Free AI for Tax Professionals: Where the Free Tier Stops at Client Data

Not safely. Return information is protected by Section 7216 and by the FTC's Safeguards Rule, and a free consumer AI tool gives you no contract, no controls and no record to show for it. Use free tiers for material that has no client in it. Anything that does needs consent, a paid business plan or no AI.

Whether sending return information to an AI tool counts as a "disclosure" has not been settled by IRS guidance that I could find, so the cautious reading is the sensible one: treat it as a disclosure. This is general information for practice owners, not legal or tax advice. Check your own position with a qualified adviser.

Follow me on Instagram@sagnikteaches

Section 7216 and the Safeguards Rule in plain terms

Two sets of rules matter, and they are separate. Satisfying one does not satisfy the other.

Connect on LinkedInSagnik Bhattacharya

Section 7216. A tax return preparer generally cannot disclose or use a taxpayer's return information without the taxpayer's written consent obtained beforehand. The regulation (26 CFR 301.7216-3) sets what valid consent looks like: it names the preparer and the taxpayer, states the purpose, names the specific recipients, describes the information involved, and carries the taxpayer's signature and date. Consent cannot be obtained after the fact, and it cannot be made a condition of service in most cases. Unless the consent states a period, it lasts one year from signing. Uses and disclosures need separate documents. Civil and criminal penalties apply under sections 6713 and 7216.

Subscribe on YouTube@sagnikteaches

There are exceptions that let a preparer share information without consent, such as with colleagues in the same firm for preparation or auxiliary services, with another preparer helping to prepare the return, or with a contractor for programming, maintaining or testing tax-preparation software "only to the extent necessary", and only after those individuals receive a written notice of the rules. Whether a general-purpose AI assistant fits any of these is exactly the open question. Some commentators read an AI vendor as a recipient that needs consent; others note that the IRS has not addressed it. The cautious practice is to assume consent is needed, naming the vendor.

The FTC Safeguards Rule. The FTC's guidance lists tax preparation firms among the financial institutions the rule covers. Its main requirements: a Qualified Individual to run the security programme, a written risk assessment, encryption of customer information in transit and at rest, multi-factor authentication for anyone accessing customer information, and a written incident response plan. Institutions holding information on fewer than 5,000 consumers are exempt from certain provisions, which the page does not list, so check the rule's text before relying on it. A breach involving unencrypted information on 500 or more consumers must be reported to the FTC within 30 days of discovery. IRS Publication 5708 is the practical guide, with a sample template, to the written information security plan (the "WISP") that most small practices use; Publication 4557, Safeguarding Taxpayer Data, covers the wider security steps and points to it.

Put together: you need a written plan, you must know where client data goes, and an AI tool is a place client data can go. If the tool is not in the plan, you have a gap.

What a free tier lacks, whatever the brand

The issue is not that free tools are worse at tax questions. It is that they are built for individuals, not for a regulated practice. Specifically:

• No contract with you. Consumer terms are accepted by clicking. There is no data-processing agreement, no confidentiality commitment tailored to you, and you cannot negotiate one.
• Training settings you must manage. Consumer ChatGPT and Claude let users switch off use of chats for model training (ChatGPT's switch is called "Improve the model for everyone" and is on by default for Free and Plus), but each setting lives in each person's own account. A policy that depends on every preparer finding a toggle is fragile.
• Personal accounts. Free tools sit on individual logins. When a preparer leaves, the history stays with her account, not with the practice.
• Retention you do not control. Chats are kept for periods set by the vendor. Temporary chat modes shorten that, but ChatGPT's Temporary Chat can still be kept for up to 30 days.
• No admin view. You cannot see who used what, or export a record for your risk assessment.

Business plans address some of these. ChatGPT Business and Claude Team are listed at $20 a user a month billed annually or $25 monthly (for teams of two or more), and both state that business content is not used for training by default. That is a better footing, but it is a footing, not a clearance.

Sort every task into three tiers

A practice is better served by a simple rule than by case-by-case worry. Use three tiers and give every task a home.

Twelve tasks, sorted, for a three-preparer practice

An illustration. A three-preparer practice with about 380 returns a year sorts its tasks like this. Use it as a template and adjust.

Six greens, one amber and five reds. The greens are where a free tier earns its place: they save time and expose nothing. The reds are where most of the real time savings live, which is why the practice must decide how to handle them rather than ignoring them.

A free-tier prompt and what comes back

Here is a green-tier example, the missing-documents email. Note that it contains no client facts.

Write a short, polite email to a client who has not sent their documents for their tax return. Placeholders: [first name], [deadline date], [your name]. Ask for: income statements from employers, interest and dividend statements, any self-employment income and expenses, and mortgage interest statements. Keep it under 130 words, friendly but clear about the deadline.

An illustrative reply looks like a four-sentence email opening with "Hi [first name]" and a list of the four document types, closing with the deadline. What you would fix: the assistant may add a sentence promising a refund or a particular outcome, or invent an incentive. Delete anything that implies advice on the return itself. Then fill the placeholders in your practice software, not in the AI tool.

The de-identification trap

The amber tier is where practices go wrong. "I'll just remove the name" feels safe. It often is not.

Take a hypothetical: a client who is a self-employed signage fitter, with two dependants, an unusual vehicle deduction and a one-off sale of equipment in March for $14,200. Delete the name and you still have a combination that someone in a small community, or the client themselves, would recognise. Return information is wider than a name and number.

A safer way to build an amber scenario is to start from the question, not from the file:

1. Write the question in general terms first: "How is a mid-year sale of business equipment treated for a sole proprietor?"
2. Invent a round, different figure and a different trade. Do not copy dates, employers or amounts.
3. Check that a colleague could not tell which client prompted it.
4. Have a partner approve the prompt before it goes into a free tool, and keep the approved wording in a log.
5. Check the answer against the primary source and your own judgement before it goes anywhere near a return.

If the scenario is too tied to the client to disguise, it is red. That is not a failure; it is the rule working.

Moving red tasks to a paid plan: the cost and what it leaves open

Suppose the three-preparer practice wants AI help with prior-year summaries and notice letters. The monthly arithmetic, using list prices, is simple:

At roughly two dollars a return, the price is not the obstacle. What a paid plan does not do is settle the legal position. Before you put return information into any assistant, the practice should do five things:

1. Decide whether the use is a disclosure and, if so, draft the consent. It must name the recipient, the purpose and the information, and be signed before the disclosure.
2. Read the vendor's terms for retention, training, sub-processors and where data is processed. Questions are in what to ask before buying AI that touches client data.
3. Switch on multi-factor authentication and use practice-owned accounts, not personal ones.
4. Add the tool to the written security plan with its permitted data categories.
5. Train staff and keep a short record of the training.

For the accountancy angle on free plans, when client work needs a paid plan covers similar ground for bookkeepers and accountants.

Consent wording to take to your adviser

Consent that does not meet the regulation is not consent. The wording below is a sketch only, for your adviser to check against the current rules and the IRS format guidance for Form 1040 filers:

I, [client name], authorise [firm name] to disclose my tax return information, limited to [describe: for example, income documents and prior-year return], to [vendor name], a provider of AI software, for the purpose of [for example, drafting summaries and correspondence about my return]. I understand I am not required to give this consent and that my tax services do not depend on it. This consent lasts until [date] unless I withdraw it in writing. Signed: ______ Date: ______

Several features matter: it names the recipient, limits the data, states the purpose, says consent is voluntary and sets an end date. Give the client a copy when they sign. A single blanket document for every use is not the right shape, since uses and disclosures need separate documents.

A page for the security plan

Whatever you decide, write it down. A one-page AI section for your plan can be as short as this:

AI tools (approved list reviewed each January by [Qualified Individual])
- Free-tier assistants: allowed only for Green-tier tasks (no client information). Training setting off. No uploads of client documents.
- [Business plan name]: allowed for [categories], only with signed client consent, practice-owned login and multi-factor authentication.
- Anything else: not allowed for client data. Ask before trying a new tool.
- Mistakes: report to [name] the same day. Do not delete the chat; note the tool, time and what was entered.
- Vendor review: terms, retention, training and processing location checked before approval and at each renewal.

Pair it with a staff policy. An AI acceptable use policy for a small professional firm has a fuller template, and what to check in an AI tool's privacy terms helps with the vendor review.

Worked example: one notice letter, handled two ways

An illustration. A client receives a notice proposing an adjustment of $3,850 to a self-employed return. The preparer needs to send a response letter within a few weeks.

Way one, the free tier. The preparer pastes the notice text and the relevant Schedule lines into a free assistant, asks for a draft response and edits it. It takes 20 minutes. The notice carries the client's name, identification number and address, so the practice has just given return information to a consumer tool with no consent and no record. If the client ever asks where their data went, the honest answer is "we do not know".

Way two, a split. The preparer writes the facts into her own notes. She asks a free assistant, in general terms, for the structure of a response to a proposed adjustment on a business expense: heading, summary of the disagreement, supporting documents list, request for reconsideration, closing. No names, figures or notice numbers. She then fills in the specifics in her word processor and checks every statement against the file. It takes about 35 minutes, 15 more than way one.

Fifteen minutes is the price of keeping the client's data out of the tool. For a practice that handles notices rarely, that is an easy trade. For one that handles dozens in a season, that is the argument for a paid plan with consent and a documented process. Either way the decision is made deliberately.

What to tell a client who asks whether you use AI

Clients are starting to ask. Having an honest line ready helps. Adapt this wording, and check it fits what you actually do:

We use AI tools for general tasks such as drafting standard emails and checklists. We do not put your tax return information or documents into AI tools unless you have given us signed consent that names the tool and the purpose. If we ever ask for that consent, you are free to say no and we will do the work without it. Every return is reviewed by a person before filing.

The final sentence matters most. A client reassured that a qualified person owns the return rarely objects to a drafting tool.

A fifteen-minute staff briefing script

Busy-season rules fail when they are only in a document. Say them aloud once:

1. "Free AI tools are for work with no client in it. If you could not show the text to the client, it does not go in."
2. "Never upload a client document, a screenshot of a return or a transcript of a client call."
3. "If you need a worked example, build it from scratch with different numbers, and ask a partner first."
4. "Check every answer against the primary source. These tools sound confident when they are wrong."
5. "If you slip, tell me the same day. A reported mistake is a small problem; a hidden one is not."

Ask each person to repeat rule one in their own words. If they cannot, the rest will not hold.

A mistake in busy season

In late March a junior preparer at a practice is stuck on a partnership allocation. She photographs a client's schedule and uploads it to a free assistant on her own login to "explain what this line means". The answer is helpful. The upload contained the client's name, tax identification number and income figures.

Nobody noticed for a week. The partner found out when the preparer mentioned the "AI explanation" in a review. The practice now had a possible unauthorised disclosure, no consent, an account it could not see or delete, and a security plan that did not mention AI. The partner's first steps: record exactly what was uploaded and when, ask the preparer to delete the chat and check the training setting, speak to the practice's adviser about the position, and add a line to the plan and the training.

The fix going forward was cheap: a green-tier "explain a line" prompt written in general terms, a rule that source documents never go into free tools, and a reminder at the start of busy season. The damage came from a rushed moment, so the rule has to be easy enough to follow in one.

Checking that the rule is working

A rule nobody tests drifts. Once a quarter, spend thirty minutes on three checks. First, ask each preparer to show you, on screen, which AI tools they have logged in to and which training setting each one uses; do not rely on memory. Second, read ten chats from a green-tier account that the preparer is happy to share, and look for anything that slipped in: a name, an amount, a pasted email. Third, compare the approved-tools list in the plan with what you found in the first check.

Any gap is a prompt to fix the rule, not to blame someone. If three people independently pasted a client's email into a free tool because the practice had no good way to draft replies, the answer might be a template library or a paid plan, not a stricter memo. Record the date of each check and what changed, so that the plan shows a living process rather than a document written once in January.

A first week plan

1. Monday. List the AI tools staff already use, including personal accounts. Ask honestly; the answer is likely longer than you expect.
2. Tuesday. Sort your tasks into green, amber and red using the table above.
3. Wednesday. Check the training setting on every free account used for green work. Turn it off.
4. Thursday. Add the AI page to the security plan and name an owner.
5. Friday. Brief staff for fifteen minutes and put the one-page rule where they can see it.

Then decide, with your adviser, whether the red tasks are worth moving to a paid plan with consent. If you would like help sorting your practice's tasks and drafting the wording, that is what my AI implementation consultation covers; you can start at the contact page. Keep the legal and tax conclusions with your own adviser.

Free AI and client data: follow-up questions

Does a paid AI plan make client data safe to use?

No. A business plan that does not train on your content by default is a better base, but it does not remove the Section 7216 question or your written security plan duties. You still need to decide whether the use is a disclosure, get consent if it is, and record the tool in your plan. A paid plan reduces one risk, not all of them.

Is removing the client's name enough?

Usually not on its own. Income figures, dependants, employer names, unusual transactions and dates can still identify a person, and the return information rules cover more than names. Replace identifiers and also round or alter figures, or keep the example hypothetical from the start. When in doubt, treat it as client data.

Can I use free AI to explain a tax rule to a client?

Yes, if you ask about the rule in general terms and then check the answer against the primary source before relying on it. Do not paste the client's figures into the question. Free assistants can state rules with confidence and be wrong or out of date, so verification is part of the job, not an optional extra.

What should the written security plan say about AI?

Name each approved tool, say what categories of data each may receive, ban unapproved tools for client data, assign an owner for reviewing vendors and say how staff report a mistake. Keep it to a page. Template plans may not mention AI at all, so you will need to add it yourself.

Further reads

• Free AI Tools for Accounting: Which Jobs Free Plans Handle — Which accounting jobs the free plans handle.
• Can AI Summarise IRS Transcripts for a Tax Resolution Case? — AI summaries of transcripts in tax resolution work.
• Shadow AI: How to Stop Staff Pasting Client Data Into Free Tools — Stop staff pasting client data into free tools.
• AI Receptionist for Accounting Firms: Tax-Season Calls and Cost — Handling busy-season calls with an AI receptionist.
• How to Write an AI Usage Policy for Your Small Business — How to write an AI usage policy for a small business.
• Free AI for HR Professionals: Are Free Plans Safe With Staff Data? — The same free-tier question for staff data.
• Tax Services AI for a Small Practice: Intake, Chasing and Review Notes — A three-part workflow for a small tax practice: intake checklist, document-chasing emails and reviewer notes, with a named person owning each step.
• Tax Workpaper Prep With AI: Tie the Return Back to Source Documents — How AI can draft a source-to-return tie-out schedule for a tax return, and the reviewer checklist that proves each figure before sign-off.
• Using AI in Financial Planning: Tax Return Data Into a Client Plan — A planner's source-to-plan workflow for tax-return figures, with a checked example and a client question log.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Internal Revenue Code section 7216 and the regulations at 26 CFR 301.7216-2 and 301.7216-3 (Cornell LII), the FTC's Safeguards Rule business guidance, IRS Publications 5708 and 4557, ChatGPT Business and Claude Team pricing pages, and OpenAI's data controls article, all read 4 November 2026. Not legal or tax advice.