This blog post is slightly different. We recently compared Google search interest in creating fake documents with interest in detecting manipulated PDFs across four markets: the US, UK, Canada and India. The results were stark enough that I wanted to write about them myself.
Google reported combined monthly volume of 15,170 on the creation side and 2,000 on the detection side. That is 7.6 to one.
The direction held in every market and every month of the available 12-month history. In India, invoice-creation phrases outnumbered their matched detection phrases by nearly 78 to one.
These are Google Ads estimates, not a census of fraudsters. Keyword data cannot identify the person behind a search. But we also see creation-oriented queries reaching verifypdf.com in Search Console, complete with landing pages, countries, positions and clicks. That search activity is not merely hypothetical.
How we measured fake document creation against detection
You cannot read a person’s motivation from a search string. A systematic review in Social Science Research notes that “different groups of users may even search the same term for different and potentially opposing reasons”.
That is exactly what happens with a bare query such as “fake bank statement”. A person may want to make one or catch one. We excluded it.
A creation phrase had to contain both a production or procurement operator, such as generator, maker, create, template or buy, and an explicit fraud marker such as fake, forged or counterfeit. Detection used an equal number of constructions: detector, detection, checker, detect, spot and authenticity.
We applied those constructions to the same nine document labels: general documents, bank statements, US pay-stub wording, payslips, receipts, invoices, utility bills, proof of address and certificates. Fifty-four phrases per side, repeated unchanged in four markets. We retrieved the estimates through DataForSEO on 26 July 2026. The four tasks cost $0.36 in total.
The result:
- United States: 7,210 creation volume versus 830 detection, a ratio of 8.69 to 1.
- India: 6,120 versus 720, a ratio of 8.50 to 1.
- United Kingdom: 1,100 versus 260, a ratio of 4.23 to 1.
- Canada: 740 versus 190, a ratio of 3.89 to 1.
- Combined: 15,170 versus 2,000, a ratio of 7.59 to 1.
This is not a universal law. It is four separate estimates from one matched English phrase design. The lowest ratio is still nearly four to one.
The US breakdown shows where its 8.69-to-one result came from:
- Receipts: 3,120 creation searches versus 170 detection, a ratio of 18.4 to 1.
- Pay stubs: 1,650 versus 290, a ratio of 5.7 to 1.
- Bank statements: 1,090 versus 240, a ratio of 4.5 to 1.
- Invoices: 450 versus 30, a ratio of 15 to 1.
- Utility bills: 410 versus 10, a ratio of 41 to 1.
- Certificates: 190 versus 20, a ratio of 9.5 to 1.
- General documents: 180 versus 50, a ratio of 3.6 to 1.
- Payslips: 90 versus 20, a ratio of 4.5 to 1.
- Proof of address: 30 creation searches; Google reported no detection volume, so no ratio can be calculated.
Receipts are the clearest large-denominator US asymmetry at 18.4 to one. India produced the largest document-level ratio: 2,330 for its six invoice-creation phrases and 30 for detection, or 77.7 to one. The consequences of fake invoices extend well beyond search data, as the EU’s €800 million customs fraud case shows. Thirty is still a thin search denominator. I will not round seventy-eight into one hundred for effect.
Across the same canonical phrases, creation volume exceeded detection in every market and every month from July 2025 through June 2026. The monthly ratio ranged from 3.79 to one in Canada to 13.11 to one in the US. Both sides declined over the year in the US, so this is not evidence that forgery interest is growing. It is evidence that the disparity persisted.
The largest caveat points in the detection side’s favour. In the US, Planner returned measurable rows for 46 of 54 creation phrases before deduplication but only 21 of 54 detection phrases. It returned null for eight creation phrases and 33 detection phrases, with almost the same split elsewhere. Null does not mean zero. 7.59 to one is the ratio of reported volume in this matched set, not the ratio of all creation demand to all detection demand.
Google also groups near-identical keywords and may repeat one volume across several variants. We removed variants only when they had an identical non-constant 12-month history and identical CPC. An earlier analysis that failed to do this inflated creation demand by more than 10,000 searches a month. That version went in the bin.
The auction data adds another uncomfortable detail. “Fake paystub generator” carries 390 reported monthly searches, a $6.99 suggested CPC and HIGH advertiser competition. “Fake paystub maker” reaches $9.92. Source-based payroll checks can reduce exposure to fake payslips, but uploaded PDFs still need scrutiny. Google’s policy prohibits advertising fake-document creation, although competition does not prove the advertisers are forgery suppliers. Detection companies can bid on the same traffic.
Sometimes the query states the intended use
The most persuasive evidence is sometimes the qualifier. Queries reaching our pages include “for loan”, “for job” and “for apartment”. That last phrase is exactly the risk a fraud-resistant tenant screening workflow is meant to address.
“For apartment” appended to a fake-income-document search is stronger evidence than a bare noun. It still does not reveal who typed it or why.
Why these document fraud statistics matter to verification teams
If you verify documents for a living, the demand side is your problem whether or not keyword data interests you.
Fannie Mae reported income misrepresentation in 48% of loans with mortgage-fraud investigative findings in 2019, rising to 61% in 2020.
An NMHC and NAA survey published in January 2024 found that 84.3% of US apartment operators who had experienced fraud saw falsified payslips, employment references or other income documents. The figures are self-reported and the denominator is 70 operators, so treat that as an industry signal rather than a population measurement.
For covered US mortgages, Regulation Z’s ability-to-repay rule requires creditors to verify income using reasonably reliable third-party records. Its examples name “payroll statements” explicitly.
Current ISA 240 says auditors “may accept records and documents as genuine” unless they have reason to believe otherwise. Its revised replacement, effective for periods beginning on or after 15 December 2026, says auditors are not required to run procedures designed specifically to identify an inauthentic or modified document.
These are different actors with different duties, but neither rule is a document-authentication manual. Meanwhile, “fake paystub generator” carries a $6.99 suggested CPC and HIGH advertiser competition.
That is why VerifyPDF exists. Someone reading a payslip is judging a story. Our document fraud detection software adds a file-based risk rating to that human judgment.
The supply side is twenty years old, and AI is accelerating it
FinCEN documented the supply chain in 2006. One mortgage-fraud scheme offered employment and income verification in any amount for an extra fee of one percent of the claimed annual income. Priced as a percentage of the lie. Today, the same market operates through industrial fake-document template farms.
In 2018 the FTC sued operators selling fake payslips, bank statements, tax forms and employment-verification services. Their “for entertainment purposes only” disclaimer “was easy for purchasers to remove”. One FAQ coached buyers on what to say if caught.
AI changed the scale and speed. AppZen data reported by Forbes shows AI-generated receipts rising from 0% of its fake-receipt flags in March 2025 to 70.8% by mid-May 2026. The next step is already visible in AI agents submitting fake documents at scale.
Do not overcorrect. Inscribe’s 2026 document fraud report says AI-generated fraud rose nearly fivefold between April and December 2025, but still comprised less than 5% of fraudulent documents detected across its network that year. In Inscribe’s network, conventional methods still accounted for most detected fraud. AI is making it faster.
What the fake document search gap means
The headline is not 77.7 to one. That figure comes from one document category in one market with a thin denominator. The defensible headline is 7.6 to one across four matched market estimates.
That gap is not a marketing opportunity. It is a diagnosis. Explicit detection-oriented search volume is much lower than creation-oriented search volume in this matched phrase set.
The data cannot show whether detection buyers use different terminology, established suppliers or channels other than search. It can show that creation-oriented wording is explicit, attracts measurable advertiser competition and persisted throughout the available 12-month history.
We built VerifyPDF for the quieter side of that ratio. Read how a fake PDF detector identifies manipulation, or run one real file through the free check before concluding that nothing is getting past your document pipeline.