How AI Is Making Fake Rental Documents Harder to Spot — and What to Do About It

Three years ago, a fake pay stub was usually easy to spot. The fonts were slightly off. The employer logo was pixelated. The net pay didn't match the deductions. A moderately careful landlord could catch it with a few minutes of attention.

That is no longer reliably true. Generative AI tools — the same technology behind ChatGPT and image generators — can now produce a structurally convincing pay stub, employment letter, or bank statement in minutes, complete with accurate-looking logos, plausible tax deductions, and consistent formatting that would pass a visual review. According to data from SingleKey, approximately 15% of Canadian tenant applications now contain at least one falsified document — and that figure is climbing.

This does not mean tenant screening is hopeless. It means the methods that worked five years ago are no longer sufficient on their own, and the checks that still work are different ones than most landlords currently run.

What AI can now replicate convincingly

The core shift is that document fraud has been democratized. What previously required graphic design skill and hours of effort now takes a free online tool and a few minutes. Here is what AI-assisted document generation can produce today:

• Pay stubs with accurate Canadian deductions. AI tools can calculate plausible CPP, EI, and income tax figures for a given salary, producing a stub where the gross-to-net math roughly holds. The employer name, address, and Business Number field can all be populated with invented but plausible data.
• Employment letters on convincing letterhead. Generative tools can produce letters that match the tone and structure of genuine HR correspondence, with accurate-looking company logos reproduced from a public website and contact details that appear legitimate at a glance.
• Bank statements with fabricated transaction histories. AI can generate realistic-looking deposit patterns — regular payroll deposits on expected dates, normal recurring expenses — while concealing the withdrawal patterns and financial obligations that would tell the real story.
• Synthetic reference profiles. Beyond documents, coordinated fraud increasingly includes references that are friends or family coached to sound like previous landlords or employers. The document looks real; the phone call sounds real. Neither represents an actual tenancy.

Why the traditional checks are failing

The standard landlord review was built for an older generation of forgeries. It looked for visual anomalies — inconsistent fonts, misaligned columns, logo quality — and arithmetic errors in the deduction math. AI-generated documents are designed to pass exactly these tests.

More specifically: legacy forgeries were built document by document, which meant errors were common and obvious. AI-generated documents are structurally sound. The fonts are consistent because they're generated from a template. The layout is clean because it's machine-produced. The deductions are plausible because the tool calculated them. A quick visual scan — which is what most landlords have time for — finds nothing wrong.

There is also a pattern problem. When a landlord reviews a single application, each document is evaluated in isolation. Coordinated fraud often involves a set of documents that are individually plausible but inconsistent with each other in subtle ways that only surface when you cross-reference them — which most reviewers don't do systematically.

What still works: the signals AI consistently gets wrong

AI is good at surface plausibility. It is less reliable at producing documents that are consistent with each other across every detail, and at replicating the specific characteristics of real payroll software output. The checks that still catch altered documents are the ones that go deeper than visual review:

• Year-to-date consistency across multiple stubs.Authentic pay stubs generated by the same payroll system accumulate YTD totals in a mathematically consistent pattern. Across three stubs, the YTD gross should increase by exactly the period gross each time, for salaried employees. AI-generated stubs frequently get this wrong — the YTD figures are plausible individually but don't add up sequentially. Our pay stub verification guide walks through the exact math.
• Employer verification through official channels only.Look up the employer in the Ontario Business Registry (ontario.ca/businessregistry) independently — not using contact details the applicant provided. Call the company's main line, found through a public directory search, and ask to be transferred to HR or payroll. An AI-generated letter can include a convincing phone number. A real company switchboard cannot be faked. Our five-step employment letter guide covers each verification point.
• Bank statements as pattern checks, not balance snapshots.A bank statement that shows payroll deposits but no rent payments, no utility payments, and no recurring expenses is describing a financial life that doesn't exist. Review the full statement for what is absent, not just what is present.
• Cross-document consistency on the details that don't matter.Fraudulent document sets are usually assembled to make the important numbers look right. The details that seem less important — the exact spelling of the employer name, the address format, whether the position title on the pay stub matches the employment letter exactly — are where inconsistencies appear. Check every field against every other document, not just the income figures.
• PDF metadata.Every PDF records when it was created, when it was last modified, and what software produced it. A pay stub claiming to come from ADP with a PDF producer field showing “Adobe Acrobat” or “Canva” has been re-exported through a document editor — which is not what a legitimate payroll system does. This check takes 30 seconds and catches a significant proportion of AI-assisted forgeries, because the generation tool leaves its own metadata signature.

The new threat: documents that are technically real but misleading

Not all misrepresentation involves outright fabrication. A growing category of problematic applications involves documents that are genuine but selectively presented: pay stubs from a two-week window of unusually high earnings, employment letters that describe temporary contract work as a permanent position, or bank statements that show the account balance on a good day while omitting the month that preceded it.

These documents pass every authenticity check because they are authentic. The problem is in what they omit. The only defence against selective disclosure is requesting a complete and structured document set upfront — three consecutive pay stubs covering a recent period, bank statements for the last 90 days rather than a screenshot, and the most recent T4 or Notice of Assessment alongside current stubs — and reviewing them as a complete picture rather than individual passes.

What a more effective screening process looks like

The landlords and agents who are catching AI-assisted misrepresentation are not doing more work — they are doing different work. Specifically:

• Require a complete package upfront. Applications that arrive document-by-document over several days give applicants the opportunity to adjust submissions based on what questions are being asked. A structured, simultaneous submission requirement removes that flexibility.
• Verify employment independently, every time.Not through the contact number on the letter. Through the company's publicly listed phone number, confirmed via an independent search. This single step catches a disproportionate share of fraudulent employment letters.
• Check metadata before visual review. PDF metadata takes 30 seconds to inspect and eliminates a large proportion of AI-generated documents before you spend time on anything else. Make it the first step, not the last.
• Use document integrity tools for cross-checking at scale. When you are reviewing multiple application packages, manual cross-referencing of every figure across every document is not realistic. Automated integrity checks that flag YTD inconsistencies, metadata anomalies, and cross-document discrepancies let you focus your manual attention on the applications that actually warrant it.

The broader context

This is not a problem unique to Toronto or to Canada. AI-assisted document generation is a global phenomenon affecting rental markets wherever screening relies on self-submitted PDFs. What makes the Ontario context particularly acute is the combination of a competitive rental market — even as rents soften in 2026, quality units still attract multiple applications — and an LTB dispute process that takes months to resolve when a tenancy goes wrong. The cost of a bad placement is not just lost rent. It is a prolonged legal process with no guarantee of a timely resolution.

Screening more carefully upfront is not paranoia. It is the rational response to a dispute resolution system that makes getting it wrong very expensive.