Fourteen months ago, fake receipts flagged by expense-fraud detection systems were almost entirely low-tech: photocopied templates, receipts bought from shady websites, or a digit clumsily edited in image-editing software. By mid-May 2026, that had flipped. AI-generated receipts — synthetic images created in seconds with a text prompt — went from 0% of detected fake receipts in March 2025 to 70.8% just fourteen months later, according to expense-management platform AppZen. Over a single year, the company caught 1,471 AI-generated receipts submitted by 745 employees across 174 companies, totaling $148,143 in claimed reimbursements.
If you run a small business and think "we're too small for anyone to bother," the data says otherwise. AI receipt fraud isn't concentrated in a handful of dramatic five-figure schemes — it's spread thin, on purpose, across thousands of small claims that are individually easy to approve and collectively expensive to ignore.
Why This Wave of Fraud Looks Different
Older receipt fraud had a built-in speed bump: making a convincing fake took effort, so most people who tried it either went big (worth the risk) or didn't bother. Generative AI removed the speed bump entirely.
"AI generators are free, instant, and good enough to fool a person," said Kunal Verma, AppZen's CTO, in comments on the trend. "AI basically flipped the game from one fake big enough to be worth the risk to a pile of tiny ones."
That shift shows up clearly in the numbers. AI-generated fake receipts average around $101 each (median $32) — noticeably lower than the $182 average for old-school template fakes. That's not a coincidence. Most expense platforms auto-approve anything under a set dollar threshold, often somewhere between $25 and $100. Keep every fake receipt just under that line, and it sails through without a human ever looking at it.
A few other findings worth knowing if you handle reimbursements:
- Employee admission rates are high. In one 2026 survey, roughly 40% of U.S. employees said they'd used AI to generate or alter a receipt for a work expense — 19% fabricating a purchase entirely, 15% inflating the amount of a real one.
- Company tools are part of the problem. 40% of employees who created fake receipts used AI tools their own employer was paying for, meaning the fraud is sometimes running on your own software license.
- Repeat behavior is common. Roughly a third of employees caught submitting a fake receipt did it again later — at one large employer studied, the repeat rate reached 41%.
- It's a global pattern, not a one-country problem. In the same dataset, submissions spanned 174 companies and dozens of countries, from a handful of $30 lunch receipts to a single $12,900 claim from one employee.
None of this means your team is unusually dishonest. It means the tools for cutting corners got dramatically better and cheaper at the same time your review process probably didn't change at all.
Why "Does It Look Real?" No Longer Works
For years, the standard for reviewing a receipt was essentially visual: does the paper texture look right, do the fonts match the vendor, does the total add up. Modern image generators produce receipts with realistic paper texture, plausible line-item breakdowns, and timestamps that match the claimed date — because that's exactly the kind of image these models are good at producing. Asking "does this look like a real receipt?" is no longer a meaningful fraud check, because a convincing fake is now free and instant to make.
The practical implication for a small business is straightforward: stop verifying the document and start verifying the transaction. A receipt is just a claim. The transaction record — the card charge, the bank debit, the merchant name that actually processed the payment — is the fact.
A Small Business Playbook for Reimbursement Reviews
You don't need enterprise fraud-detection software to close most of this gap. A handful of policy and process changes cover the majority of the risk.
1. Match receipts to actual transaction data, not just document appearance
If an employee pays with a company card, the bank or card processor already has a record of the merchant, amount, and date. Cross-check the receipt against that record rather than trusting the image on its own. A mismatch in merchant name, amount, or timing is a far stronger signal than anything visible on the receipt itself.
For expenses paid out of pocket and submitted for reimbursement, ask for the underlying payment confirmation (a card statement line, a digital wallet receipt, a bank transfer confirmation) alongside the vendor receipt — two independent records are much harder to fabricate consistently than one.
2. Revisit your auto-approval threshold
If your expense tool auto-approves anything under, say, $75, know that fraudulent AI receipts are being sized specifically to land under exactly that kind of line. That doesn't mean you need to review every $12 coffee receipt personally — but it does mean a fixed dollar threshold alone is a known, exploitable rule. Combine a lower auto-approval ceiling with random sample audits above and below the line, so employees can't reliably predict which small claims get a second look.
3. Watch for repeat submitters and pattern clustering
Because a large share of fraud is repeat behavior, tracking who has previously had a receipt flagged — even for something minor — is one of the higher-leverage things you can do with limited review time. Similarly, watch for clusters: multiple employees submitting suspiciously similar receipt formats, amounts that all land just under your threshold, or a spike in "lost receipt, here's a reconstruction" submissions.
4. Require prompt submission with full detail
Set a clear policy — expenses submitted within a set window (commonly 30 days) with the date, amount, merchant, and business purpose stated explicitly. Late submissions, vague purposes ("client meeting," no name or context), and receipts with digits that don't quite match a plausible itemized total are all worth a closer look.
5. Use pre-approval for anything unusual or high-value
Ordinary recurring expenses (mileage, standard travel, routine supplies) can stay on a light-touch process. Anything outside the normal pattern — a large one-off purchase, an unfamiliar vendor, a category the employee doesn't usually expense — should require pre-approval or a manager sign-off before reimbursement, not just after-the-fact review.
6. Don't rely on AI tools you provide employees without also verifying their output
If your business gives employees access to AI writing, image, or document tools, be aware that the same tool that helps someone draft a client email can also generate a convincing fake receipt in seconds. Policy language alone ("don't misuse company tools") is not a technical control — pair it with the transaction-matching habit above.
What This Looks Like in Practice
The scale problem is easiest to see in a real example. At one Fortune 10 company studied in the AppZen data, 142 employees across 22 countries submitted 340 fraudulent receipts worth a combined $34,953 — an average of roughly $103 per claim. No single submission would have triggered a manual review on its own. It took looking at the pattern across the whole organization, not any individual receipt, to see the problem.
A small business obviously isn't processing thousands of expense reports a month. But the same math applies at a smaller scale: five employees each padding a handful of $40–$80 claims a year is easy to miss one report at a time and adds up to real money by December. The geographic spread in the data is also a reminder that this isn't a niche behavior tied to one team or role — it shows up wherever people submit expenses and know the review is light.
A Quick-Reference Checklist for Reviewing Receipts
When something crosses your desk for reimbursement, a few seconds on each of these catches most fabricated claims:
- Does the amount, date, and merchant on the receipt match the actual card or bank transaction? If there's no matching transaction at all, that's disqualifying on its own.
- Is the total suspiciously close to your auto-approval threshold — say, $3 under a $75 cutoff?
- Has this employee had a flagged or corrected receipt before? Repeat behavior is common enough to weight into your review.
- Does the business purpose field say something specific (client name, project, meeting context) or just a generic phrase like "client meeting" or "office supplies"?
- Was it submitted promptly, or does it show up weeks after the fact as part of a batch of "reconstructed" receipts?
- If several employees submit similar-looking receipts around the same time, treat that as a pattern to investigate, not a coincidence to wave through.
None of these checks require special software — they just require actually looking at the transaction record instead of trusting the document.
The Bigger Point: Expense Fraud Is a Bookkeeping Problem, Not Just an HR One
Expense reimbursement fraud already showed up in roughly 13% of occupational fraud cases before generative AI made fabrication effortless, according to the Association of Certified Fraud Examiners. What's changed isn't whether people are tempted to pad an expense report — it's how cheap and convincing the fabrication has become, and how much smaller each individual fraudulent claim can be while still adding up.
That's exactly why clean, well-documented, easily auditable financial records matter more than ever. When every transaction is tracked in a system where the entry, the account it hits, and the supporting record are all linked and reviewable — rather than scattered across email attachments, spreadsheet rows, and photos of receipts nobody cross-checks — mismatches stand out instead of hiding in volume. A reimbursement claim that doesn't tie back to an actual card transaction or bank line is a red flag you can catch in minutes if your books are structured for it, and one you'll never notice if they aren't.
Keep Your Finances Organized and Auditable
Catching fabricated receipts is ultimately a records problem: you need every expense traceable back to a real transaction, not just a document that looks plausible. Beancount.io gives you plain-text accounting with full version history, so every entry — and every edit to it — is transparent, traceable, and easy to reconcile against your actual bank and card records. Explore the documentation to see how a version-controlled ledger makes mismatched or duplicated expense claims easy to spot, or check out Fava for a visual dashboard on top of your books. Get started for free and build reimbursement reviews on records you can actually trust.