TLDR Google's Failures on Full Display

It's been a terrible month for Google. Their failings were laid bare by the hard work of a single, independent researcher, Krzysztof F. of Adalytics . The crux of the story is that advertisers spent BILLIONS of dollars over many years, trusting the self-reported data from the platforms and other adtech vendors. Hundreds of pages of detailed documentation of these platforms' failures to correctly report on their own data has opened the eyes of advertisers. Perhaps they shouldn't be so trusting of self-reported data, including the data in log files. Obviously, there is unlikely malicious intent, but the data is clearly wrong, as reported to Google's customers. Let me summarize the last 3 cases.

Video ads sold as TrueView, but did not fit Google's own definition of TrueView

TrueView ads are the special form of ads that Google sells where the advertiser doesn't pay unless the user chooses to watch the ad -- i.e. NOT skip it. Most people are familiar with this because they have seen it when watching YouTube videos. The original Adalytics research here -- Did Google mislead advertisers about TrueView skippable in-stream ads for the past three years? -- showed that a large quantity of ads were sold as TrueView ads, but did not meet Google's own definition of what a TrueView ads should be. TrueView ads are video ads that run in-stream, audible, and skippable. In-stream means the ads run next to YouTube video content, either before (pre-roll), during (mid-roll), or after (post-roll), much like TV ads before, during, and after TV shows. Further, Google doesn't charge for the impressions, only for "views" which are the cases where the user chooses not to skip the ad. The ads are also audible and visible.

The screenshot above is one of hundreds of examples documented by the Adalytics research, which shows video ads that ran off-page, out-stream, auto-played, muted, and not-skippable. Off-page, means it ran outside of the part of the webpage the user is looking at. Out-stream means the video ad ran by itself, not next to a YouTube content video. Auto-played means the ads were automatically played, without any user initiated event -- e.g. clicking to play a YouTube content video (because there was no associated YouTube content video). Muted is self-explanatory. And non-skippable because the user couldn't even see the ad to skip it, or in some cases the skip button was obfuscated or deliberately hidden or obstructed by something else (another banner ad on top of the video ad blocking the skip button). These video ads were sold as TrueView, but they did not fit Google's own definition of TrueView ads.

This problem occurred over a period of three years (since 2020) and was documented by Adalytics across the entire timeframe. It affected 1,100 brand advertisers, whose names were listed because their ads were detected and documented by Adalytics. And because of the enormous size of these advertisers and their ad budgets, likely billions of ad impressions and hundreds of millions of dollars were affected. The vast majority of this "misrepresented TrueView ads" problem occurred when the ads ran on GVP ("Google Video Partner") sites and apps. Most of the ads that ran on YouTube itself were fine, and met the definition of TrueView ads. The recommendation to advertisers is that they turn off GVP entirely and force their ads to run on YouTube only.

Google's blog post in response to the 200 pages of evidence from Adalytics was a non-response, meant to misdirect readers. While it contained true statements, none of those statements addressed the evidence at hand. For example, Google said that the vast majority of CAMPAIGNS ran on YouTube. That is accurate. But the vast majority of video ad IMPRESSIONS ran off of YouTube and a small portion ran on YouTube. Google also stated they have very strict policies and publishers must follow them. Obviously those sites and apps outside of YouTube did not follow these strict policies, because they served video ads off-page, out-stream, auto-played, muted, and non-skippable. The fact that Google didn't detect this themselves is problematic. What is even more problematic is when Google tried to hide behind what they called "independent verification." Google cited IAS and DoubleVerify which reported that most ads were audible and viewable. Turns out, Google gave them the data and these 2 vendors simply "performed calculations" and "provided reporting." Advertisers now realize that this is not "independent verification" and that neither of these vendors actually measured anything with JavaScript tags.

Non-kid ads shown on kids channels and videos

The previous study showed that video ads that ran outside of YouTube on video partner sites and apps mostly did not conform to the standard of TrueView ads but were sold as TrueView. Google and the outside vendors they cited, IAS and DV, also did not catch this problem for years, so these non-conforming video ads were sold to advertisers as if they were TrueView. In this second case, Adalytics found problems with the ads that ran on YouTube itself -- i.e. ads for adults (non-kid ads) were shown on "made for kids" channels and videos.

For all the gory details, screen shots and examples, see the original 120 page research post here: Are YouTube Advertisers Inadvertently Harvesting Data From Millions of Children?

I will highlight a few points here which document Google's failures, both of their own tech and also of the enforcement of their own "strict" policies. It's one thing to have strict policies, it's another thing to actively and properly enforce them. Quoting from the research:

• YouTube is serving ads from major brands such as Mars, Procter & Gamble, Ford, Colgate-Palmolive, Samsung, and many others on YouTube channels that are labeled as “made for kids.”
• Viewers of ads on "made for kids" channels are clicking on the ads and going to brands’ websites – such as Michigan State Police, Disney, BMW, Hyundai, and Verizon. Those sites have code from dozens of data brokers that harvesting and sharing meta-data on those viewers upon the click through.
• Google’s Performance Max (“Pmax”) ad targeting algorithm appears to be placing adult brands’ ads on “made for kids” YouTube channels; advertisers report they cannot audit this issue because Pmax does not provide them with granular placement reports.
• Google's own content classifiers appear to be incapable of reliably detecting "made for kids" videos or even entire channels; this is leading to advertisers' non-kid ads being shown on "made for kids" content even though the advertiser has explicitly tried to avoid placing their ads on such videos.
• media buyers report that Google’s software controls make it very difficult if not impossible to completely avoid “made for kids” videos. Keyword exclusion lists for terms like “kids” or “children” do not appear to prevent brands’ ads from being served adjacent to YouTube videos that explicitly contain the words “kids” or “children” in the title of the YouTube video.

The takeaways that advertisers and their media buyers are arriving at after seeing this evidence are:

• Google's own tech -- e.g. content classifiers -- are failing to properly classify "made for kids" content even if the videos had keywords like "Nursery Rhymes" and "Kids Songs" in the title; so Google is not able to help advertisers prevent their ads from being served adjacent to kids videos even if the advertiser wanted to avoid this adjacency.
• Google's own optimization algorithms used in PMax ("performance max") are also placing non-kid ads on kids channels and videos; advertisers using PMax don't know this is happening because the placement reports don't provide this kind of granular detail.

Quoting a senior advertising executive:

“Google has failed advertisers, again. There is no reasonable excuse for ads running on content intended primarily for kids other than to extort advertisers through a toddler-enabled click farm. The observations around PMax (Preschooler Max) are damning given the hard sell Google is putting on us to trust their so-called AI black box. We're overdue real transparency and Google needs to be made accountable - refunding us for all ads on this content and explaining themselves to the FTC.”

Google's blog post in response to the 120 pages of evidence from Adalytics was yet another "we have strict policies" post with statements that are true, but not pertinent to the evidence at hand, and another attempt at hiding behind verification vendors -- "It was subsequently debunked by multiple independent third parties, including DoubleVerify, IAS and Pixability." -- which are not independent at all. Google also doubled-down on their claim to not serve "personalized" or "behaviorally targeted" ads on "made for kids" YouTube channels.

Behaviorally targeted / “personalized” ads serving on “made for kids” YouTube channels even though Google said they didn't allow this.

Despite Google's claim to not serve "personalized" or "behaviorally targeted" ads on "made for kids" YouTube channels 4 separate experimental campaigns conducted by different parties corroborated that personalized and behaviorally targeted ads are still being served on "made for kids" channels as of this week.

Quoting from one of the 4 experimental campaign results -- the placement report revealed that around 50% of the impressions and views generated by this behaviorally targeted ad campaign were apparently served on “made for kids” channels such as "miraculous ladybug" and “stacyplays”.

Another juicy detail that was surfaced in this research was the fact that many demographic attributes had large portions of "unknown." See the data table above as just one example -- "parent, not a parent, and unknown." Note 100% of the ads in this case were served to "unknown." This raises many questions about the accuracy of the media buys done on large platforms like Google. If an advertiser doesn't specify a particular demographic attribute, is it OK for the platforms to just serve ads to users which they have no data on or have data that says "unknown"? In this particular third case, we are left with a conundrum: 1) is Google/YouTube doing "personalized" or "behaviorally targeted" ads on made for kids content? (which contradicts their own recent claims and written policies), or 2) if they are not using cookies or identifiers for targeted ads, are they serving ads to "unknown" users (which contradicts the fact that advertisers thought they were buying audience targeted segments.)

To re-iterate: "either Google is harvesting data and serving personalized ads on "made for kids" videos or they are serving targeted ads to users for whom they have ZERO data on." Either way, they are contradicting something -- 1) their own claims, or 2) advertisers that think they are buying audience targeted ads. Either way, these are failures that Google still has to address. We are waiting for Google's third blog post. We don't anticipate anything of substance. We do expect more misdirection and talking out of the side of their mouths.