Bookkeeping has two halves: knowing what happened and recording it correctly. AI helps with the paperwork around the second half and with every conversation attached to it: the uncategorized-transaction questions you send clients, the collections emails, the cleanup proposal, the explanation of why cash fell while profit rose. A chat assistant drafts each in a minute and you correct it from what you know about the client.
It does not know the client's chart of accounts, what the owner meant by that $840 Amazon charge, or the difference between an owner draw and a reimbursement unless you tell it. It also miscounts. Anything numeric that matters, from a bank rec to a 1099 total, belongs in your accounting software or a tool that runs actual code, and you check it.
Bank feeds, payroll files, and vendor W-9s contain account numbers and Social Security numbers. Those stay out of consumer chat tools unless your firm has approved the tool and its data terms. Nearly everything below works on de-identified or summarized data.
Quick wins this week
- Paste your uncategorized transactions (description and amount only, no client name or account numbers) and ask for a suggested category plus a one-line client question on each one you are unsure about. Send the questions as one email instead of twelve.
- Turn your best collections email into three versions: friendly at 15 days past due, firm at 30, final notice at 60. Save them as templates in your invoicing tool.
- Describe a messy set of books in a few sentences and ask for a cleanup plan with phases, what you need from the client, and an outline for a fixed-fee proposal.
- Ask for a plain-English paragraph explaining why the profit and loss shows a profit while the bank balance fell. Reuse it every month with new numbers.
What AI can do for bookkeepers, task by task
Categorizing transactions and writing bank rules
Export uncategorized transactions from QuickBooks Online or Xero with only description, date, and amount, remove the client name, and paste them in with the chart of accounts and any client-specific rules (Shell is fuel, not meals). Ask for a category, a confidence level, and a client question for anything below high confidence, then write the recurring ones as bank rules in your software. Check every category against tax treatment before posting; the model guesses from the vendor name, which fails for mixed-use vendors like Amazon and Costco.
Bank and credit card reconciliations
When a rec will not balance, export the statement lines and the register for that account (no account numbers), upload both to a tool that runs code, and ask it to match on amount and date within a few days and list what is unmatched. Confirm the unmatched items sum to your difference. Plain chat without code execution miscounts anything longer than a screen.
Accounts receivable follow-up
Paste the aging by invoice with customers renamed A, B, C and ask for a prioritized call list and a draft message for each aging bucket in your client's voice. Include the client's actual terms and late-fee policy, because the model will invent them if you do not. Read every draft before it goes into the invoicing tool's reminder schedule.
Cleanup and catch-up projects
Describe the state of the books: months behind, unreconciled accounts, a growing undeposited funds balance, loan payments booked as expense. Ask for a phased plan, a document request list, and the questions to ask before quoting. Use the plan as your scope document, then check it against the file; the actual mess is always different from the described mess.
Month-end close checklist and client reports
Give the entity type, the apps in the stack (payroll provider, payment processor, inventory), and your current steps, and ask for a close checklist ordered by dependency with a review step after each reconciliation. For monthly reports, paste the summary numbers without names and ask for three to five sentences of commentary, then verify each against the ledger; any driver the model offers is a guess until you confirm it.
Payroll, sales tax, and 1099 questions
Ask for the framework: which forms, which agencies, what triggers a filing. Then verify every date and threshold on the agency's site or in your payroll provider's documentation, because these change yearly and the model's training data is stale. Use NotebookLM with the actual publication uploaded when you need page-level answers.
Prompts for bookkeepers
Replace the bracketed placeholders, paste into any chat assistant, and iterate on the result.
Categorize a batch of transactions with client questions
You are a bookkeeper categorizing transactions for a [ENTITY TYPE, E.G. S-CORP] in [INDUSTRY] that uses [ACCOUNTING SOFTWARE]. Chart of accounts: [PASTE ACCOUNT LIST]. Client rules: [PASTE KNOWN RULES, E.G. SHELL IS FUEL, ADOBE IS SOFTWARE]. For each transaction below, return a table: Description, Amount, Suggested Account, Confidence (high, medium, low), Client Question. Write a client question only for medium and low confidence, in one plain sentence the owner can answer from memory. Do not invent a vendor's purpose; if the description is ambiguous, mark it low. Do not comment on tax deductibility. Transactions (description, date, amount): [PASTE TRANSACTIONS WITH NO ACCOUNT NUMBERS OR CLIENT NAME]
Tip: Post the high-confidence rows after a skim; the questions column becomes one email to the client.
Write a three-stage collections sequence
You are writing accounts receivable follow-up emails on behalf of a [CLIENT BUSINESS TYPE] to its customers. Payment terms: [TERMS, E.G. NET 30]. Late fee policy: [POLICY OR NONE]. Tone: [FRIENDLY AND PROFESSIONAL]. Write three emails: a reminder at [DAYS PAST DUE], a firmer notice at [DAYS PAST DUE], and a final notice at [DAYS PAST DUE] that states the next step ([NEXT STEP, E.G. PAUSE SERVICE OR REFER TO COLLECTIONS]). Each under 120 words, with placeholders for invoice number, amount, and due date. Do not threaten legal action or mention interest or fees beyond the policy above.
Tip: Load them into your invoicing tool's reminder feature so they go out on schedule without you.
Scope a cleanup project
You are a senior bookkeeper scoping a cleanup engagement. The books are in [SOFTWARE], last reconciled in [MONTH AND YEAR], for a [ENTITY TYPE] doing roughly [ANNUAL REVENUE RANGE]. Known problems: [LIST, E.G. UNDEPOSITED FUNDS BALANCE, LOAN PAYMENTS BOOKED AS EXPENSE, DUPLICATE VENDORS]. Produce: 1) a phased plan in order of dependency, 2) the documents to request from the client before starting, 3) ten questions to ask on the scoping call, 4) risks that would change the price. If anything above is too vague to plan against, ask me clarifying questions first. Do not estimate hours or fees.
Tip: Answer its clarifying questions honestly; that is where the scope creep hides.
Explain a financial report to an owner
You are a bookkeeper writing to a small-business owner with no accounting background. Explain [THE QUESTION, E.G. WHY PROFIT IS UP BUT CASH IS DOWN] using these figures: [PASTE ROUNDED NUMBERS, NO NAMES]. Under 150 words, one concrete example, and a plain definition for any term a non-accountant would not know. Do not give tax advice or tell them what to do; end by offering a ten-minute call with [ME OR THEIR CPA].
Tip: Round the numbers before pasting; the model treats precise figures as facts to reuse.
Build a month-end close checklist
You are a bookkeeping operations lead. Build a month-end close checklist for a [ENTITY TYPE] client in [INDUSTRY] using [ACCOUNTING SOFTWARE], [PAYROLL PROVIDER], and [PAYMENT PROCESSOR OR OTHER APPS]. My current steps: [PASTE YOUR STEPS]. Order the checklist by dependency, add a review checkpoint after every reconciliation, flag which steps could be handled by bank rules or recurring transactions, and list what the client must provide by which day of the month. Format as a table with Step, Owner, Due Day, Done By. Do not include tax filings.
Tip: Paste the finished table into your task manager as a recurring project.
Want a prompt for something else? Use the Prompt Builder or browse finance and analysis prompts.
Skills to build
De-identifying data before you paste
Why: Client names, account numbers, and SSNs are the difference between a permitted use and a confidentiality breach, and the model does not need them to categorize or reconcile.
How: Build an export habit: description, date, amount only. Replace customer and vendor names with letters.
Writing rules the way you would brief a new hire
Why: The model's categorization is only as good as the client-specific rules you give it, which are exactly what a new team member would need too.
How: Keep a one-page rules file per client (vendor to account mappings, owner quirks) and paste it at the top of every categorization prompt.
Knowing when to use code execution
Why: A chat model predicting text approximates sums; a tool that writes and runs code computes them. Bank recs, aging totals, and 1099 thresholds need the second kind.
How: Find the data-analysis mode in ChatGPT, Claude, or Julius, or Copilot in Excel, and practice on an account you already reconciled by hand.
Verifying dates and thresholds against the source
Why: Sales tax, payroll, and 1099 rules change yearly and vary by state, and the model answers confidently from stale training data.
How: Fixed rule: any date, rate, or dollar threshold gets checked on the agency's site or in your payroll provider's documentation before it reaches a client.
Saving prompts as templates
Why: You run the same dozen workflows for every client every month. A saved prompt with placeholders makes each one consistent and fast.
How: Put your five best prompts, your writing style, and a verification checklist into a Claude Project, custom GPT, or Gemini Gem.
Tools worth knowing
ChatGPT
The general-purpose AI assistant most of your coworkers already use.
Claude
A careful writing and analysis assistant that shines on long documents.
Microsoft Copilot
AI inside Word, Excel, Outlook, and Teams, with your company's data protections.
Julius AI
Chat with your spreadsheets and data files and get charts, stats, and answers back
NotebookLM
A research notebook that only answers from the sources you give it, with citations.
Zapier
Connect thousands of apps with no-code workflows, now with AI steps, agents, and chatbots
Cautions for bookkeepers
If you prepare or help prepare tax returns, tax return information is protected under IRC Section 7216 and your practice falls under the FTC Safeguards Rule, which requires a written information security plan that should name the AI tools you allow. Even if you only keep books, bank feeds, payroll files, and W-9s carry account numbers and Social Security numbers. Never paste client financial data into a consumer AI tool unless your firm has approved the tool and its data terms, and de-identify anyway.
Payroll registers carry names, SSNs, pay rates, and direct deposit account numbers. Keep them inside your payroll provider. If you need help with a payroll question, describe the situation in words (an hourly employee in one state, a bonus run) instead of uploading the register.
The model will total a column wrong, drop a row, and file a vendor under the wrong account because the name sounded like something else. Every total goes through your software or a code-execution tool, and every category you did not personally set gets a look before it posts. The tax treatment of a transaction is a judgment, not a vendor lookup.
Browser-driving agents that can log in and click are tempting for bank downloads and bill pay. Bank terms of service, your engagement letter, and your liability coverage rarely contemplate that. Use the bank feeds and integrations your accounting software already supports, and keep credentials in a password manager the tool cannot read.
Your 30-day plan
- Week 1: Read your firm's AI policy or write a one-page version that names approved tools and banned data (client names, SSNs, account numbers). Confirm the tool's data terms exclude training on your inputs.
- Week 1: Do three quick wins on non-confidential work: a collections sequence, a report explainer, and a close checklist for one client.
- Week 2: Run one client's de-identified uncategorized transactions through the categorization prompt. Write down where the model was wrong; those notes become the rules file.
- Week 2: Reconcile one account with a code-execution tool alongside your normal process. Note which exceptions each caught and how long each took.
- Week 3: Save your five best prompts and one rules file per client into a Project, custom GPT, or Gem with a verification checklist.
- Week 4: Automate one recurring handoff, for example a Zapier flow that files receipts from a shared inbox into your document folder, and show your team what you built, including the errors you caught.
Frequently asked questions
Will AI replace bookkeepers?
Can ChatGPT do my bookkeeping?
Is it safe to put client bank data into an AI tool?
Which AI tool is best for bookkeepers?
Should I tell clients I use AI?
Terms used on this page
Related roles
- AI for AccountantsAI drafts the flux commentary, writes the XLOOKUP, explains the K-1 to your client in plain English, and reads the 40-page lease for you. You still own the numbers, the judgment, and the signature.
- AI for Tax PreparersAI drafts the organizer letter, the CP2000 explanation, and the note about why the refund shrank, and it reads a 900-page bill so you can ask it questions. Every threshold, date, and citation still gets checked against this year's source before it touches a return.
- AI for Small Business OwnersAI answers the one-star review calmly, drafts the month of posts, turns your voice memo into an SOP, and preps the questions for your CPA. You still make the calls, sign the checks, and own what goes out under your name.
- AI for Financial AnalystsAI writes the first pass of your variance commentary, audits your model for hard-codes and sign errors, and turns a 10-K into a table with page references. You decide what the numbers mean and what to tell the CFO.