Dr. Augustine Fou, the ad fraud researcher who sells the FouAnalytics verification tool, argued in a column published on October 16, 2026 that the widely cited estimate of $63 billion lost to invalid traffic understates fraud, because it counts invalid clicks arriving on websites rather than invalid ad impressions. His evidence is one set of FouAnalytics charts from an unnamed campaign, covering January 8 to February 7 of an unstated year.

In Short

An ad fraud researcher says bots that are paid for showing ads do not bother clicking on them, so a website's own count of fake visitors misses most of the fake ad views. He uses one campaign's charts to show about 80% fake ad views next to under 10% fake site visits. That matters because the most quoted fraud estimate, $63 billion, was built from click data, and he argues the real loss on display ads bought per thousand views is hidden.

The column and its author

The piece, titled How 80% Impression Fraud Hides in Site Traffic, appeared on State of Streaming, a streaming TV advertising publication where Fou writes a recurring column under the "see Fou yourself" label and carries the site's "Founding Voice" badge. The page shows only "Oct 2026" on the byline; its page metadata dates publication to October 16, 2026. The chart images in the article are hosted on LinkedIn's media servers, which suggests the material circulated there first.

The column is the author's own analysis, not a peer-reviewed study or a company disclosure. Fou also runs a commercial product in the same market as the vendor he criticises. In August, FouAnalytics priced unlimited ad verification at a flat $2 million a year, and in September Inuvo added FouAnalytics post-bid measurement to its IntentKey product. The column closes with a subscription pitch for weekly FouAnalytics posts on State of Streaming. That context matters for reading the argument, though it does not decide whether the argument holds.

The structure is simple. "Many folks cited a study that said '$63 billion was lost to invalid traffic.' It was wrong. Here are three reasons why," according to Fou. The three reasons are the difference between invalid clicks and invalid impressions, the reliability of IP address detection, and the arithmetic used to turn rates into dollars.

Whose $63 billion?

Fou does not name the company behind the study, and the PDF capture of the column contains no hyperlink to it. Two details identify it, though. The column quotes the vendor as saying "TikTok recorded the highest average IVT rate at 24.2 percent." Those exact words appear in Advanced Television's January 21, 2026 report on Lunio's 2026 Global Invalid Traffic Report, and MediaPost reported on January 20, 2026 that Lunio put waste at $63 billion. Lunio's own website states that "$63 billion was lost to invalid traffic in 2025, according to our 2026 Global Invalid Traffic Report."

According to that report, Lunio analysed more than 2.7 billion paid ad clicks between August 1, 2024 and July 31, 2025, across Google, Meta, Microsoft Bing, LinkedIn, TikTok and X, in eight industries and ten countries. It found that 8.51% of paid traffic was invalid. Platform rates reported by Advanced Television were 24.2% for TikTok, 19.88% for LinkedIn, 12.79% for X, 10.32% for Bing, 8.2% for Meta and 7.57% for Google Ads. Within Google, Display came in at 12.02% and Video at 20.62%, against 5.21% for Search.

Lunio is no stranger to PPC Land's pages. Its survey of 131 marketing leaders found that only 5.3% run a dedicated invalid traffic platform, and a separate retail dataset showed retail search campaigns using AI Max with 72% more invalid traffic than those without it. Like Fou, Lunio sells detection, so both sides of this disagreement are vendors.

Reason one: bots click only when clicks pay

Fou's central argument rests on how ads are bought. For impressions sold on a CPM basis, "all the bots have to do is cause the ad to load in order to get paid; they don't need to waste time and effort to click on those ads, so they don't click," according to Fou. For cost-per-click inventory the incentive reverses: "the bots have to click in order to get paid, so they do click." He adds that nearly 20 years ago he was already seeing click-through rates above 100% for pharmaceutical clients on search ads - a claim offered from memory, without data in the column.

If that logic holds, a rate of invalid traffic measured on landing pages says little about the share of display impressions that never reached a person. Fou puts it bluntly: "The invalid rate measured on the site severely underestimates the percentage of invalid impressions - wasted ad spend."

What the two charts show

The evidence is a pair of stacked-area charts from FouAnalytics, both covering January 8 to February 7. The year is not stated. Neither the advertiser, the campaigns, the exchanges, nor the total number of impressions measured is disclosed.

The first chart, labelled "impressions," shows the share of ad impressions in each category, with dark red marking invalid. "The chart above shows 80% of the impressions are invalid (dark red)," according to Fou. The chart itself is less uniform than that sentence. On January 8, the dark red band sits at roughly 60%. It climbs over the following two weeks and runs between roughly 80% and 85% from late January to February 7. A reading of the period as a whole would put the invalid share below 80% for the first half of the window and around or slightly above it for the second. The headline figure fits the later weeks, not the full 31 days.

The second chart, labelled "on-site," shows visits to the advertiser's website over the same dates. Here dark blue, which FouAnalytics uses for humans, dominates. "Note the dark red is sub-10%," according to Fou, and the chart supports that: the dark red base mostly stays below 10%, with short peaks. The chart also carries orange and lighter red layers above the dark red that reach 20% to 40% on several days and spike above 60% around January 11 and February 4. The column does not explain what those colours mean, so the share of site traffic FouAnalytics treats as questionable, rather than confirmed invalid, cannot be read from the post.

There is a further gap. The on-site chart covers all site traffic, not just visits from the display campaigns in the first chart. The comparison is therefore between invalid impressions for one set of buys and the overall mix of visitors from every source, which supports Fou's direction of argument but is not a like-for-like ratio.

The data grid

The column adds a table of site visits whose URLs carried a utm_source tag. According to the grid's header, 20,641 visits had the tag, 92% of 22,418 recorded visits. Most source names are blurred. The largest rows are 7,274 visits (35.2%) and 6,941 visits (33.6%), followed by "tiktok" at 2,019 visits (9.8%), then rows of 1,161, 1,000, 439 and 134 visits.

Fou highlights six rows in yellow, identified only by numeric IDs, as clicks from display campaigns bought on a CPM basis. Their counts are small: 94, 92, 74, 49, 31 and 31 visits, or 371 visits combined, about 1.8% of tagged traffic by PPC Land's calculation. Some of those rows are mostly blue; others show long red segments. "Note how small the counts are (bots didn't need to click through to the site in order to make money on CPM campaigns)," according to Fou, who adds that the colour coding showed "surprisingly, more blue than expected," meaning some humans who saw the ads did click.

The 13 rows visible in the capture add up to 19,339 visits, so the remainder of the table is not shown.

TikTok

The TikTok row is where Fou addresses Lunio's platform ranking directly. "That's the clicks coming to the site, not the percentage of invalid impressions," according to Fou. In his grid, the 2,019 TikTok visits are almost entirely dark blue, with a thin red segment at the end of the bar. He attributes that to turning off Pangle, TikTok's audience network, and says good results on TikTok are possible with Pangle disabled. The column gives no invalid rate for the row and no comparison with Pangle switched on.

This is one advertiser's account set against a 2.7 billion click aggregate, so it cannot show that Lunio's 24.2% is wrong. What it shows is the definitional point: the two numbers measure different things.

Reason two: IP addresses no longer sort bots from people

Fou's second objection concerns detection. Fifteen years ago, he writes, IP addresses were enough because "bots came from data center IP addresses and humans came from residential IP addresses." That division, according to Fou, no longer holds.

"Most botnets disguise themselves using residential proxy services, so their IP addresses look like residential IPs," according to Fou. A residential proxy network is typically built, in his account, from malicious "free VPN" apps that people install without knowing what they do. His example is a household iPad carrying such an app, used to route bot traffic so it appears to come from the home connection. A household might own 10 to 15 devices, and only the iPad is compromised. Blocking that IP address therefore blocks the humans in the house as well.

PPC Land has documented the supply side of this market. The BADBOX 2.0 botnet infected more than 10 million uncertified devices, mainly streaming boxes, and access to compromised home connections was resold to other criminals. Any botnet of that kind gives fraudsters residential addresses at scale.

The reverse error also exists, according to Fou. A data center IP address can belong to a person using a VPN, and "based on FouAnalytics data, that percentage has gone up from 0.1% to 1-2% these days." The column gives no dates or sample for either figure.

His conclusion is that IP-based tools err in both directions: they block data center addresses, catching humans, and miss residential ones, letting bots through. He extends the criticism to blocking itself. If invalid clicks are filtered out of site reports, he argues, advertisers lose sight of problems such as a Facebook or Instagram campaign running with Audience Network switched on, where in his description "virtually all of your clicks were invalid." Fou says he has recommended that advertisers not block invalid clicks to their sites, because "blocking invalid clicks from showing in your reports doesn't make campaigns better; it just hides the truth from you."

The column does not say whether Lunio's detection relies on IP addresses, or to what extent. It treats IP methods as a general weakness of click-detection vendors rather than demonstrating it for the report in question.

Reason three: rates multiplied by spend

The final objection is short. Fou writes that it is "incorrect to just multiply the invalid click rate by country by the ad spend by country to get the 'dollars lost to invalid traffic.'"

Lunio's report, as published, supports the description. In a check by PPC Land of the report PDF, the country table pairs each invalid rate with 2024 digital ad spend: China's 16.37% against $143 billion produces the stated $23.4 billion. The global headline follows the same shape, with the report saying an average rate of 8.51% "means that $63 billion dollars were lost," against total spend it expected to exceed $740 billion in 2025. The report does not set out how the 8.51% average was weighted, and the country table uses 2024 spend while the global figure uses a 2025 projection.

Fou's objection is that a click-based rate is being applied to total digital spend, much of which is bought by impression. On his reasoning, that undercounts waste in CPM display, though the column offers no alternative figure.

No number of his own

That absence is deliberate. Fou lists three reasons he does not publish global estimates of losses to fraud: "I don't need the publicity"; "I don't have complete measurement, so I don't have the data to substantiate it"; and it "would be irresponsible of me to just extrapolate and make incorrect calculations to arrive at a big scary number for fraud that everyone then breathlessly cites in blog posts and articles."

He goes further, arguing that no fraud number has changed behaviour. "Whatever the fraud number - regardless of how big it is - it doesn't cause any substantive corrective action," according to Fou. "It's nothing more than publicity for the company that published it and will be forgotten by the next news cycle."

The position has a cost for his own case. The column title promises 80% impression fraud, but the only evidence is a single unidentified campaign over one month. Without a sample of campaigns, there is no basis for treating 80% as typical, and Fou does not claim it is.

Why the dispute matters for buyers

For advertisers, the practical question is which number reaches the budget conversation. Click-based figures are what most performance marketers see, through tools on their own sites and through platform credits such as Google's Invalid Activity Credit Report, which covers Search and Performance Max. Impression-level invalid rates on display and video buys sit with verification vendors and the exchanges.

Published fraud figures already point in different directions. DoubleVerify reported this summer that fraud and invalid traffic fell 41% in North America and 45% in EMEA year over year, while Lunio's click data showed LinkedIn at 15.34% invalid traffic over nine months of 2025 and 2026. Adalytics argued in 2025 that verification vendors labelled declared bots as valid humans, and Fou has made a similar case about CTV, where ads keep running when the TV is off. Each source measures a different layer, with a different method, and none of them reconciles with the others.

Fou's column does not settle the size of the problem. Its contribution is narrower and harder to dispute: a rate built from clicks on landing pages measures click fraud on paid traffic, and cannot stand in for the share of impressions that bots consumed upstream. Whether the gap is as wide as 80% against under 10% across the market is a question his one campaign cannot answer.

Timeline

Summary

Who: Dr. Augustine Fou, founder of the ad verification tool FouAnalytics, writing a column on State of Streaming. His target is the $63 billion invalid traffic estimate that matches Lunio's 2026 Global Invalid Traffic Report, which the column does not name.

What: Fou argues the $63 billion figure understates fraud because it is built from invalid clicks on websites, while bots on CPM display buys only need ads to load. He shows FouAnalytics charts where about 80% of impressions were invalid in the later weeks of one campaign, against under 10% dark red invalid traffic on the advertiser's site. He also says IP-based detection misses residential proxies and blocks VPN users, and that multiplying click rates by ad spend is wrong.

When: The column was published on October 16, 2026. The charts cover January 8 to February 7 of an unstated year. Lunio's report was released on January 20, 2026, covering August 2024 to July 2025.

Where: The column appeared on State of Streaming. The disputed figures concern paid media worldwide, including programmatic display, TikTok, Meta and Google.

Why: The most cited fraud figure shapes how advertisers judge their exposure. If click-based rates cannot capture impression fraud on CPM buys, the waste in display budgets is not visible in site analytics. Fou's evidence comes from one undisclosed campaign, and he publishes no market-wide figure of his own.