How Can Accountants Use ChatGPT Safely and Effectively in 2026?
A ready-to-adapt policy for how accounting and bookkeeping firms should and should not use ChatGPT with client data.
Table of contents
Your staff are already using ChatGPT. They are drafting emails with it, summarizing memos, and looking up tax guidance. Most of them started months ago. The question is not whether your firm will use AI. The question is whether you will have a written policy before something goes wrong.
Most accounting and bookkeeping firms still have no formal AI use policy. That gap creates real risk. One associate pastes a client’s profit-and-loss statement into a free-tier chatbot, and the firm has a data governance problem it never saw coming.
The fix is not a ban. Bans do not work when the tool is free, fast, and two clicks away. The fix is a clear, written policy that tells your team what is acceptable and what is off-limits.
TLDR: ChatGPT and similar AI chat tools are genuinely useful for drafting, summarizing, and brainstorming. They are not safe for client-identifiable data on consumer-tier plans. Write a policy, share it with everyone, and revisit it quarterly.
Why this needs to be written down
“Use common sense” is not a policy. It is a liability waiting to happen. Every person on your team has a different idea of what common sense means when it comes to new technology.
One associate might think pasting a trial balance is fine as long as client names are removed. Another might upload an entire tax return to ask a follow-up question. Without a written standard, both believe they are being reasonable.
Written policies also protect the firm. If a data incident occurs, regulators and insurers will ask what controls were in place. “We told everyone to be careful” is not a satisfying answer.
A policy does not need to be long. It needs to be specific, shared, and signed. Even a one-page document is better than a verbal reminder at the last team meeting.
What is generally safe to use ChatGPT for
These use cases involve no client-identifiable data and carry low risk when handled on any tier of a major AI chat tool.
- Drafting client-facing emails. Write the structure and tone first, then add client-specific details yourself after the draft is generated.
- Summarizing public tax guidance. Ask ChatGPT to explain a new IRS notice or FASB update in plain language. The source material is already public.
- Explaining technical concepts. Need a quick refresher on Section 174 capitalization rules? AI tools handle this well. Always verify against primary sources.
- Building memo and report structures. Use it to outline a research memo or advisory letter. You fill in the substance and citations.
- Brainstorming and checklists. Generate a year-end close checklist or a list of common deductions for a specific industry. Review before using.
The common thread is that none of these tasks require you to share anything confidential. The AI is helping with structure and language, not making decisions about real client situations.
What is not safe on a consumer-tier tool
Consumer-tier AI tools - including ChatGPT Free and Plus plans - do not offer the data protections that accounting firms need when handling client information.
- Pasting profit-and-loss statements or balance sheets. Even with the client name removed, financial data combined with industry details can be identifiable.
- Uploading tax returns or workpapers. These documents contain Social Security numbers, EINs, income figures, and other protected information.
- Asking for tax advice using real client numbers. The tool may use your input for model training. You have now shared client data with a third party.
- Pasting engagement letters or fee agreements. These contain client names, service scope, and fee structures that are confidential.
- Using free-tier tools for any client data. Free tiers almost always permit the provider to use your inputs for training. This is a non-starter for client confidentiality.
The risk is not theoretical. Under most state privacy laws and the AICPA Code of Professional Conduct, you have a duty to protect client information. Feeding it into a consumer chatbot likely violates that duty.
A policy template you can adapt
Below is a starting point. Replace the bracketed placeholders with your firm’s details. Have your team read and sign it.
[Firm Name] Artificial Intelligence Acceptable Use Policy
Approved tools. Only AI tools on the firm’s approved list may be used for work purposes. As of [date], the approved tools are: [list tools and tier, e.g., ChatGPT Team, Microsoft Copilot for Microsoft 365]. Personal accounts on free tiers are not approved for any work use.
Prohibited inputs. Staff must not enter client-identifiable information into any AI tool. This includes names, tax ID numbers, financial statements, engagement terms, and any data that could identify a specific client. When in doubt, do not paste it.
Review requirement. All AI-generated content used in client deliverables must be reviewed by a [manager / partner / senior associate] before delivery. AI output is a draft, never a final product.
Accuracy obligation. The person submitting the deliverable is responsible for its accuracy, regardless of whether AI assisted in creating it. AI-generated tax positions, calculations, and citations must be independently verified.
Training and updates. All staff will complete a [15/30/60]-minute orientation on this policy within [30] days of hire or policy update. The firm will review and update this policy at least [quarterly / semi-annually].
Reporting. If you believe client data has been entered into an unapproved AI tool, report it to [name or role] within 24 hours. Early reporting helps the firm respond quickly and limits potential harm.
This template covers the essentials. Your firm may want to add sections on record retention, specific regulatory obligations, or approved use cases by department. Start simple and expand over time.
Rolling it out without a lecture
Nobody wants another compliance lecture. Here are three ways to make this land with your team.
-
Show, then tell. Start the meeting by demonstrating a genuinely useful ChatGPT workflow - like drafting a client update email in 30 seconds. Once people see the value, they are more receptive to hearing about the boundaries.
-
Make the approved tools easy to access. If the firm is paying for ChatGPT Team or Microsoft Copilot, make sure every team member has a login before the policy meeting. Removing friction is the best way to keep people off free-tier alternatives.
-
Keep it short and revisit often. A one-page policy reviewed quarterly is more effective than a ten-page document that nobody reads after onboarding. AI tools change fast. Your policy should keep pace.
Where AI tools built for finance fit in
General-purpose chatbots are not the only option. Several AI tools are built specifically for accounting and bookkeeping workflows. They carry a different risk profile because they are designed with client data handling in mind.
Karbon uses AI to help manage practice workflows and client communications within a system that is purpose-built for accounting firms.
Digits applies AI directly to financial data analysis and reporting, with controls designed for the sensitivity of that data.
Intuit Assist is embedded within QuickBooks and operates under Intuit’s existing data agreements, which most firms already have in place.
These tools are not a replacement for a general AI policy. But they reduce the temptation to paste client data into a consumer chatbot because the specialized functionality is already available inside a controlled environment.
For a broader look at AI tools purpose-built for this space, see our guide to the best AI bookkeeping tools.
Settings to check today
Each major AI platform has settings that control whether your data is used for model training. Check these before your next client interaction.
| Platform | Setting to find | Where to find it | Default |
|---|---|---|---|
| ChatGPT | “Improve the model for everyone” | Settings > Data Controls | On (Free/Plus) |
| Claude | “Allow training” toggle | Privacy settings in account | Off (Pro), varies by plan |
| Gemini | “Gemini Apps Activity” | Google Account > Data & Privacy | On |
| Microsoft Copilot | Data sharing and training | Admin Center > Copilot Settings | Off (Enterprise), On (Consumer) |
Even on paid tiers, verify these settings. Defaults change with updates, and a new team member’s account may not inherit the firm’s configuration. Build a quarterly check into your IT review process.
Note that turning off training does not eliminate all risk. The provider still processes your input on their servers. For truly sensitive data, a self-hosted or on-premise solution may be the only appropriate option.
Conclusion
AI chat tools are here to stay in accounting. The firms that benefit most will be the ones that set clear rules early - not the ones that pretend the technology does not exist or panic into blanket bans.
Start with a written policy. Keep it short and specific. Make the approved tools easy to access so staff do not reach for unapproved alternatives. Review the policy regularly as the tools and regulations evolve.
The goal is not to control how your team thinks about AI. It is to make sure the firm’s clients stay protected while your team gets the productivity benefits that these tools genuinely offer.
Frequently asked questions
Is it illegal to use ChatGPT for tax work?
Not inherently. There is no law that bans accountants from using AI tools. However, professional standards require you to protect client confidentiality and verify the accuracy of any advice you give. Using ChatGPT as a drafting aid is different from relying on it for tax positions without independent verification.
Does ChatGPT’s paid Team or Enterprise tier change the risk?
Yes, meaningfully. OpenAI’s Team and Enterprise tiers include contractual privacy commitments that your data will not be used for model training. They also offer admin controls, audit logs, and SSO. This does not eliminate all risk, but it significantly reduces the data governance concern compared to free or Plus tiers.
Should we just ban AI tools entirely to be safe?
Banning rarely works. Staff will use personal devices and free accounts outside your visibility. A ban also puts your firm at a competitive disadvantage as peers adopt these tools for efficiency. A clear acceptable-use policy is more realistic and more protective than a prohibition you cannot enforce.
Who should approve which AI tools the firm uses?
This should be a joint decision between firm leadership and whoever handles IT and compliance. At smaller firms, that may be one or two people. The key is that someone is explicitly responsible for evaluating tools, approving them, and reviewing that approval periodically.
What about AI features already built into QuickBooks or Xero?
Embedded AI features in platforms like QuickBooks or Xero operate under those vendors’ existing data processing agreements. Your firm likely already has terms in place with these providers. The risk profile is different from pasting data into a standalone chatbot. Still, review the vendor’s AI-specific terms to understand what data is used and how.
Can we use ChatGPT to draft client deliverables?
Yes, with guardrails. Use it to create structures, outlines, and initial language. Do not feed it client-specific data to generate the draft. A qualified professional must review every deliverable before it goes to a client. The person who signs off is responsible for accuracy, regardless of whether AI helped produce it.
