Blog · Signals

MRC invalid traffic rules require a decision rate. Unknown impressions are not clean.

August 31, 2026 · RTBlint

Invalid traffic headlines are almost always a percentage of classified impressions. The Media Rating Council requires something else in the same report: a decision rate, the share of rendered impressions where the vendor had enough information to reach a verdict. In the MRC worked example, 85 of 100 rendered impressions qualify and 15 are unknown. Unknown traffic must not be assumed valid. That rule is what separates a vendor dashboard from a population you can compare.

What the metric actually measures

Invalid traffic is advertising activity that should be removed from the numbers advertisers pay against. The MRC splits it into General Invalid Traffic, identified through lists and standardized parameter checks, and Sophisticated Invalid Traffic, which requires advanced analytics and human review. GIVT detection is mandatory for every accredited measurement organization. SIVT detection is optional and separately accreditable.

The consequence is regulatory, not cosmetic. GIVT is meant to be consistent between vendors because it leans on shared resources: the IAB/ABC International Spiders and Bots List and the Trustworthy Accountability Group data-center address list. On data centers the MRC requires filtration of invalid traffic from the three largest hosting entities: Amazon AWS, Google, and Microsoft. SIVT findings must be segregated and disclosed only at campaign total level, because transaction-level detail would let operators reverse-engineer the logic.

Back-end detection is mandatory. Sophisticated adjustments made after a campaign closes must land within 14 days of completion. Up-front blocking is not required, and the MRC is openly cautious about it, warning that blocking telegraphs the method to the traffic source and invites A/B testing against the filter.

Because filters only work on the signals they receive, vendors must also publish a decision rate: recorded impressions where enough information existed to reach a verdict, divided by total impressions intended for measurement. Fifteen unknown out of one hundred rendered impressions is a 15 percent unmeasured bucket sitting beside an IVT rate computed on the other 85. Treating the headline IVT percentage as if it described all one hundred is the mistake practitioners make when they compare vendor charts without reading the denominator footnote.

Where the telemetry lives on the bidstream

The MRC lists the signals that determine whether an impression is classifiable. They map cleanly onto a programmatic stack: IP address including X-Forwarded-For, unmodified user agent, device and app identifiers, referrer, consent string, publisher and placement IDs, ads.txt information, pre-fetch headers, and OpenRTB attributes.

// Fields that feed IVT decision rate, not just pre-bid lists
"device": {
  "ip": "...",
  "ipv6": "...",
  "ua": "...",              // unmodified; SSAI server UA → unknown
  "ifa": "...",
  "lmt": 0
},
"app": {
  "bundle": "...",          // April 2024 MRC update: CTV bundle spoofing
  "publisher": { "id": "..." }
},
"source": {
  "ext": { "schain": { ... } }   // ads.txt / sellers.json chain
},
"regs": {
  "ext": { "gdpr": ..., "us_privacy": "..." }  // consent context
},
"imp": [{
  "id": "...",
  "video": { "placement": ..., "plcmt": ... }
}]

A bid request that clears the auction with a thin device object is not automatically fraud. It is often unclassifiable. Pre-bid datacenter lists and device reputation feeds answer a narrow question on the fields they receive. Post-bid IVT measurement answers a broader one on the same fields, and marks the gap as unknown when the chain breaks. That gap does not appear on a seller fill report. It appears inside the buyer's measurement vendor as impressions that never entered the IVT numerator or denominator the dashboard headline assumes.

The April 24, 2024 MRC interim update added bundle ID spoofing in connected TV to the IVT guidance. CTV supply paths that strip or rewrite app.bundle do not only confuse targeting. They remove one of the corroboration points SIVT detection depends on, which pushes impressions toward unknown or toward misclassification when the vendor guesses from partial data.

Why the published rates disagree by an order of magnitude

Reported IVT rates diverge depending on who measures what. Verification vendors report low single digits on protected inventory. Integral Ad Science measured global invalid traffic at 1.203 percent before the 2026 FIFA World Cup and 1.134 percent after kickoff, against an expected seasonal decline to 0.876 percent. DoubleVerify reported fraud and SIVT violation rates down 41 percent year over year in North America and 45 percent in EMEA on July 29, 2026, with North America still the highest tracked market at 0.6 percent on protected campaigns.

Click-level vendors report far higher figures. Lunio recorded LinkedIn at a 17.62 percent invalid traffic rate in the first quarter of 2026, and found retail search campaigns running Google AI Max exposed to 72 percent more invalid traffic than campaigns without it, at 5.28 percent against 3.07 percent in the second quarter of 2026.

Part of the gap is definitional: clicks and impressions are different denominators, and protected campaigns are not a random sample. A March 2025 investigation spanning more than a petabyte of traffic concluded that at least 40 percent of web traffic is non-human, and found cases where verification tools identified a visitor as a bot and served an ad anyway.

Reclassification moves the numbers too. Declared AI crawlers count as GIVT rather than fraud, and their volume is new: DoubleVerify recorded an 86 percent year-over-year rise in GIVT during the second half of 2024, 16 percent of it linked to AI tools. Changes to three MRC accredited metrics in Display and Video 360 on July 14, 2025 were expected to roughly double reported invalid and GIVT begin-to-render impressions, a detection improvement that reads on a dashboard as deterioration.

None of those comparisons is honest without the decision rate beside them. A protected campaign at 0.6 percent IVT on 95 percent classified impressions is a different claim from 0.6 percent on 60 percent classified. The second leaves 40 percent of spend in the unknown bucket the MRC says you cannot treat as clean.

The mechanism on defended versus open traffic

Pre-bid filtration and post-bid measurement sit on opposite sides of the same auction. A buyer who blocks datacenter ranges before pricing never sends those impressions to post-bid IVT at all. A buyer who measures only after delivery sees everything that served, including traffic missing the telemetry needed to classify it. Sellers who compare their low platform-side IVT credits to a buyer's post-bid report are often comparing different populations without knowing it.

Google's buy-side documentation lists six types of invalid activity on search, including double-click duplicates, competitor clicks, automated tools, known data-center traffic, and impressions designed to depress clickthrough rate. Activity caught before an invoice generates is never billed. Activity caught afterwards produces credits, which Google states are adjustments rather than refunds. Campaign-level visibility into those credits arrived when Google published help documentation for the Invalid Activity Credit Report on June 1, 2026, limited to Search and Performance Max.

Sell-side exposure runs in reverse. Publishers receive no credits; they lose revenue and, inside Google's owned channels, serving capacity. A Lunio survey of 131 senior marketing leaders fielded in May 2026 found 5.3 percent using dedicated invalid traffic tools, with 39.7 percent citing confidence in platform-side prevention as the reason not to, while 38.9 percent rated their trust in platform invalid click credits at 4 or below on a ten-point scale. Platform-side prevention and third-party decision rates are not the same instrument.

What to do

Ask for decision rate beside every IVT headline. If the vendor report shows 1.134 percent global IVT without stating how many impressions were classifiable, you do not have a rate. You have a rate on an undisclosed subset.
Audit the telemetry list on live CTV traffic. Sample bid requests from every path you sell through, including SSAI and reseller hops. Count how often device.ua is a server string, how often device.ip is missing while the impression still clears, and whether app.bundle is stable across exchanges for the same title.
Split pre-bid blocks from post-bid unknowns in reporting. Datacenter blocks remove inventory before measurement. Unknowns are inventory that served without classifiable signals. Merging them in one "clean" line hides structural gaps.
Validate requests before they leave your stack. Structural errors (missing device fields, inconsistent pod objects, deprecated-only video enums) are not IVT scores. They are reasons a downstream classifier marks unknown. Paste production traffic into the bid request tester or run the CI validation guide on every integration release.

RTBlint does not detect invalid traffic, bots, or fraud. It flags malformed or inconsistent OpenRTB payloads. A request that fails validation is a request another system may refuse to classify, which is a different failure mode from a high SIVT rate on traffic that looked well formed.

Sources

MRC Invalid Traffic Detection and Filtration Guidelines (Version 1.0 final October 27, 2015; IVT 2.0 final June 25, 2020; interim updates April 24, 2024) define GIVT, SIVT, decision rate, and required telemetry. Field names in this post follow OpenRTB and AdCOM as published by IAB Technology Laboratory. The analysis is our own. RTBlint is independent and not affiliated with IAB, IAB Tech Lab, MRC, or PPC Land.