ChatGPT for Tax Preparation: Chat, Work and Business Data

Each time a staff member opens ChatGPT for prep work, they make two choices. One is the mode: Chat or Work. The other is the folder, if any, that ChatGPT gets to open. Before either choice, there's one fact to check: which plan the Firm's account is on.

Vendor claims below are quotations or narrow paraphrases from two named OpenAI pages. Each workflow below is a choice the Firm makes and records. A Preparer checks each ChatGPT draft against its sources. The Reviewer reviews the workpaper and signs off.

For a comparison with another vendor, see the side-by-side. This piece stays with ChatGPT.

Check which plan the account is on

OpenAI's ChatGPT agent help page has a "Safety & privacy" section with a part headed "Business, Enterprise, and Edu plans". That part says: "By default, we do not use your business data for training our models, including data accessed during agent mode sessions."

It covers business data on Business, Enterprise, and Edu plans. It doesn't cover a Plus or Pro account. The same page has a separate part for those users, which says: "For Pro and Plus users, your data ... is used in accordance with OpenAI's privacy policy". And the business line is about training only.

Three steps, in order:

1. Ask each staff member which login they use for work, and match it to a plan the Firm holds.
2. Read OpenAI's current terms for that plan before any client file goes near it.
3. Write down which plan is cleared for client files. Keep every other login to Doe files.

Chat and Work are two different jobs

OpenAI's ChatGPT quickstart page describes both modes. For Work: "For research, analysis, or deliverables such as documents, presentations, spreadsheets, and Sites, select ChatGPT, then switch to Work at the top of the new chat page, above the composer." For Chat: "For a quick question or chat, select ChatGPT, then select Chat in the switcher at the top of the new chat page, above the composer."

The vendor ties Work to deliverables and Chat to quick questions. A Firm can turn that into a standing choice per task. Here's one way to set it. Every cell below is the Firm's choice, not something OpenAI says.

Neither quote says which mode can open a folder, or what plan a mode needs. Don't assume either. Before staff use a row with client files, try it on Doe files and note which mode opened the folder, what changed in it, and whether the output matched what you expected. Where AI fails in tax prep covers how to build that kind of test.

The folder ChatGPT can modify

OpenAI's ChatGPT quickstart page says: "Start a chat, create a project, or open a folder. ChatGPT can read and modify files in the folder you choose." It also says: "The ChatGPT desktop app is available for macOS, Windows, and Linux. Use it for projects, local files, longer tasks, and quick chats."

Since ChatGPT can modify files in that folder, make a copy for this job alone: one client, one period. Never point it at the original.

Then set four things for that copy:

• What goes in: only the documents the task needs. For a missing-items request, the Firm can choose the organizer checklist and documents received so far, then decide whether the task also needs a prior-period return.
• What stays out: other clients' files, the Firm's originals, engagement letters and billing the task doesn't use, and any file holding the Firm's passwords, credentials or admin settings.
• The name: client, period and task, so anyone can tell what it holds at a glance. For example, John-Doe_Period-B_missing-items.
• When the job ends: name who deletes or archives the copy, and when. File the checked output in the workpaper, not in the copy folder. Put the choice in the Firm's procedures so it doesn't rest on memory.

If staff open local files in the desktop app, confirm the computer holding the copy is one the Firm approves for client data.

Older guides that name agent mode

If a guide or an internal note tells staff to use ChatGPT agent or agent mode, check it against the vendor. OpenAI's ChatGPT agent help page now opens with this line: "ChatGPT agent is no longer available. Use ChatGPT Work for longer, multi-step tasks and finished deliverables."

Act on that line:

• Search the Firm's procedures, saved prompts and training notes for "agent mode" and "ChatGPT agent". Retire or rewrite instructions that tell staff to use ChatGPT agent or agent mode.
• Before staff follow any outside guide, re-check its steps against the current quickstart page. If a step names a mode or button the quickstart doesn't, stop and ask whoever owns the procedure.
• Re-run the Doe test after any rewrite.

One week at Doe & Co. CPA

Sam Doe owns Doe & Co. CPA. The Firm's account is on a Business plan, and Sam has set a mode and a folder for each recurring task. The table tracks client John Doe's file. Jane Doe is the Preparer; Sam is the Reviewer.

Jane opens every row of each draft against its source file before it moves on. Jane enters the checked figures in the Firm's tax software, which computes the return. Sam, as the Reviewer, reviews the workpaper and signs off.

Write the Firm's instruction card

For each recurring task, fill in one card before staff run it. Keep the cards wherever the Firm keeps its procedures. These are the fields, not prompt wording; the Firm writes its own.

• Task name: the job, in the words staff already use.
• Mode: Chat or Work.
• Approved plan: the plan, from the first section above, that this task may run under.
• Folder name: the copy-folder pattern, or "none" when the Firm chooses a task without file access.
• What the output must include: the source file and page for every figure, with "could not find it" accepted instead of a guess.
• Who checks: the Preparer who opens each cited page, and the Reviewer who signs off on the workpaper. For reviewers sets out the Reviewer's part.
• Where the checked output is filed: the workpaper location, plus what happens to the copy folder.

Name who re-reads both OpenAI pages and how often, and have that person update any card the pages now contradict.

Where TracePrep fits

TracePrep's one shipped workflow today is the Reasonable Compensation Study for an S-corporation shareholder-employee. It's a different job from this workflow, and it keeps the same kinds of records:

We've built a TracePrep plugin that runs these steps in your Firm's workflow.

This is general information, not tax or legal advice. Your licensed professional and your Firm's Reviewer are responsible for every return and every conclusion. Keep client data inside the systems your Firm has approved. As general guidance only, not advice: if you use an AI tool with client data, firms generally look for a business or commercial plan whose terms cover how that data is handled; read the vendor's current terms before you do.

Join the waitlist