This guide goes deeper on three areas where mistakes do the most damage when they slip through: numbers, names and sources. It also covers a related risk: scammers who use AI to fool your staff.
Numbers: do the math yourself
Never trust a total just because it is neatly formatted. When AI gives you a number that matters:
- Recompute totals, subtotals and tax with a calculator or your own spreadsheet.
- Check that every figure came from your source, not from the AI. If you gave it five line items and the summary mentions six, something was invented.
- Watch units and decimals: hours or days, monthly or yearly, 1.5 or 15.
- Compare percentages with the numbers behind them. If "half the invoices" turns out to mean three out of ten, something is off.
A good habit: ask the AI to list the numbers it used, then check that list against your records.
Names and details: use your own records
AI can mix up names, swap first and last names, or produce an address that looks believable but is wrong. Before anything goes out:
- Check every name, title, address, phone number and email against your own system, such as your client files or contact list.
- Don't let AI "fill in" missing contact details. If you don't have it, get it from a source you trust.
- Watch for names the AI added that you never gave it.
Sources: make sure they exist
If an AI answer cites a law, a study, an article, a quote or a court case, treat the citation as unconfirmed.
- Find the original source yourself, from an official or trusted site.
- Read the actual text. Make sure it says what the AI claims.
- Be careful with links in AI answers. They can lead to pages that don't exist, or to the wrong site. Instead of clicking, type the address of a site you already trust.
If you can't find the source, don't use the claim.
Can this go into AI? An office manager wants to paste last month's vendor invoice totals, with vendor names but no bank or account numbers, into the company's business AI tool to spot duplicate bills.
Answer: Only if your AI policy allows business figures in that tool — vendor names and totals are business data, not personal data. Then check any duplicates it finds against the original invoices yourself.
Scams that use AI
The same tools that help you write can help criminals write. Scams that were once easy to spot because of bad spelling now arrive polished. Watch for:
- Fake invoices that copy a real vendor's style and ask for payment to a new account.
- Urgent payment changes: an email or text saying a vendor "switched banks" and you must pay today.
- Voice clones: a call that sounds like your boss or a relative, asking for an urgent transfer or gift cards. AI can copy a voice from a short recording.
- Fake messages from "the owner" asking staff to buy gift cards or send sensitive files.
The red flags are about pressure and process, not spelling:
- Urgency: "today," "right now," "before the bank closes"
- A request to change how or where money goes
- A reason not to call or check: "I'm in a meeting, just text"
- Secrecy: "don't tell anyone yet"
The fix is simple and old-fashioned: verify through a channel you already trust. Call the vendor on the number in your records, not the one in the email. Call your boss on their known number. Never change payment details based on an email or call alone, and require a second person to confirm any change to bank details.
Can this go into AI? A bookkeeper gets an email saying a long-time vendor has new bank details, and wants to paste it into a scam-checking tool to look for red flags.
Answer: Yes — once any account numbers are masked. Checking a suspicious message is fine, but the real test is calling the vendor on the number already in your records.
To practice, try our live Scam Check. Paste a suspicious email, invoice or text, or describe a strange call, and it points out the red flags and what to do next.
Key points
- Recompute every total yourself, and make sure each figure came from your own source.
- Check names and contact details against your own records, and never let AI fill in missing ones.
- Find and read every source an AI cites, and don't trust AI-made links.
- AI makes scams more convincing. Verify any payment change through a channel you already trust.
Payment change call-back script
Use this whenever a vendor, client or coworker asks to change where or how money is sent.
- Request received from [sender] on [date] by [email / text / call], asking to change [payment details].
- Call back on the number already in our records: [number from our files, not from the message].
- Script: Hi, this is [your name] from [Business Name]. We received a request to change your payment details. Can you confirm whether you sent it?
- Confirmed by [name of person you spoke with] at [time]. Change is genuine: [yes / no].
- Second person check: [coworker name] reviewed and agreed on [date].
- If anything feels off (urgency, secrecy, a reason not to call), stop and tell [owner or manager] before paying.