Refuting Every Lie, Misdirection, and Non-Answer in Google's Response to YouTube Scandal

Google wrote a blog post -- https://blog.google/products/ads-commerce/transparency-and-brand-safety-on-google-video-partners/ -- in response to the 202-page research post from Adalytics --https://adalytics.io/blog/invalid-google-video-partner-trueview-ads.

Below, I will refute every lie, mis-direction, and non-answer by Google. But first, some necessary context.

In-stream versus Out-stream video ads

Let's start by clarifying what we are talking about in this "scandal." The screen shots below show what out-stream video ads look like. Note the "Skip Ad" button, which most folks are familiar with in ads seen on YouTube. These are clearly YouTube ads. But in the cases below, the video ads were played by themselves (out-stream) separate from any YouTube video content. These ads did NOT run pre-roll, mid-roll, or post-roll to a YouTube content video -- THOSE would be called "in-stream." The video ads shown below are out-stream video ads played by themselves without any YouTube content.

The Adalytics report focused on the MISREPRESENTATION of out-stream ads as if they were TrueView in-stream ads. In the pie chart below, from a campaign analyzed by Adalytics, the dark blue slice shows that 16% of the impressions were valid TrueView ads that ran on YouTube and the light blue shows 11% of impressions were valid TrueView ads that ran on GVP (Google Video Partner) websites. Both of these kinds of ads were valid. If you add the two together, it's 27%; so about 1 in 4 ads were valid "TrueView, skippable, in-stream video ads." Running on YouTube versus GVP is not the main issue here.

The main issue found by this research is that the other 3 of the 4 ads were mis-represented as "TrueView, skippable, in-stream video ads" when they were not; they were out-stream ads that ran by themselves without any YouTube content, some were not skippable, muted, off-page, not viewable. Re-iterating that this is a different issue than whether the ads ran on YouTube itself or elsewhere on video partner websites. Further, I confirmed with the researcher Krzysztof Franaszek that properly embedded YouTube content videos that had TrueView ads running pre-roll, mid-roll, or post-roll were NOT considered invalid in this case.

A review of the Adalytics research post shows the word "in-stream" used 196 times and "out-stream" used 54 times - because this was the key point, about misrepresentation. A review of the Google response shows 0 mentions of "in-stream" and 0 mentions of "out-stream." They did NOT address the data presented by the research -- that 3 in 4 ads were out-stream ads misrepresented as in-stream TrueView ads.

Google's own documentation further states that

• Video ad placements must be audible by default.
• Scroll-to-play ads are not permitted for TrueView.
• For mid-rolls, the video content’s duration must be at least 10 minutes.”

See the 3rd screen shot above -- (muted, autoplay, not initiated by user, and in lower corner, often outside of the page content). Quoting from the Adalytics research:

"Many TrueView skippable “in-stream” ads that Adalytics reviewed were delivered on sites and apps in which the ads were rendered in a method that violates Google’s own definitions of in-stream. Specifically, ads were placed on pages with such characteristics as

• in small, out-stream video players in the corner or side of the consumer’s device viewport
• in a fully muted video player
• with little to no video content in between consecutive TrueView ads
• where the video ads auto-play without any viewer interaction or initiation
• the ads played continuously, on a loop"

Those are not TrueView ads, that advertisers paid for. None of these issues were addressed in the Google blog post.

Finally, Google's own documentation shows that IVT ("invalid traffic") explicitly includes "Misrepresentation of in-stream and out-stream video."

“Out-stream video invalid traffic is the misrepresentation and display of in-stream video in an out-stream format. For example, this happens when in-stream video ads are shown in a banner ad -- an advertiser is led to believe that their ad was displayed in-stream with other video content, when in fact it was displayed in a different format.” So by their OWN definition of IVT, 3 in 4 of the video ads Google sold as TrueView should be deemed "invalid" IVT. 

With this context, let's review each sentence in Google's blog post response.

Google: "The overwhelming majority of video ad campaigns serve on YouTube."

This is technically accurate, almost all video ad CAMPAIGNS serve some ads on YouTube. What is NOT addressed is the PROPORTION of ads within a campaign that ran on YouTube versus outside of YouTube. According to the research, in the observed campaign, only 16% of the ads served on YouTube, while 84% served outside of YouTube, on GVP ("Google Video Partners"). Of course, there could be other campaigns where the majority of ads showed on YouTube instead of GVP.

Google: "When advertisers create video ad campaigns, they can clearly see that their ads may run on third-party sites via GVP during the campaign setup."

As an advertiser, I have set up YouTube campaigns and you can see in the screen shots below that the option for "Where do you want your ad to appear?" is NOT present during campaign set up. It appears after the ad is created and users must go into each ad to verify whether this is set to "Only on YouTube" or "On YouTube and on Google video partner sites and apps." If anything, this should be considered a "dark pattern" because some advertisers may not know they have this option to select during campaign set up because the option is not present during campaign set up, contrary to what Google claimed.

Google: "Ad inventory across the Google Video Partner network is more than 90% viewable."

This is laughable. Clearly their viewability detection is broken. In every case documented by the research, 1) "82.5% was delivered against Google Video Partner websites which were playing TrueView ads in muted, out-stream, auto-playing, or hidden video slots" (hidden video ad slots are not viewable). 2) "77.7% was delivered against Google Video Partner websites which were playing TrueView ads in muted, out-stream, auto-playing, or hidden video slots" (hidden video ad slots are not viewable). 3) "As another example, one can observe TrueView ads for American Express being served, where the (muted, auto-playing, out-stream) TrueView ad is covered by another banner display ad that prevents the consumer from being able to click “skip” on American Express’s ad" (covered by another ad is not viewable).

Here's why Google's "90% viewable" number is self-reported and incorrect. In April 2022, DoubleVerify proudly sent out a press release proclaiming to be "the first to earn accreditation by the Media Rating Council (MRC) for its independent, third-party calculation and reporting of YouTube video viewability for desktop and mobile (web and app) using Google’s Ads Data Hub (ADH)." The operative terms are "calculation and reporting" and "using Google’s Ads Data Hub (ADH)." This means DV is performing calculations and reporting, using data supplied by YouTube. DV is NOT running any measurement pixel or javascript tag to actually measure viewability. So the use of the word "independent" is a lie; and the viewability they report after performing the calculations are no different than if YouTube reported it themselves -- i.e. "grading their own homework."

It is now obvious, after the hundreds of pages of detailed research with screen recordings and code samples from Adalytics, that measurement vendors like DV, IAS, and Moat are clearly failing to detect viewability issues like the example mentioned above -- "TrueView ad is covered by another banner display ad." Oh, right. They're not running any pixel or javascript tag to collect data. They are just performing calculations on data supplied by YouTube. What's even more disturbing is that MRC accredited them for "independent" YouTube viewability measurement, when no measurement actually took place. MRC reached out and clarified that the accreditation given to DV does not include measurement of ads outside of YouTube, for example on GVP.

Google: "We use real-time ad quality signals to determine if people are present and paying attention that help us decide whether to serve a video ad in a Google Video Partner site or app. Those signals include viewability, the size and position of the player and whether people are engaging with the ads."

More flowery BS words that make false claims. There is no HTML code or javascript code that can detect anything outside of the browser, let alone "if people are present and paying attention."

The viewability signal they use is clearly incorrect -- for example, when they failed to detect the TrueView ad was covered up by another display ad. The "size and position" of the ad are likely measured correctly, but did you know that the smaller the ad player, the higher the viewability, because all of it fits in the viewport (e.g. viewable portion of the browser window). So an ad stuffed in a 1x1 pixel or 0x0 pixel is 100% viewable because it fits entirely and easily in any viewport no matter how small. Note that third party vendors DV, IAS, and Moat are NOT actually measuring any ad on YouTube with a pixel or javascript tag. They are also not measuring ANY TrueView ad served on a GVP site with a pixel or javascript tag. They have no actual measured data; they should NOT represent video ads where they have NO measurement to be "viewable."

Do you now see how Google's claim of 90% viewability is entirely self-reported, unsubstantiated, and not independently measured by ANY of the third parties like DV, IAS, and Moat, even though those vendors also have MRC accreditation? Google could have just lied and said it was 100% viewable; but that would appear to be too implausible no one would believe it. No one should believe the 90% number either, for all the reasons documented by the research.

Google: "In addition to the high bar we set on YouTube, we have strict policies that all third-party publishers, including Google Video Partners, must follow."

When was the last time bad guys "followed" the rules. Having strict policies doesn't do ANYTHING to prevent bad guys from trying and succeeding in gaming the system. Years ago, we witnessed fake sites embedding dozens of YouTube videos on a single HTML page. There was nothing else on the page. They would then repeated load the page with bots. This generated enormous numbers of views for those videos and ads ran alongside these videos too. Years ago, the recommendation was to turn off ads in "embedded videos" to reduce this form of fraud. The same fraudsters are now even more clever. Why load the YouTube content video when you can just load the video ads and repeatedly load only the ads to maximize profit? This is exactly the same evolutionary sequence as in programmatic display and video ads outside of YouTube. Bad guys used to set up websites, and load dozens of ads on them. Then they dispensed with the websites and webpages and loaded just the ads to save time and bandwidth. Why load the entire page when you can just load the ads and make money? Bad guys have further reduced their operating costs by not even loading the ads -- just fake the bid requests and that was sufficient to make money fraudulently.

It will be interesting how many sites Google kicks out of the Video Partner network after being exposed like this.

Google: "We actively enforce these policies."

Apparently not "actively" enough, because the data from this research dates back to 2020. Over the course of at least 3 years, across more than 1,100 advertisers/brands' campaigns, and billions of impressions, TrueView ads that were NOT in-stream, audible, skippable, and viewable were transacted on crappy Google Video Partner sites. Running on GVP instead of YouTube was NOT the main point of the research. Out-stream ads mis-represented to be "TrueView skippable in-stream ads" is the main problem documented by the research. Google has not responded in any way to this.

Google: "Our policies prohibit engaging in disruptive, invasive or deceptive ad-serving practices along with other specific practices which some would consider indicate that they are made for advertising. This includes serving ads on hidden browsers, pages with more ads than content or ads that lead to accidental clicks. To give you a sense of how serious we are about this, in 2022 we stopped serving ads on more than 143,000 sites for violating our policies."

No one is disputing that Google HAS strict policies. Everyone now realizes that despite the strict policies, enforcement is lax or non-existent. Look at the list of advertisers affected; and look at the list of agencies that transacted these ads without knowing they were not TrueView ads but were reported as TrueView ads. How did those 143,000 get approved in the first place? And did any of the advertisers get refunds for ads that ran on those policy-violating sites?

How did sanctioned, side-loaded (not from Google Play) Iranian apps and Russian state sponsored propaganda sites get approved into GVP and run TrueView ads? Will these be kicked out of GVP now that they have been documented?

Google: "We offer third-party validation for ads running on Google Video Partners content. We partner with outside organizations to help us ensure publishers are complying with our policies. Google Video Partners supports independent, third party verification from DoubleVerify, Integral Ad Science, and Moat for viewability and invalid traffic."

As we already mentioned above NONE of these outside vendors are running any pixel or javascript tag, on YouTube or on GVP sites. Therefore they are not "measuring" anything. The press release from DoubleVerify clearly states they are only performing calculations (on data supplied to them by YouTube) and reporting on viewability based on those calculations. Nothing was directly measured. This should not be labeled "independent, third party verification" and the MRC accreditation of this is evidently a sham to cover up that NO measurement was actually done.

So What?

As I said at the top of this article, the Google blog post responding to the research basically said nothing. Worse, it contained lies, mis-direction, and non-answers ("in 2022 we stopped serving ads on more than 143,000 sites for violating our policies" has nothing to do with the detailed documentation of out-stream ads being mis-represented and sold as "TrueView skippable in-stream ads" which advertisers thought they were buying.) They are hoping that advertisers and buyers (media agencies) are not technical enough to cut through the crap and that this will blow over by the next news cycle. The 1,100 advertisers that spent billions on YouTube advertising over at least the last 3 years should be pissed. They should also take action and collectively demand refunds on at least 3 in 4 ads that did not meet Google's own policies and definitions for "TrueView skippable in-stream ads." Google's own documentation labels these as IVT (mis-represented in-stream and out-stream video ads). At the very least Google should refund the 3 in 4 that they themselves would define as IVT.

The research findings were shown to media buyers for peer review, to get their feedback, and to understand their perceptions and expectations when buying TrueView ads. Here are some excerpts quoting directly from the research:

“In your own words, how would you feel if Youtube was running large numbers of your ad campaign's YouTube TrueView in-stream video ads on: - 3rd party websites or apps, - with small, out-stream video players, - with sound off by default, - that were auto-playing with user involvement?”

The media buyers responded:

• “I would demand a refund of all impression and third party costs associated with buying those ads. I would also demand that youtube/google block buying of these placements by default.”
• “Cheated by Youtube. Not confident in future Youtube buys without plans to improve ad placements.”
• “I would feel like I was ripped off.”
• “I would feel like the inventory was unreliable, and not worth paying for.”
• “Cheated out of my money.”
• “would feel screwed”
• “Fraud”
• “That inventory ran in a way that was not in line with how it was purchased.”
• “Pissed”
• “Would feel like its not a transparent investment”
• “Owed a massive credit”
• “I would feel cheated. I would feel like i wasn't getting what i paid for”
• “Betrayed and lied to”

Furthermore, a large majority of media buyers reported they would not buy TrueView skippable in-stream ads if they knew that a fair number of those video ads would serve in out-stream, muted, auto-play video players."

“If you were sold "TrueView skippable in-stream" video ad inventory that actually ran on 3rd party websites as muted, auto-play, out-stream video, would you consider that to be "ad fraud" ?"

84% of surveyed media buyers responded “Yes” - suggesting that they would consider mis-declared TrueView skippable in-stream inventory to be “ad fraud”.

When asked to explain their reason in free form text responses, some of the media buyers responded:

• “This is ad fraud because it is not playing in the way the product is described. It is not in stream - i.e. - it's not playing before/during/after content. Outstream, autoplay inventory would be worth much less than true instream video.”
• “It doesn't match the description of what is being sold.”
• “It is documented to be something other than what was received”
• “outstream ads are different then youtube trueview ads. They appear in sidebar, corners, or within editorial content.”
• “Per Google's definition it is required for TrueView Ads to be audible. If they are muted by default - this is a conflict of the understanding of how TrueView operates.”
• “its outstream not instream as purchased”
• “There is little chance the ad will actually have an opportunity to be effective”
• “If its muted, its not true-view”
• “The videos were clearly out-stream placements. They appear on sites with primarily written content (ie NOT video) and disconnected from the purpose of the websites.”

The screen shot of the Google post below was taken 8:37a ET, June 29, 2023, in case Google edits the post to claim they never said what they said. https://blog.google/products/ads-commerce/transparency-and-brand-safety-on-google-video-partners/