← All use cases

中文

Finance & bookkeeping · Claude

Have Claude double-check your tax numbers against receipts

Idea + starter prompt, not run by us yet

For: Self-employed people and couples with a side business who file their own taxes

The poster shared only figures and dates, plus a handful of public business names, from a tangled household tax picture. Claude reviewed the first pass, found double-counted income and costs and client mix-ups, and produced a spreadsheet. The poster typed the numbers into filing software by hand and checked receipts and bank statements too.

Steps

  1. Gather your receipts, bank statements and your own first-pass numbers.
  2. Strip out ID numbers, full names and addresses before you paste anything. Give only amounts, dates and category labels.
  3. Ask Claude to look for double-counting, wrong categories and missing items, and to explain each issue.
  4. Re-check its calculations with a second pass, then compare the final figures to your receipts and statements yourself.
  5. Enter the numbers into your filing software by hand and decide yourself whether you need a professional.

Starter prompt

Starter prompt. Change the [brackets] first.

Trial first: use [5] made-up lines of income and expenses, with one double-counted item and one wrongly filed item, and run the whole check on those. Save the result in [trial-check]. Do not use my real figures yet.

This job only reads and drafts. It changes nothing outside [output folder] and files nothing. After my yes, here are my tax figures for [year]: income streams, expenses and dates, with names and ID numbers removed. [paste figures]

Check them for double-counted revenue or expenses, items under the wrong category and anything that does not add up. Explain each problem in plain words and ask me when unsure. Then give a summary table I can compare with my receipts and bank statements. Do not guess missing numbers. I will not give you ID numbers, passwords or account numbers.

Original post: Reddit · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.