How to Spot a Fake Receipt (And Why It's Getting Harder)
Tanvir Alam•Sep 14, 2026•8 min read•Receipt Management
AI-generated fake receipts now make up the majority of flagged expense fraud, and a manual review on its own can no longer reliably catch them.
Knowing how to spot a fake receipt used to be straightforward. Blurry text, odd fonts, a VAT number that did not add up. A trained eye could catch most fakes in seconds. That is no longer the case.
AppZen data has been reported as showing AI-generated receipts accounted for 0% of flagged expense fraud on its platform in March 2025, climbing to 70.8% by May 2026 (Accounting Today, 2026). On that reporting, the crossover happened in April 2026, when AI image tools overtook template-based forgeries as the dominant method of receipt fraud. For practices processing expense claims across a client base, that shift matters.
The reliable signals have not disappeared. You just need to know where to look, because some of them are far less obvious than they used to be.
Why fake receipts are getting harder to detect
Until recently, forging a convincing receipt needed either Photoshop skills or a paid template service. Both left traces. Photoshop edits were often visible under scrutiny. Template sites produced recognisable patterns. Expense review tools got good at catching both.
Then generative AI changed the economics of fraud. Tools like ChatGPT and Google Gemini can produce a photorealistic receipt from a text prompt in seconds, at no cost. A Medius survey of 2,000 UK and US finance professionals has been reported as finding 30% saw an increase in falsified receipts after the launch of OpenAI's GPT-4o in 2024. A SAP research report from July 2025 has been reported as finding nearly 70% of chief financial officers believed their employees were using AI to falsify travel expenses or receipts.
The shift has also changed the shape of the fraud. AI-generated fakes have been reported as tending to be lower value, averaging around $100 per receipt against a $182 average for older template-based fakes (AppZen, 2026; figures are US platform data). That is deliberate. Smaller claims slip under manual review thresholds more easily. High volume, low value, and a low risk of detection, at least without the right checks in place.
The fake receipt checklist: what to look for
This checklist works for manual review and for briefing your team on what to flag before escalating. Work through it in order rather than relying on instinct.
1. Check the VAT number
Every UK receipt from a VAT-registered business should carry a valid VAT registration number in the format GB followed by nine digits, for example GB 123 4567 89. You can verify any number in seconds using HMRC's VAT number checker.
Red flags:
VAT number missing on a receipt from a business that should be VAT-registered
Number formatted incorrectly, with the wrong digit count or no GB prefix
Number that does not appear in HMRC's register
The same VAT number appearing across receipts from supposedly different businesses
This is one of the most reliable checks you can run. AI tools can generate plausible-looking VAT numbers, but they cannot make a fake number pass HMRC's live register.
2. Check the business details against Companies House
A receipt should carry the legal trading name, the address, and where relevant the company registration number of the supplier. Cross-reference the details against the free Companies House register.
Red flags:
Business name that does not match a registered company or sole trader
Address that returns nothing on a basic map search
Phone number that goes to a disconnected line or voicemail
Website on the receipt that is inactive or newly registered
3. Examine the maths
This sounds obvious, but it catches a surprising number of fakes. AI-generated receipts sometimes produce totals that do not match the itemised lines. Check that net plus VAT equals gross, that line items sum correctly, and that the VAT rate applied produces the right amount.
Red flags:
VAT amount that does not match 20% of the net, or whichever rate is shown
Line item subtotals that do not add up to the pre-tax total
Round numbers that look implausible for the type of purchase
If the receipt is for a taxi, a restaurant, or retail, all categories with well-understood pricing, a total that does not fit the description of what was bought is worth querying.
4. Look at the format and layout
Genuine receipts follow patterns specific to their point-of-sale system. Retailers on the same till software produce receipts with consistent fonts, layouts, and spacing. AI generators do not always replicate these.
Red flags:
Font inconsistencies within the same receipt, especially between item names and totals
Unusual spacing or alignment that differs from other receipts from the same supplier
Logos that look slightly off, meaning blurred edges, wrong proportions, or a colour that does not match the brand
Thermal receipt formatting, the narrow single-column style, applied to categories that do not typically produce thermal receipts
A receipt image that is suspiciously sharp and clean for something supposedly photographed in the field
That last point matters more now. A real receipt photographed on a phone tends to have shadows, slight angles, and background context. An AI-generated image rendered at perfect resolution on a plain white background is worth a second look.
5. Check the date, time, and location for plausibility
This is a behavioural check rather than a document check, and it is often where expense fraud gives itself away. Does the receipt date match a working day when the person was actually in that location? Does the timing make sense alongside other receipts or calendar entries?
Red flags:
Receipt dated on a weekend or public holiday for a business that would not typically be open
Two receipts from different cities on the same day with no travel claim between them
Receipt from a supplier in a location the person had no reason to visit
Times that conflict with other verified activity, such as flights, hotel check-ins, or meeting records
This cross-referencing is exactly where manual review struggles at scale. Reviewing one receipt takes seconds. Reviewing it against three months of expense history and calendar data takes far longer.
6. Check for duplicate claims
Duplicate submission is one of the oldest forms of expense fraud and still one of the most common. The same receipt submitted twice, once as a photo and once as a PDF. A receipt resubmitted in a later period. Two receipts from the same supplier on the same date for the same amount.
Red flags:
Identical amounts from the same supplier within a short window
The same receipt image submitted in different file formats
Claims with the same date and description but slightly different totals
Manual duplicate checking across a large client base is slow. It is also one of the areas where automated tools add the most reliable value.
7. Watch for behavioural patterns
Individual receipts are one thing. Patterns across a client's expense data are often more revealing. Fraud tends to be consistent, which creates statistical signals that stand out when you are looking for them.
Red flags at the pattern level:
Claims that cluster just below the client's approval threshold
A sudden increase in the volume or value of claims from one person
The same supplier appearing repeatedly with no obvious business reason
Claims that spike at month-end or quarter-end when attention is typically lower
> Receiptflow flags suspicious receipts during processing, including duplicate detection and format anomaly alerts. Try it free and see what your current review process might be missing.
Why manual review alone is no longer enough
Manual review worked when the scale of fraud was manageable and the techniques were predictable. Neither condition holds in 2026.
AppZen data has been reported as showing more than 3.5 million fake receipts created across the top four expense fraud websites in a single six-month period. The technical barrier is gone. Anyone with a free account and a text prompt can produce a plausible receipt in under a minute. The volume, quality, and variety of fakes have all risen at once.
The ICAEW has been reported as noting that detection tools are increasingly used to catch AI-generated fakes, scanning receipts for patterns in how an image was created and for inconsistencies in lighting, texture, or metadata that no human reviewer would notice (ICAEW, 2025). That is where the field is heading: AI-generated fraud caught by automated detection, with human review focused on the exceptions that automated systems escalate.
For accountants and bookkeepers, the practical point is simple. Manual spot-checks still matter. The checklist above will catch a share of fakes. But the volume and quality of AI-generated receipts mean a practice relying only on manual review is working with a real blind spot.
The firms that handle this well combine a strong manual checklist with automated tools that process every receipt at scale. Neither replaces the other. Together they close most of the gap.
What to do when you suspect a fake receipt
If a receipt fails two or more of the checks above, treat that as grounds for a formal query before processing. You do not need certainty to pause and ask questions.
Query the receipt with the client or employee, asking for supporting evidence such as the original email confirmation, a card statement showing the matching transaction, or a booking reference.
Run the VAT number through HMRC's live checker.
Search the supplier address on Companies House and a map service.
If multiple checks fail and the client cannot provide corroborating evidence, document your findings and escalate under your firm's anti-money-laundering procedures.
Under the UK's Proceeds of Crime Act 2002 and the Money Laundering Regulations, accountants have reporting obligations when they suspect financial crime. A fake receipt submitted for tax purposes is not a minor administrative error. It is a potential fraud. Knowing when to escalate, and to whom, is part of the professional responsibility that comes with the role.
> Processing receipts across multiple clients? Receiptflow flags anomalies automatically, so your team can spend review time where it actually matters. Start your free trial today.
FAQs
Common Questions with Clear Answers
Can you tell if a receipt was generated by AI?
Sometimes, but not always by eye. AI-generated receipts increasingly pass visual inspection. The most reliable checks are cross-referencing the VAT number against HMRC's register, verifying the maths, and checking supplier details against Companies House, because AI tools can produce plausible documents but cannot make a fake VAT number pass a live HMRC lookup.
What are the most common signs of a fake receipt?
The most common red flags are invalid or missing VAT numbers, arithmetic errors in the totals, business details that do not match any registered company, implausible dates or locations relative to the claimant's movements, and receipts submitted just below approval thresholds.
Is submitting a fake receipt illegal in the UK?
Yes. Submitting a fraudulent receipt for a tax deduction or expense reimbursement can constitute fraud under the Fraud Act 2006 and potentially money laundering under the Proceeds of Crime Act 2002. UK accountants also have reporting obligations under the Money Laundering Regulations if they suspect a client has submitted false documentation.
How do I check if a VAT number on a receipt is real?
Use HMRC's free VAT number checker at vat.gov.uk/check-vat-number/lookup. Enter the number shown on the receipt and the tool will confirm whether it is registered and, in many cases, show the trading name associated with it.
What should an accountant do if they suspect a receipt is fake?
Pause processing and ask the client for corroborating evidence such as a card statement showing the matching transaction. Verify the VAT number and supplier details independently. If the receipt fails multiple checks and no supporting evidence is provided, document your findings and consider whether a Suspicious Activity Report is required under UK anti-money-laundering regulations.
