How should accountants use ChatGPT?

For the language work around the numbers, never for the numbers. In a small firm as of 2026 that means five jobs: drafting client emails and reminders, turning notes into engagement letters and proposals from the firm's own templates, first-draft research memos you then verify against the primary source, writing procedure and onboarding docs, and explaining or building spreadsheet formulas. Give it your firm's context first, verify anything with a citation, and keep client identifiers out of any consumer plan.

Is ChatGPT safe for client data at an accounting firm?

Not on a consumer plan by default. As of 2026 OpenAI's published shape is that Free, Plus, and Pro conversations can be used to improve models unless you turn that off in settings, while Team, Enterprise, and API usage is not used for training by default. So the safe pattern is: no client names, SSNs, EINs, account numbers, or financial statements on a personal login, ever; if the firm wants to paste real client material, it moves to a workspace plan with the data terms in writing, and it reads the current policy page before it does.

Can ChatGPT do bookkeeping or tax calculations?

It can explain them, draft the memo about them, and write the formula that computes them. It should not be the thing that computes them. A language model predicts text; it does not carry a ledger, it will confidently misadd a column, and it cannot sign anything. Keep arithmetic in the accounting software and the spreadsheet, keep filings in the filing software, and use ChatGPT for the words around the work.

Is Claude or ChatGPT better for accountants?

Both do the five jobs on this page well. Claude tends to win on long-document work and on following a firm's style once it has the context loaded; ChatGPT wins on ecosystem, plugins, and the fact that half your clients already use it. The prices are nearly identical as of 2026: Claude Pro $20 and Team $20 a seat billed annually ($25 monthly); ChatGPT Plus $20 and Team $25 a seat. Run three of your real tasks through both for a week and let the output vote. The tool is the small decision.

What should an accounting firm never paste into ChatGPT?

On any consumer plan: client names, Social Security numbers, EINs, bank and account numbers, full financial statements, tax returns, payroll files, anything under an NDA, and anything you would not want in a training set. On a workspace plan with written no-training terms the list shrinks to what your engagement letters and state rules allow, which for most small firms still means de-identify first. Write the list down, put it in the onboarding doc, and treat it like the password policy.

Will ChatGPT replace accountants?

No. It replaces the hour a CPA spends drafting the same client email for the fortieth time, the afternoon a bookkeeper spends writing the month-end procedure nobody documented, and the first pass at a research memo. The judgment, the signature, the client relationship, and the liability stay with the human. Firms that use it well get the same work out the door with fewer late nights; firms that skip the never-paste list get a very fast way to make an expensive mistake.

Justin McKelvey

Fractional CTO · 15 years, 50+ products shipped

AI for Business • 6 min read • Published Sep 07, 2026

ChatGPT for Accountants: The Five Jobs It Does in a Small Firm, the Never-Paste List, and the $20 Question (2026)

TL;DR: ChatGPT for accountants works when you use it for the language around the numbers and never for the numbers. In a small firm as of 2026 that is five jobs: client communication, engagement letters and proposals from your own templates, first-draft research memos you verify, procedure and onboarding docs, and spreadsheet formulas. It must never be the thing that does the arithmetic, and client data must never touch a consumer login. The $20 Plus plan is not the decision; the 20-minute setup and the never-paste list are. If the firm wants to paste real client material, it moves to a workspace plan with written no-training data terms, which is Team at $25 a seat. Here is the whole thing, written for the CPA who'll actually use it and the owner who has to approve it.

Why "ChatGPT for accountants" is usually asked wrong

Line one of every guru post on this topic is "50 ChatGPT prompts for accountants." I've been inside enough small firms to know what happens next: someone pastes a client's trial balance into a personal ChatGPT account to "check it," gets a confident wrong answer, and the partner bans the tool by Friday. Two mistakes in one move: using a language model for arithmetic, and putting client data on a consumer plan. Neither is a ChatGPT problem. Both are setup problems, and the fix costs about twenty minutes, not a prompt list. The firms that get real hours back are the ones that decided what the tool is for before anyone opened it.

The five jobs ChatGPT does well in a small firm

1. Client communication. The fortieth "we still need your 1099s" email, the reply to a client who wants to know why their estimated payment went up, the year-end reminder sequence. Give ChatGPT your firm's tone in a short standing note, tell it what you're asking for, and the draft is 80% done in seconds. This is where most small firms recover the most hours, and it's the safest job on the list because it involves no client numbers.

2. Engagement letters and proposals from your own templates. Paste your template, describe the engagement, get a clean first draft in your language. Do not let it write the scope from scratch; a model does not know your state's rules or your firm's liability appetite. It is a very fast assembler of what you already have.

3. First-draft research memos. "Summarize the treatment of X and give me the sections to check" is a legitimate starting point. Then you go to the primary source, because as of 2026 every model still invents citations under pressure. The rule is the same one law firms learned the hard way after the 2023 sanctions cases: it drafts, a human verifies, nothing with a citation leaves the building unverified.

4. Procedure and onboarding docs. The month-end close checklist that lives in one person's head, the "how we onboard a new payroll client" doc nobody wrote. Dictate it in five messy minutes, have ChatGPT structure it, edit once. This is the job owners underrate and it's the one that makes the firm sellable.

5. Spreadsheet formulas and explanations. "Write the Excel formula that flags any vendor paid twice in 30 days" or "explain what this nested IF does." It's excellent at this. Notice the pattern: it writes the formula, the spreadsheet computes the result. The model never touches the total.

The two jobs it must never touch

The numbers. A language model predicts text. It does not carry a ledger, it will misadd a column with total confidence, and it cannot tell you it was wrong. Reconciliation, tax computation, anything that ends in a figure a client relies on stays in the accounting software and the spreadsheet. ChatGPT can explain the figure, draft the memo about the figure, and write the formula that produces the figure. It is not the calculator.

The signature. Attestation, opinions, filings, anything your license is attached to. Obvious when written down, routinely blurred in practice when the draft looks finished.

The never-paste list

On any consumer plan (Free, Plus, or Pro), the list is short and absolute: client names, Social Security numbers, EINs, bank and account numbers, full financial statements, tax returns, payroll files, anything under an NDA, and anything you would not be comfortable seeing in a training set. As of 2026 OpenAI's published shape is that consumer conversations can be used to improve models unless you turn that off in settings, while Team, Enterprise, and API usage is not used for training by default. Read the current policy page before you rely on that sentence; it has changed before and it will change again. On a workspace plan with the terms in writing, the list shrinks to what your engagement letters and your state board allow, which for most small firms still means de-identify first. Write the list down. Put it in the onboarding doc from job number four. Treat it like the password policy.

The $20 question: which plan does the firm need?

Free is enough to learn on and to do jobs one and four with zero client data. Plus at $20 a month removes the usage ceilings and gets one person the current models. Team at $25 per seat (two-seat minimum) is the first tier built for a firm: a shared workspace, admin controls, and the no-training-by-default data terms that let you stop pretending the never-paste list will be followed perfectly by everyone forever. Pro at $200 is for heavy agentic use; I have not met the small firm that needs it. The honest answer for a three-person firm that touches client data is Team, and it is the small decision. Whether ChatGPT drafts your client emails or runs an installed workflow that drafts, routes, and files them without you standing over it is the big one.

Claude or ChatGPT for accountants?

I get this one weekly, and the honest answer is that both do the five jobs. Claude tends to win on long-document work and on holding a firm's style once its context is loaded, which is why the accounting-firms guide has a section on it. ChatGPT wins on ecosystem and on the fact that half your clients already use it. Prices are nearly identical as of 2026: Claude Pro $20 and Team $20 a seat billed annually ($25 monthly) for teams of two or more; ChatGPT Plus $20 and Team $25 a seat. Run three of your real tasks through both for a week and let the output vote. The full comparison for a small business is in Claude vs ChatGPT for small business.

The 20-minute setup that beats a prompt list

Give it your context once, in writing: who the firm serves, how you sign emails, the three things clients always ask, the never-paste list. Store that as the standing instruction (Projects or a custom GPT on ChatGPT; a Project on Claude). Then every draft starts from your firm instead of from the internet's average accountant. That is the whole trick, and it is why "50 prompts" lists don't work: they give the tool nothing to be specific about. The strategy-level version of that setup is free on this site; the templates, skill packs, and exact prompts I install are what clients pay for, and I'm not going to pretend otherwise.

The owner's version

If you own the firm, the decision isn't whether ChatGPT is good for accountants. It is. The decision is whether AI runs repeated tasks in your firm without you standing over it, with a human approving anything that goes to a client. That is the difference between a $20 subscription three people use inconsistently and a system. The self-serve path to the system is the CFSB Playbook ($397). The done-for-you version is the Claude for Small Business install (from $4,500, about two weeks and three hours of your time), which starts with a plan: the AI Readiness Assessment ($2,500, credited in full against a build within 90 days). If you'd rather find out in thirty minutes whether any of this fits your firm, book a free strategy call. No pitch, and if the honest answer is "fix your intake first," I'll say exactly that.

Free Resource

Get the Free AI Content Toolkit

The exact system I use to turn one idea into a month of content — atomization framework, voice template, prompt library, weekly system.

Frequently Asked Questions

How should accountants use ChatGPT?: For the language work around the numbers, never for the numbers. In a small firm as of 2026 that means five jobs: drafting client emails and reminders, turning notes into engagement letters and proposals from the firm's own templates, first-draft research memos you then verify against the primary source, writing procedure and onboarding docs, and explaining or building spreadsheet formulas. Give it your firm's context first, verify anything with a citation, and keep client identifiers out of any consumer plan.
Is ChatGPT safe for client data at an accounting firm?: Not on a consumer plan by default. As of 2026 OpenAI's published shape is that Free, Plus, and Pro conversations can be used to improve models unless you turn that off in settings, while Team, Enterprise, and API usage is not used for training by default. So the safe pattern is: no client names, SSNs, EINs, account numbers, or financial statements on a personal login, ever; if the firm wants to paste real client material, it moves to a workspace plan with the data terms in writing, and it reads the current policy page before it does.
Which ChatGPT plan does a small accounting firm need?: Free is enough to learn on. Plus at $20 a month removes the usage ceilings and gets the current models for one person. Team is $25 per seat with a two-seat minimum and is the first tier built for a firm: a shared workspace, admin controls, and the no-training-by-default data terms. Pro at $200 is for heavy agentic use and almost no small firm needs it. The honest answer for a three-person firm that touches client data is Team, and the plan is the small decision; the setup is the big one.
Can ChatGPT do bookkeeping or tax calculations?: It can explain them, draft the memo about them, and write the formula that computes them. It should not be the thing that computes them. A language model predicts text; it does not carry a ledger, it will confidently misadd a column, and it cannot sign anything. Keep arithmetic in the accounting software and the spreadsheet, keep filings in the filing software, and use ChatGPT for the words around the work.
Is Claude or ChatGPT better for accountants?: Both do the five jobs on this page well. Claude tends to win on long-document work and on following a firm's style once it has the context loaded; ChatGPT wins on ecosystem, plugins, and the fact that half your clients already use it. The prices are nearly identical as of 2026: Claude Pro $20 and Team $20 a seat billed annually ($25 monthly); ChatGPT Plus $20 and Team $25 a seat. Run three of your real tasks through both for a week and let the output vote. The tool is the small decision.
What should an accounting firm never paste into ChatGPT?: On any consumer plan: client names, Social Security numbers, EINs, bank and account numbers, full financial statements, tax returns, payroll files, anything under an NDA, and anything you would not want in a training set. On a workspace plan with written no-training terms the list shrinks to what your engagement letters and state rules allow, which for most small firms still means de-identify first. Write the list down, put it in the onboarding doc, and treat it like the password policy.
Will ChatGPT replace accountants?: No. It replaces the hour a CPA spends drafting the same client email for the fortieth time, the afternoon a bookkeeper spends writing the month-end procedure nobody documented, and the first pass at a research memo. The judgment, the signature, the client relationship, and the liability stay with the human. Firms that use it well get the same work out the door with fewer late nights; firms that skip the never-paste list get a very fast way to make an expensive mistake.

More on AI for Business

ChatGPT for Bookkeepers: What It Does in a Small Practice, What It Must Never Touch, and the Plan That Fits (2026)

ChatGPT for bookkeepers in 2026, written for the bookkeeper who will actually use it and the practice owner who has to sign off: the five jobs it does well in a small practice, the two it must never touch, the never-paste list for client data, why the $20 plan is the small decision, and when a Business workspace with written data terms is the right one.

7 min

AI Fluency: What It Means, the 4D Framework, and What It Looks Like in a Real Business (2026)

AI fluency is the ability to work with AI effectively, efficiently, ethically, and safely — Anthropic's free course teaches it as four skills: Delegation, Description, Discernment, Diligence. Here's what each one looks like inside a 10-person business, why a certificate won't change Tuesday, and the 30-day version that does.

9 min

AI for Paralegals: The Four Jobs It Does, the One It Never Touches, and the Never-Paste List (2026)

AI for paralegals in 2026, written for the paralegal who will actually run it and the firm owner who has to approve it: the four jobs AI does in a small firm's back office, the one job it never touches, the never-paste list, what it costs, and the 20-minute setup that survives the supervising attorney.

8 min

AI for Consultants: How a One-Person Practice Runs on Claude Without Losing the Judgment Clients Pay For (2026)

I'm a consultant. Here's how my practice actually runs on AI as of 2026 — the three jobs it does every day, the one job it's banned from, what it costs, and the pricing mistake most consultants make the week they get faster.

7 min

Written by

Justin McKelvey

Fractional CTO & AI consultant in Austin, TX. 15 years building software, 50+ products shipped, $53M+ in client revenue generated. I help $1M–$50M founders ship production software and automate operations with AI — without hiring a full-time executive team.

Work with me

If this was useful, here are two ways I can help:

1) Get the free toolkit 2) Book a strategy call