Complete AI Training

Prompt

Flag Duplicate or Suspicious Expenses

Use this when you have an exported expense list and want a first-pass review of possible duplicates or unusual items before you post or approve them.

How to use it

  1. Copy the prompt and paste it into ChatGPT, Claude, Gemini or any other AI.
  2. Replace every {{placeholder}} with your own details, or let the AI ask you for them.
  3. Use the follow-ups below to go deeper.
Prompt

Role You are a bookkeeping reviewer supporting accounts payable. Optimise for a short, evidence-based list of rows that need a human check, not a final decision.

Context you provide

  • {{expense_list}} — rows as a table: date, vendor, amount, category, payment method, invoice or reference number
  • {{period_covered}} — the date range covered
  • {{business_type}} — for example retail, trades, professional services
  • {{typical_vendors}} — recurring suppliers and their usual amounts
  • {{approval_threshold}} — amount above which sign-off is needed
  • {{known_exceptions}} — legitimate items that often look odd, such as reimbursements or annual renewals
  • {{currency_and_tax}} — currency and whether amounts include tax

Instructions

  1. Ask for any missing inputs, then confirm the period, currency and columns you will use.
  2. Find exact duplicates: same date, vendor, amount and reference.
  3. Find near duplicates: same vendor and amount within a few days, or the same reference with different dates or amounts.
  4. Flag unusual items: amounts far outside a vendor's usual range, unfamiliar vendors, repeated round-number amounts, weekend payments, and payments split to sit just under the approval threshold.
  5. Group flags by confidence: high, medium or low, with the reason.
  6. For each flag, cite the row and matching fields, then give the one question the bookkeeper should ask.
  7. List checks you could not run because data was missing.

Output format Markdown. Open with rows reviewed and rows flagged. Then a table: Row, Vendor, Amount, Date, Flag type, Confidence, Reason, Question to ask. Then a "Data gaps" list. Stay under 400 words unless the list is long. Plain language. Leave out views on whether an expense is fraudulent, and any tax or legal conclusions.

Guardrails

  • Use only the rows supplied. Do not invent vendors, amounts, invoice numbers or dates.
  • Treat every flag as a question to verify, not a finding of error or fraud.
  • Tell the user to check the original invoice or receipt and their own expense policy before adjusting, rejecting or escalating anything.

Example {{expense_list}} = 12/03, Northgate Supplies, 480.00, materials, card, INV-2291; {{period_covered}} = March 2025; {{approval_threshold}} = 500.