Finance & APStarter

Pull invoice details into a table you can actually import

Extracts the fields you name from a batch of invoices and flags what it could not read instead of guessing.

Works with Claude, ChatGPT, GeminiUpdated 2026-09-27

When to use it

Use this on a batch of invoices you would otherwise key in by hand. The instruction that matters is the one telling it to mark a field unreadable rather than infer it, because a confidently wrong invoice total is far more expensive than a blank cell.

Opening this in Claude or ChatGPT sends the prompt with its {{PLACEHOLDERS}} still in it. Replace them with your own details before you send, or copy the prompt and fill them in first.

Extract the following fields from each invoice below: {{FIELDS}}

<invoices>
{{INVOICES}}
</invoices>

Output a table, one row per invoice, with those fields as columns in the order listed.

Rules:
- Copy values exactly as printed. Do not reformat dates, do not convert currencies, do not strip leading zeros from invoice numbers.
- If a field is missing or you cannot read it with confidence, write NOT_FOUND. Never infer a value from the others. An inferred total is worse than a blank one, because nobody checks it.
- If the same invoice appears twice, keep one row and add a DUPLICATE column.
- After the table, list any invoice where a total does not equal the sum of its lines, with both numbers. Do not correct it; flag it.

Then state how many invoices you processed and how many cells came out NOT_FOUND, so the person checking knows how much to check.

End of prompt: Pull invoice details into a table you can actually import

Related prompts

Browse all 50 prompts

Get new prompts as we publish them

New prompts as they go up, plus founder essays on shipping AI systems. ~2 emails / month.

One-click unsubscribe. We never share your email.

Prompts get you started. If you want this wired into how your team actually works, see what we do.