ChatGPT Prompts for Bookkeepers: The Working Set

Most "ChatGPT prompts for bookkeepers" lists are five generic prompts nobody has run on live books. This is the bookkeeping cut of the prompt library we use inside an operating CPA firm: 30 prompts, grouped by the work you actually do, each with a guardrail note.

One thing up front. These prompts draft, explain, and build checklists. They don't do the books. ChatGPT can't see your bank feed, can't match a deposit to 14 invoices, and won't remember the correction you made yesterday. Use it for the words around the work, and keep the work itself in your tools.

And one rule that never bends: nothing client-identifying goes into the chat. More on that below.

What are ChatGPT prompts for bookkeepers?

ChatGPT prompts for bookkeepers are reusable instructions that turn the chat window into a drafting and explaining tool for bookkeeping work. Think client emails, categorization decision checks, month-end close checklists, and cleanup triage plans. The prompts use bracket placeholders like [CLIENT NAME] and [LINE ITEM] so nothing client-identifying enters the chat. ChatGPT is not a categorizer, a ledger, or a reconciliation tool. It drafts the email, builds the checklist, or explains the concept, and you review everything before it touches a client or a book. Each prompt saves minutes, not hours, but the minutes compound across 15 to 25 client books every month.

Key Takeaways

• 30 prompts, 4 workflow groups. Client communication, categorization sanity checks, month-end close, and cleanup triage. No tax-prep prompts here; this is the bookkeeping set.
• Placeholders are the rule. [CLIENT NAME], [LINE ITEM], [AMOUNT]. Nothing real goes in until you're back in your own email client or ledger.
• Nothing client-identifying enters the chat. No name next to a number, no account numbers, no statement uploads, on any plan tier.
• Prompts draft and explain. They don't do the books. No bank feed access, no deposit matching, no memory of your corrections.
• Every output is a first draft. You review and approve before anything reaches a client.

How to use these ChatGPT prompts for bookkeepers

Each prompt has three parts: a short title, the prompt itself in a blockquote you can copy, and a one-line guardrail note. Swap the bracketed placeholders after the prompt runs, in your own tools, never in the chat.

The prompts come from the library we run at an operating CPA firm. I cut the tax-prep groups (K-1 codes, IRS notices, depreciation) because this is the bookkeeper's set; the tax side of that work lives with TracePrep, our sister product for tax workpaper automation. If you want the deeper story on chat tools in bookkeeping workflows, we've written an honest walkthrough of how to use Claude for bookkeeping that covers the same ground from the tool side.

Group 1: Client communication (8 prompts)

1. Missing-Document Request

Draft a friendly email to [CLIENT NAME OR CODE] requesting their [DOCUMENT TYPE] by [DATE]. Under 100 words, no guilt-tripping. Use a placeholder or internal code, not the real name, until you paste it into your own email client.

2. Uncategorized-Transaction Question Batch

Turn this list of [N] transaction descriptions [descriptions only, no amounts or names] into one short email of questions for the client, grouped so they can answer everything in five minutes. Descriptions only. Strip amounts and anything identifying before pasting.

3. Plain-English P&L Explainer

A [LINE ITEM] rose [PERCENT]% this month for a [BUSINESS TYPE]. Given [REASON CATEGORY], explain the likely drivers in 3 sentences a non-finance owner would understand. Possibilities, not a diagnosis. Feed the reason category, not the client's actual books. Confirm the real cause in the ledger before it goes out.

4. Fee Conversation Opener

Write a direct, non-apologetic email explaining a fee change from $[OLD FEE] to $[NEW FEE] for [SERVICE]. Keep dollar figures as placeholders until final review.

5. Month-End Delivery Note

Draft a short month-end delivery email for a [BUSINESS TYPE] client: books are closed, two things worth their attention, one question I need answered. Under 120 words. Template only. The two things and the question come from your review, not ChatGPT's.

6. Re-Engagement Check-In

Write a check-in email for a client I haven't heard from since [TIMEFRAME], naming a specific reason for the outreach (an upcoming close, a document gap, a season change) and one clear ask. Generic timeframe and reason category. No account details.

7. Scope Decline Email

Draft a polite email declining [SERVICE TYPE] work for an existing client, stating the service boundary and offering a referral category, with transition language for anything in progress. Keep the reason factual. Don't name a referral unless it's someone you've already vetted.

8. New-Client Onboarding Welcome

Write an onboarding welcome email for a new monthly bookkeeping client: what I need by when [DOCUMENT CATEGORIES], how we communicate, and what day their books close each month. Template. Swap in your real process before sending.

Group 2: Categorization sanity checks (7 prompts)

ChatGPT can't categorize your feed. It can pressure-test your reasoning when a transaction is genuinely ambiguous.

9. Category Decision Tree

Build a decision tree for categorizing [EXPENSE TYPE] for a [BUSINESS TYPE]: the questions to ask, in order, before picking an account. Process reasoning only. Your chart of accounts makes the final call.

10. Net-vs-Gross Fee Explainer

Explain how to book a [PROCESSOR] payout that arrives net of fees: what posts to revenue, what posts to fees, and what goes wrong when it's posted gross. General mechanics, not a specific client. Mechanics check, not a booking instruction. Verify against how the processor actually reports.

11. Transfer-vs-Income Flag List

List the signs that a deposit is a transfer, an owner contribution, or a loan rather than income, and the follow-up question to ask the client for each. Loans, draws, and transfers all look alike in a feed. This builds your question list, not your answer.

12. Memo-Line Decode Checklist

Give me a checklist for identifying a bank transaction whose memo only says [GENERIC MEMO PATTERN, like ACH PAYMENT plus a number]: what to check, in order, before I ask the client. Pattern only. Never paste the real memo line with real reference numbers.

13. Chart-of-Accounts Fit Check

For a [BUSINESS TYPE], list the accounts a standard chart of accounts usually misses and the transactions that end up misfiled because of it. General patterns. Compare against the client's actual COA yourself.

14. 1099-Vendor Flag Question Bank

Build the questions I should ask about a new vendor to decide whether to track them for 1099 purposes. Process only, no dollar thresholds. Thresholds change and this isn't the place to get them. Flag the vendor, and let the client's tax preparer own the filing rules.

15. Personal-vs-Business Boundary List

List common transactions that look like business expenses but are usually personal for a [BUSINESS TYPE] owner, and a neutral way to raise each with the client. The phrasing help is the point. Nobody enjoys the 'was this business?' conversation.

Group 3: Month-end close checklists (8 prompts)

16. Close Checklist by Client Type

Build a month-end close checklist for a [INDUSTRY TYPE] business with [NUMBER] accounts and [PAYROLL/INVENTORY/NEITHER]. Order it by what breaks first. Template. Adapt per client without naming them in the chat.

17. Reconciliation Checklist Generator

Build a month-end bank reconciliation checklist for a [ENTITY TYPE] with [NUMBER] accounts, including the checks for items that cleared in the wrong month. Checklist only. Run the actual reconciliation in your software.

18. Trial-Balance Red-Flag Scan

List the 8 most common trial balance red flags to scan for before closing a [ENTITY TYPE]'s books, why each matters, and what evidence to pull for each. General checklist. Never upload an actual trial balance.

19. Journal-Entry Clarity Rewrite

Rewrite this journal entry description for clarity: "[VAGUE DESCRIPTION, vendor and employee names and amounts replaced with placeholders]." Make it specific enough for a reviewer to follow cold. Strip names and figures first. Never the real JE text.

20. P&L Variance Question Bank

A [LINE ITEM] moved [PERCENT]% month over month. Give me the 5 questions to ask before flagging it as an error. Percentage only, never dollar figures tied to a real client.

21. Flux Analysis Question Bank

Generate 6 flux-analysis questions for a [INDUSTRY TYPE] client whose gross margin moved [PERCENT] points this quarter. Percentage and industry only.

22. Accrual Reversal Checklist

Build a checklist of accrual entries that typically need reversing in the first week of a new month for a [ENTITY TYPE], with columns for timing, source workpaper, and your own sign-off. General accounting checklist. Your workpapers hold the real entries.

23. Client Deliverable QC Pass

Review this paragraph for tone, typos, and clarity before it goes to a client: "[PARAGRAPH with names and figures swapped for placeholders]." Flag anything that reads hedgy. Anonymize first, always. Never the real deliverable text.

Group 4: Cleanup triage (7 prompts)

24. Cleanup Scope Questions

I'm estimating a cleanup for a [BUSINESS TYPE] with [N] months of neglected books. Build the list of questions to answer before I estimate hours. Scoping discipline. The estimate itself comes from your rates and the actual mess.

25. Backlog Triage Order

Give me a triage order for clearing [N] months of uncategorized transactions: what to batch first, what to hold for client questions, what can wait until the end. Sequencing help. The judgment calls stay yours.

26. COA Migration Mistake List

List common chart-of-accounts mapping mistakes when moving a client's books from [SOFTWARE A] to [SOFTWARE B]. Software names only. No actual chart of accounts.

27. Duplicate-Transaction Hunt

List the places duplicate transactions hide when a client connected the same bank account through two different feeds, and the order to check them. Checklist. The dedupe itself happens in your ledger.

28. Clearing-Account Investigation

Build a checklist for investigating a clearing account that never returns to zero, and what each leftover balance pattern usually means. Clearing accounts that never hit zero are the tell. This builds your investigation path.

29. Prior-Bookkeeper Handoff Questions

I'm taking over books from another bookkeeper. Build the handoff question list: access, open items, unreconciled periods, and undocumented habits to ask about. Question bank. Keep the prior bookkeeper's name out of the chat.

30. Cleanup Pricing Conversation

Draft a direct email explaining why a cleanup engagement costs more than monthly bookkeeping, without disparaging whoever did the books before. Tone help for an awkward email. Figures stay placeholders.

What prompts can't fix

A few prompts we ran and killed, so you don't have to learn the same way.

"Generate the journal entries for this transaction." It invented account names that didn't match the chart of accounts, twice in one week. Cheap to catch, annoying to keep catching. Use it to explain what an entry should look like conceptually. Never to generate the actual entry.

"Summarize this document" with a real document attached. I uploaded a full tax return once, before I'd thought through what that meant. The summary was fine. The problem: a client's name, numbers, and history sitting on a vendor's servers the moment I hit enter. Killed it the same day. If a prompt only works by uploading the real thing, don't run it.

Anything that ends with a number you'd repeat to a client. Cash position, what they owe, what something costs. ChatGPT produces confident numbers that are wrong often enough to matter. Numbers come from the ledger.

And the structural limits no prompt gets around: ChatGPT can't see your bank feed. It can't match a $3,847.92 deposit back to the 14 invoices inside it. It starts every chat from zero, so the correction you made yesterday is gone today. It's a drafting tool with no memory of your clients and no access to your books. That's not a flaw you can prompt away. It's the design.

Where Growthy picks up

The gap in every one of these ChatGPT prompts for bookkeepers is the books themselves. That's what we build.

Growthy categorizes your clients' transactions automatically and learns from your corrections. Move a transaction, and it remembers, per client. It's 85% accurate on first import, you review the rest, and it asks when it's uncertain instead of guessing. Deposit matching included: the lump-sum processor payouts that no chat window will ever unpick.

ChatGPT drafts your client emails. Growthy does the clicking that ate your afternoon.

Alpha demos are booking now, Growthy is free during alpha, and every demo is run by the founder. Book a demo and bring one messy client book.

Guardrails: what not to put into ChatGPT

Every prompt above works because of one rule that doesn't bend: nothing that identifies a client goes into the chat window. No name next to a number. No account numbers. No uploaded statements, invoices, or exports with real data in them.

Whether the vendor trains on your inputs depends on your account settings and plan, and that policy can change. Verify it yourself before relying on it. But no setting changes your confidentiality obligations to your client. And no plan tier makes a consumer chat tool part of your engagement terms. The rule of thumb: if it has a name and a number next to it, it doesn't go in the box. Everything in this list works by typing a category, a pattern, or an anonymized placeholder. This is operator experience, not legal advice.

FAQ

Can ChatGPT do my bookkeeping?

No. It miscategorizes often enough that you'd end up reviewing everything anyway, and it has no connection to your ledger, so nothing it says posts anywhere. It's genuinely useful for drafting emails, building checklists, and explaining concepts. Use it for words, not books.

Is it safe to paste transactions into ChatGPT?

Descriptions and patterns, yes, if you strip anything identifying: names, account numbers, and real reference numbers. A name next to a number is the line. Uploading statements or exports crosses it, on any plan.

What's the best prompt to start with?

The journal-entry clarity rewrite (#19). It's low risk, takes ten seconds, and fixes a real problem: entries nobody can follow six months later. The close checklist builder (#16) is the best second step.

Should I use ChatGPT or Claude for bookkeeping prompts?

Both handle this list fine; the differences show up in tone and how they handle long checklists. We've compared them head-to-head for this exact work in Claude vs ChatGPT for bookkeeping and, for firm-side work, ChatGPT vs Claude for accounting.

Growthy is bookkeeping software, not a CPA firm. This content is educational, not professional advice.

Related: whether ChatGPT's Finances tab works for client books