Ad tech research group Adalytics has today released new evidence which it claims strongly suggests that YouTube has been running personalised ads on made-for-kids videos, despite pledging not to do so. Today’s release comes as a follow up to the report Adalytics issued last week which claimed that YouTube potentially leaked data from millions of child users, while also running personalised ad campaigns on made-for-kids content.
In a new blog post released today, Adalytics runs through analysis of four reports from different media buyers. In each case, data sent back to the buyer by Google seems to suggest that all impressions for each campaign were served on made-for-kids channels. But the reports also provide data outlining audience segments which were used to target the campaign – suggesting that these ads were personalised.
What does the new evidence say?
Adalytics’ report last week made a number of separate but related claims about how YouTube handles child users. While YouTube says it doesn’t run personalised ads on made-for-kids content, Adalytics claimed it does. Alongside this, the report claimed that YouTube was seen dropping advertising cookies for viewers watching made-for-kids content, and created tracking IDs for users who had clicked on ads on made-for-kids content. And it also made claims that YouTube’s tools were ineffective at enabling advertisers to avoid made-for-kids content.
Google promptly responded, denying the central claims from the report. A statement from the company described the report as “deeply flawed and misleading”, and reiterated that “personalised advertising has never been allowed on YouTube Kids, and in January 2020 we expanded this to anyone watching ‘made for kids’ content on YouTube”. Google also said that “the portions of this report that were shared with us didn’t identify a single example of these policies being violated”.
Hence today’s response from Adalytics.
The four examples outlined by Adalytics are all fairly similar. In each case, the buyer set up an ad campaign which was configured to target some of YouTube’s audience segments. In each case, the buyer also set up an inclusion list (which states specific channels which the buyer wants their ads to run on) which included several hundred made-for-kids YouTube channels. The buyer also did not select or enable audience expansion.
In each case, Adalytics says that Google’s reports show that 100 percent of impressions and views were served on made for kids channels (a lot of the specific numbers have been redacted on screenshots which were included on Adalytics blog post, but in some instances it’s at least clear that the majority of views came from made-for-kids channels).
But audience reporting provided by Google suggests that placements were targeted using Google audience segments. For example, one of the campaigns targeted affinity segments including ‘High-End Computer Aficionados’. The corresponding audience report shows that 71.0 percent of views and 76.5 percent of impressions came from users who fit into this affinity segment.
It’s hard to square this circle. If all ads in question were served on made-for-kids channels, and Google never uses personal data to target ads, then it’s very difficult to explain the data presented by Adalytics.
A slam dunk?
Adalytics itself doesn’t seem to claim that this proves beyond a doubt that YouTube has been serving personalised ads on made-for-kids content. Rather, it says that these results “raise questions about the YouTube CEO’s statement that no personalised ads would be served on made-for-kids YouTube videos”.
The evidence however (assuming it’s accurate) doesn’t look good. Several media buyers cited by Adalytics say that the data proves Adalytics’ original claims – one source described the evidence as “highly conclusive”.
There may be other explanations – but these won’t necessarily reflect well on YouTube either.
One out for Google would be to say that the audience reporting shows the extent to which a campaign hit a buyer’s targeted segments, even if that data wasn’t used to target the campaign. It would in principle be possible for YouTube to use non-personalised targeting parameters for campaign targeting (thereby fulfilling its own pledge), but then still report back on audience segments using the data it has available. It would however be extremely surprising that in the example outline above, YouTube was able to deliver 76.5 percent of impressions to high-end computer aficionados, despite not using that as a targeting parameter,
Rob Webster, global vice president at marketing consultancy CvE, said it could be a case of inaccuracy in Google’s booking settings and reporting. It could be possible that ads weren’t behaviourally targeted, but that an error in booking and reporting made it seem like they were. He added that “every effort should be made to rule out a setup error”.
Webster added that YouTube could surface some small print explanation of what constitutes behaviourally targeted advertising, arguing that this small print applies differently to made-for-kids supply. But if this was the case, Webster said that it would appear that “the tech vendor is being deliberately misleading in an effort to maximise returns”.
So it’s not case closed by any means. But as one buyer put it, any response from Google “should be similarly backed by data and platform examples”.
