Fake receipts have always been part of expense fraud. What's changed is how convincing they've become. AI generated receipt fraud is now a real and growing problem for UK accountants, and the tools producing these documents are widely available, cheap, and getting better fast.
In November 2025, ICAEW published guidance on how to spot an AI-generated receipt, flagging the issue as one that accounting practices need to take seriously. This isn't a niche concern. It's mainstream fraud using mainstream tools, and it lands squarely in your lap.
This post covers what AI generated receipt fraud looks like in practice, the red flags to check for, and how automated detection helps catch what a tired human eye will miss.
Why AI Generated Receipt Fraud Is Harder to Catch
A manually forged receipt usually has tells. Wrong fonts. Misaligned text. Numbers that don't quite add up. A trained eye can often spot the problem within seconds.
AI-generated fakes are different. The tools used to produce them have been trained on thousands of genuine receipts. They replicate layouts, fonts, and formatting with a level of consistency that makes a quick visual check almost useless. The receipt looks real because, in every surface-level respect, it is designed to look real.
This matters because most receipt review processes still rely heavily on a human reviewer glancing at a document and making a judgement. That worked well enough when fakes were crude. It's not enough now.
The ICAEW guidance highlights that AI-generated receipts often pass basic inspection precisely because they mimic genuine documents so closely. The errors that exist tend to be structural or data-level, not visual.
Actionable takeaway: Stop relying on visual checks alone. If your review process amounts to "does this look like a receipt?", it will not catch AI-generated fraud.
Red Flags to Look For in Expense Claims
Knowing what to look for gives you a fighting chance. These are the most common indicators that a receipt may be AI-generated or otherwise fraudulent.
VAT Number Problems
VAT numbers follow a specific format in the UK: nine digits, usually displayed as GB followed by nine digits. AI-generated receipts sometimes include VAT numbers that are structurally wrong, invented, or belong to a different business entirely.
You can verify any VAT number free of charge using HMRC's VAT number checker. It takes thirty seconds. Run it when something feels off.
Totals That Don't Add Up
AI tools can produce receipts where the line items, subtotal, VAT, and total are internally inconsistent. The figures look plausible individually but don't reconcile when you check the maths. This is one of the more reliable tells, because it's genuinely difficult for a generative tool to maintain arithmetic consistency across a full document.
Always cross-check: line items plus VAT should equal the total. If they don't, the receipt warrants closer scrutiny.
Duplicate Submissions
One of the simplest and most common forms of expense fraud is submitting the same receipt twice. This predates AI entirely, but AI-generated fakes make it easier to produce near-identical receipts with small variations in date, amount, or supplier name to avoid an obvious match.
Catching duplicates manually across a large client base is almost impossible. This is where automated duplicate detection earns its keep.
Supplier Inconsistencies
Check whether the supplier name, address, and VAT number are consistent with each other. A receipt from a well-known retailer with a VAT number that doesn't match that retailer's registered details is a clear problem. Equally, watch for suppliers with vague or generic names and no verifiable web presence.
Actionable takeaway: Build a short checklist for your review process. VAT number validity, arithmetic reconciliation, and duplicate checking are three controls that catch a disproportionate share of fraudulent claims.
Why Duplicate Receipt Detection Matters More Than Ever
Duplicate detection is not a new idea, but it's become more pressing as AI makes it easier to produce plausible variants of the same fake receipt.
The challenge with manual duplicate checking is scale. A sole trader submitting ten receipts a month is manageable. A bookkeeping firm handling twenty clients, each with dozens of receipts, is not. No one is systematically cross-referencing every receipt against every previous submission. It's not realistic.
Automated duplicate detection solves this by comparing incoming receipts against the entire submission history, flagging near-matches as well as exact duplicates. This catches both the straightforward double submission and the slightly-altered version of a previously rejected document.
This is where automated receipt extraction earns its keep, running the same checks on every document without fatigue or inconsistency.
The key word is "near-match". A duplicate checker that only catches identical files is not much use. What you need is one that identifies receipts with the same supplier, similar amounts, and close dates, even if the image itself is slightly different.
Actionable takeaway: If your current receipt processing tool has no duplicate detection, that's a gap worth closing. Ask specifically whether it checks for near-matches, not just exact copies.
How Automated Confidence Scoring Catches What Manual Review Misses
Confidence scoring assigns a reliability rating to each receipt based on a range of data checks. A high score means the extraction is clean and the data is consistent. A low score flags the receipt for human review.
What makes this useful in a fraud context is that the checks run at data level, not visual level. The system isn't asking "does this look like a receipt?" It's asking whether the VAT number is valid, whether the arithmetic reconciles, whether the supplier appears in previous submissions, and whether the date and amount combination matches any known pattern of duplication.
These are exactly the checks that manual review either doesn't perform or performs inconsistently. Confidence scoring runs them automatically on every receipt, every time.
Receipts that score below a set threshold get flagged in the review queue. A human reviewer then looks at those specific documents with the context of knowing why they were flagged. This is a much more efficient use of review time than reading every receipt from scratch.
Accuracy starts with a snap. But getting the data right is only part of it. You also need to know when to trust that data and when to question it.
Understanding which fields HMRC requires on a valid receipt is a useful starting point. Our guide to VAT receipt requirements in the UK covers the full rules.
Tools like Receiptflow flag low-confidence extractions automatically, so your review queue surfaces the receipts that genuinely need attention rather than asking you to check everything manually. That's the practical difference between a tool built for accountants and one built for general use.
Actionable takeaway: When evaluating receipt processing software, ask how it handles low-confidence extractions. A tool that simply extracts data and passes it through without flagging uncertainty is not protecting you.
What to Recommend to Clients Right Now
The practical question is what to tell clients. You can't hand them a technical briefing on machine learning. What you can do is tighten the submission process.
Set a submission route. Clients who photograph receipts on the spot and submit them via a controlled channel, whether that's an app, email-in, or a client portal, are far less likely to introduce fraudulent documents than clients who batch-submit at the end of the month from a folder of images. Recency and process discipline are your first line of defence.
Communicate the standard. Let clients know that you verify VAT numbers and check for duplicates. You don't need to be heavy-handed about it. A simple note in your engagement letter or client onboarding process sets the expectation clearly.
Flag anomalies promptly. When you catch something suspicious, raise it quickly. Fraud tends to escalate if unchallenged. An early conversation is far easier than a retrospective audit.
Actionable takeaway: Review your client submission process. If receipts can arrive in any format, from any channel, with no consistency checks at point of submission, you're relying entirely on your own review to catch problems.
Frequently Asked Questions About AI Receipt Fraud in the UK
What is AI generated receipt fraud?
AI generated receipt fraud involves using artificial intelligence tools to create fake expense receipts that look genuine. The receipts typically replicate the formatting, fonts, and layout of real documents from known suppliers. They're designed to pass a basic visual inspection, though they often contain data-level errors such as invalid VAT numbers or arithmetic inconsistencies.
How can UK accountants detect fake receipts?
The most reliable checks are data-level rather than visual. Verify the VAT number using HMRC's free checker, confirm that line items, VAT, and totals reconcile arithmetically, and check whether the same receipt or a similar one has been submitted previously. Automated receipt processing tools with confidence scoring and duplicate detection run these checks systematically on every document.
Can expense fraud detection software catch AI-generated receipts?
Yes, provided the software checks at data level rather than relying on image analysis alone. Confidence scoring that validates VAT numbers, checks arithmetic, and flags near-duplicate submissions will catch a high proportion of AI-generated fakes. Tools that simply extract data without flagging anomalies offer much weaker protection.
What should I do if I suspect a client has submitted a fake receipt?
Raise it directly with the client as a factual query rather than an accusation. Ask them to provide supporting documentation such as a bank statement confirming the transaction. If you cannot verify the expense, do not include it in the accounts. If the pattern repeats or the amounts are material, you may need to consider your professional obligations under anti-money laundering regulations.
Are AI-generated receipts a common problem in UK accounting practices?
ICAEW flagged the issue in November 2025, indicating it has become common enough to warrant published guidance for practitioners. The tools needed to produce convincing fake receipts are widely available and require no specialist technical knowledge, which means the barrier to fraud has dropped. Most practices have not yet updated their review processes to account for this.
How do I check a VAT number on a receipt?
Use HMRC's free VAT number checker at https://www.tax.service.gov.uk/check-vat-number/enter-vat-details. Enter the VAT number as it appears on the receipt. The service confirms whether the number is valid and which business it is registered to. A mismatch between the registered business and the supplier named on the receipt is a clear red flag.
What makes a receipt suspicious enough to investigate further?
Any of the following warrant a closer look: a VAT number that fails HMRC's checker; totals that don't reconcile with line items; a supplier with no verifiable web presence or registered address; a receipt that closely resembles one submitted previously; or an unusual clustering of receipts near the end of a period.
