Prompt
You are an experienced bookkeeper's assistant doing a first-pass categorization of transactions for review. You propose; the bookkeeper decides.

Business context: {{business_type}}

Chart of accounts to use (the ONLY allowed categories): {{chart_of_accounts}}

Transactions (date, description, amount): {{transactions}}

Return a table with: date, description, amount, proposed account, confidence (High / Medium / Low), and one-line reasoning.

Rules:
- Use ONLY accounts from the chart of accounts provided. Never invent an account.
- If a description is ambiguous or you would be guessing, mark it REVIEW with Low confidence and state the question that would resolve it.
- Flag anything that looks like a duplicate, a personal expense, an owner draw, or a transfer between accounts rather than income or expense.
- Do not infer a vendor's business type from a generic name — mark those REVIEW.
- After the table, list every REVIEW item with the exact question to ask the client.

This is a draft for my review before anything is posted. Accuracy beats coverage: a REVIEW flag is always better than a wrong category.

Fill in your details and the prompt updates live — then copy.

What you get back (excerpt)

| 03/04 | AMZN Mktp US | $84.12 | Supplies | Medium | Amazon purchases for this client are usually grooming supplies — confirm | | 03/07 | SQ *BLUE BOTTLE | $6.75 | REVIEW | Low | Coffee shop — client meeting (Meals) or personal (Owner Draw)? | | 03/09 | TRANSFER TO SAVINGS | $2,000.00 | REVIEW | Low | Inter-account transfer, not an expense — confirm destination account | | 03/12 | CHEVRON | $61.40 | Vehicle Fuel | High | Fuel purchase consistent with mobile grooming route | REVIEW questions: (1) Was the 3/4 Amazon order grooming supplies or something else? (2) Was Blue Bottle a client meeting or personal? (3) Confirm the 3/9 transfer went to the business savings account.

The full workflow

  1. Export descriptions, dates, and amounts only — strip account numbers and any customer names before pasting.
  2. Paste the client's actual chart of accounts, not a generic one.
  3. Run the prompt in batches of 50-100 transactions.
  4. Review every line, resolve REVIEW items with the client, then post in your accounting software.
  5. Save recurring vendor decisions as bank rules so next month needs no AI at all.

Watch out for

FTC Safeguards Rule: a raw bank export is customer financial information. Strip account numbers, customer names, and balances before pasting anything, and document the tool you use in your written security plan (WISP).

The AI has no memory of the client's history — a vendor that was 60% supplies and 40% inventory last year will be coded on name alone. Review every line before posting.

Confidence labels are self-reported, not calibrated. Treat "High" as "plausible," not "verified."

Where this comes from

Every use case on this site is grounded in real reports from working bookkeepers — not invented by us.

More AI use cases for bookkeepers

← All 6 use cases: How Bookkeepers Use AI