your workflow.

What are the biggest ethical risks when using AI in tax prep?

The three biggest risks are data confidentiality breaches from using consumer-grade AI tools, over-reliance on AI-generated output without human verification, and failure to disclose AI use to clients. Each of these risks can result in professional liability exposure, regulatory penalties, or loss of client trust. Mitigating them requires specific policies around tool selection, workflow design, and client communication.

How do I disclose AI use to tax clients without scaring them?

Frame the disclosure as a quality and efficiency improvement, not as a warning. Use simple language that explains the role of AI in your workflow, confirms that client data is protected, and emphasizes that a human professional reviews everything. Most clients will be neutral or positive about the disclosure. The key is to be transparent without being defensive. If you are confident in your process, that confidence transfers to the client.

Your Next Step

Ethical AI use in tax preparation is not complicated. It comes down to three commitments. Protect client data. Review everything. Disclose honestly. If you do those three things consistently, you will stay on the right side of the regulators and the right side of your clients.

If you want a practical walkthrough of how to set up these guardrails in your firm, including the specific tools I recommend and the exact workflow I use, I put together a free playbook. It covers tool selection, data protection agreements, review workflow templates, and the disclosure language I shared above.

Download the free accounting AI playbook here.

I also cover this material in depth on the AI Blindspot YouTube channel. The channel walks through real implementations with real firms so you can see what actually works and what does not.

By Ben Merrick, CPI (AI)

This is education about AI strategy, not a guarantee of results. Results depend on implementation quality, firm size, and market conditions. Consult a qualified advisor before making technology investment decisions.

This is education, not a guarantee of results. Results depend on implementation quality, firm size, and market conditions. Consult a qualified advisor before making technology investment decisions.

← Back to Blog