Almost everything published about ad fraud is written by companies selling fraud detection, which makes the genre relentlessly alarming. So here's an unusual thing to report: one of the biggest problems in programmatic advertising was substantially fixed, and it happened because people simply started looking.
The number that genuinely improved
In 2023 the ANA's programmatic transparency study analysed a large pool of advertiser spend and found roughly 15% of programmatic dollars reaching made-for-advertising sites — pages built to harvest ad revenue rather than to be read. The figure became a rallying point, and the industry responded.
Subsequent ANA quarterly benchmarks tracked what happened next. MFA exposure fell to around 1% in 2024, and by late 2025 the median sat near 0.4% of spend. Over the same period, working media on web and mobile — the share of each dollar reaching a quality impression — climbed from around 36% to roughly 47%.
That is one of the clearest measurable improvements in the history of programmatic advertising, and it deserves to be reported as such.
What actually fixed it Nobody invented a fraud-proof exchange. Buyers got log-level placement data, read it, and excluded what they found. The leak responded to attention.
Other measures moved with it. Brand safety violations in programmatic display fell from around 5.8% in 2023 to roughly 3.4% in 2026. Verification tool adoption among advertisers spending over a million on programmatic rose from about 82% to 96%. By Q4 2025 the ANA reported that disciplined advertisers were converting 56.7% of programmatic spend into impressions meeting quality criteria — fraud-free, measurable, viewable and MFA-free.
Now the four qualifiers
Each of these is significant enough that reporting the headline without them would be misleading.
1. That figure describes advertisers who were being measured
The ANA benchmark tracks organisations participating in transparency audits — by definition a self-selected group with the resources and the intent to care. It isn't a measurement of the open programmatic ecosystem.
The ANA's own reporting makes this explicit: improvement was unevenly distributed, and the worst-performing quartile of measured advertisers still showed MFA exposure ranging from around 3% to over 27%. Among advertisers not being measured at all, nobody knows — which is rather the point.
2. Spend share and impression share are different numbers
This causes more confusion than anything else in the category, and it's why you'll see MFA reported at under 1% in one place and 15% or higher in another, with both being accurate.
MFA inventory is deliberately cheap. So it wins a large volume of impressions while consuming a small share of dollars. Supply-side tracking has continued to report MFA taking a double-digit percentage of open-exchange impressions even as its share of audited spend collapsed.
3. AI made the supply cheaper while buyers were shutting the demand
An awkward asymmetry. Generative tooling reduced the cost of producing a plausible-looking content site to near zero, and supply-side monitoring reported a substantial year-on-year rise in the number of active MFA domains through 2025.
So MFA properties multiplied while the share of disciplined budgets reaching them fell. The inventory is abundant and largely unsold to anyone paying attention — a better outcome than it sounds, but it means the problem is contained rather than solved. Any lapse in vigilance meets a much larger supply than existed in 2023.
It's also worth noting what these sites look like now. The old tells — slideshow articles, twelve ad slots above the fold, sticky autoplay video, listicles rewritten from public sources — still apply, but the copy reads better than it used to. Detection based on "does this look badly written" no longer works, and the reliable signals are structural: ad density, pagination patterns, traffic sourcing. The underlying dynamic is the same one described in the content sameness problem, monetised.
Adwisely
AI ad automation for Shopify stores running Google and Meta retargeting and prospecting campaigns.
Best for: Shopify stores automating Google & Meta ads
4. Fraud followed the money to CTV
The most important qualifier, and the one that determines where the next few years of this go.
Connected TV now represents around 40% of programmatic spend. It carries high CPMs. It uses server-side ad insertion, which limits the verification signal available. That combination — large budgets, weak measurement — is exactly what attracts fraud.
Verification vendors reported multiple new CTV bot variants appearing within single quarters of 2025, with individual schemes capable of draining substantial monthly sums. Meanwhile overall sophisticated invalid traffic across the ecosystem has stayed stubbornly high since 2023 despite record verification spending — because as budgets migrate, fraud capacity migrates with them.
If you're increasing CTV investment — and many advertisers are — the assumption that premium environments are inherently clean is the wrong one to carry into it.
What to actually ask about CTV inventory
Four questions, none of which require specialist tooling to put to a partner.
Can you see the app or channel name for every impression? If placement reporting shows aggregated categories rather than named properties, you cannot verify anything. This is the equivalent of the log-level data that fixed MFA on the web.
What proportion served on devices versus in browsers? A meaningful share of supposedly connected-TV impressions historically served in environments that were not televisions at all.
What verification is actually running, given server-side insertion limits what can be measured — and what does the vendor concede it cannot see?
How many intermediaries sit between your budget and the publisher? Each one takes a fee, and long paths correlate with poor outcomes independently of fraud.
These are the same questions worth asking of any newer channel where the transparency norms haven't settled, including the shoppable formats now being built on top of CTV.
The finding that undermines the good news
One statistic from the ANA benchmarks deserves more attention than it has received, because it contradicts the story the industry tells about itself.
Supply path optimisation — buying through fewer, better partners — has been the stated direction of travel for years. The data says the opposite happened.
| Measure | Earlier benchmark | Later benchmark |
|---|---|---|
| Supply-side platforms used | ~14 | ~19 |
| Domains and apps bought | ~22,600 | ~53,800 |
The median advertiser more than doubled the number of properties it bought from, while adding exchanges rather than consolidating. Buying across fifty thousand domains is not a strategy anyone chose; it's what happens when broad automated buying is left to optimise on cost.
This matters because exclusion lists work retrospectively. You can block the MFA sites you know about; you cannot block the ones that appeared last week in a supply pool you've never inspected. A wider footprint means more surface area for exactly that.
The waste that isn't fraud
Worth separating, because conflating the two produces bad decisions.
ANA reporting has put total unrealised value in the open programmatic ecosystem in the tens of billions annually — but the composition has shifted. As outright fraud and MFA declined, what remained is largely structural: indirect supply paths adding fees without adding value, impression quality gaps, and inventory that is technically valid and practically worthless.
That distinction has a practical consequence. Fraud filters catch bots. They do not catch a supply path that passes your money through four intermediaries before it reaches a publisher. Those are different problems requiring different tools — log-level data and supply path work rather than verification tags — and buying more of the second when you have the first is a common and expensive misdiagnosis.
Some waste is also simply physics. A non-viewable impression at the bottom of a long page isn't negligence, and chasing perfect measurement past the point of diminishing returns is its own form of waste.
What this means if you're not a large programmatic buyer
Most of the above concerns open programmatic display, which many marketing teams barely touch. Three things still apply.
Search and social are not immune, but the exposure is different. Click fraud on search — competitors and bots clicking paid results — is a real cost, though platform-side filtering catches a large share and the platforms credit invalid clicks. The bigger practical leak on search is usually not fraud at all but irrelevant traffic you're paying for legitimately, which is a negative keyword and match type problem rather than a security one.
Retail media inherits the same questions. As budgets shift into retail media networks, ask the same things you'd ask of any supply source — where exactly did impressions serve, what verification is available, and can you see placement-level data. Newer environments have historically had weaker transparency, not stronger.
Attribution problems and fraud problems look identical from a dashboard. Both present as spend without outcomes. Before assuming fraud, rule out the measurement explanation — increasingly likely given how much attribution has degraded. This is a good argument for measurement approaches that don't depend on user-level tracking, which are also harder to fool.
A proportionate response
Scaled to how much programmatic you actually run.
If you buy open programmatic at scale: demand log-level placement data as a contractual term, review the domain list quarterly, maintain and update exclusion lists, and consolidate supply paths rather than adding partners. Treat CTV as requiring its own scrutiny rather than inheriting the trust its premium reputation implies.
If you buy mostly search, social and retail media: the fraud exposure is real but smaller, and your time is better spent on the ordinary waste discussed above. Check placement reports where they exist, exclude obviously irrelevant placements, and ask your retail media partners what verification is available.
Everyone: watch for the pattern that reveals it — high engagement metrics with no conversion lift. Traffic that clicks and never converts, at volume, is either fraud or a targeting failure. Both are worth investigating and neither is visible if you only look at cost per click, which is the metric most likely to look good in exactly this scenario.
For smaller advertisers there's a simplification worth stating plainly: the environments with the least fraud exposure are generally the ones where you can see exactly where your money went. Concentrated spend in a few well-instrumented channels beats thin spend across many opaque ones — an asymmetry that tends to favour focused operators over large diversified budgets, as set out in how smaller brands compete against bigger competitors.
And keep the vendor incentive in view. Anti-fraud companies produce most of the research in this space, and their commercial interest runs toward maximum alarm. That doesn't make the research wrong — the ANA data is independent and the trend it shows is real — but a number that arrives attached to a product demo deserves the same scepticism as any other.
The honest summary
Something real got fixed. MFA's share of measured spend fell by a factor of thirty or more among advertisers who looked, and that happened through transparency and exclusion rather than technology. It's a genuine industry success and a useful demonstration that these problems are tractable.
It's also incomplete. The improvement is concentrated among the advertisers already paying attention, the worst quartile still bleeds double digits, MFA supply grew even as its revenue fell, and fraud has been migrating toward CTV where measurement is weakest. Meanwhile the median advertiser's supply footprint expanded rather than consolidated, which is the opposite of what the stated strategy implies.
The transferable lesson isn't about fraud specifically. It's that the leak responded to attention — and that the advertisers who moved their numbers didn't find a better exchange. They just read the placement report.
Do you know where your media budget actually went?
We audit paid media placement by placement and rebuild buying so the spend reaches real people.
Explore Performance Marketing →Frequently asked questions
Did made-for-advertising sites actually get cleaned up?
Among advertisers who measured, yes, and dramatically. The ANA's 2023 transparency study found roughly 15% of programmatic spend reaching MFA sites; its subsequent quarterly benchmarks showed that figure falling to around 1% in 2024 and a median near 0.4% by late 2025. That is one of the clearest measurable wins in adtech's history. The important qualifier is that this measures advertisers participating in transparency audits — a self-selected and disciplined group — and the ANA itself reported the worst quartile still carrying MFA exposure ranging from around 3% to over 27%.
Why do MFA statistics vary so much between sources?
Because they measure different things. Share of spend and share of impressions produce very different numbers, since MFA inventory is deliberately cheap and therefore wins many impressions while consuming comparatively few dollars. Population matters too — figures drawn from audited advertisers look far better than figures drawn from the open exchange as a whole. So MFA can simultaneously be under 1% of audited spend and a double-digit percentage of open-exchange impressions, with both figures accurate. Always check the denominator and the population before quoting a number.
Where has ad fraud moved to in 2026?
Toward connected TV, and for straightforward reasons. CTV carries high CPMs, uses server-side ad insertion that limits verification signal, and now represents around 40% of programmatic spend. Fraudsters follow money and weak measurement, and CTV offers both. Verification vendors reported multiple new CTV bot variants emerging within single quarters of 2025, with individual schemes capable of draining substantial monthly sums. Overall sophisticated invalid traffic has stayed stubbornly high across the ecosystem even as MFA specifically declined.
Has generative AI made the MFA problem worse?
It has made MFA supply cheaper to create while buyer discipline has made it harder to monetise, which produces a strange split. Content production costs for a plausible-looking site fell close to zero, and supply-side tracking reported a substantial year-on-year rise in active MFA domains. So the number of MFA properties grew while the share of disciplined advertisers' budgets reaching them fell. The inventory is abundant and mostly unsold to buyers who are paying attention, which is a better outcome than it sounds but not a solved problem.
What should advertisers actually do about ad fraud in 2026?
Get log-level placement data and read it, because the single strongest predictor of low MFA exposure is whether anyone looks. Consolidate supply paths rather than adding partners, since benchmark data shows the median advertiser increased both the number of exchanges and the number of domains bought rather than reducing them. Apply exclusion lists and keep them current. And treat CTV with specific scrutiny rather than assuming premium environments are inherently clean, since that is where fraud capacity has been migrating.
