MOW submits a critique on the testing of Topics API; Mozilla and Apple raise further privacy concerns

This week MOW submitted a response to Google’s testing of Topics and its presentation of the results, which artificially inflate the performance of the API.

In brief, the results of Google’s testing show that Topics are less effective than current technologies, and that smaller publishers are likely to be disproportionately affected. In its Topics explainer, Google notes that Contextual Data, which small publishers have less of, is key to effectively leveraging the API and inferring user interests.

This is, however, of minor significance compared to the fact that:

1. Google’s use of third-party cookies in the test group. This entirely negates the purpose of the test which is to examine Google products in a world where third-party cookies have been blocked
2. Google fails to define the volume of Google’s test group. For an industry which deals in delivering hundreds of thousands if not millions of impressions per campaign, it is no use establishing that the cost per impression to an advertiser is $1 dollar if that figure cannot reproduced at scale.
3. Google’s use of misleading definitions. Google, for instance, uses a non-traditional measure of conversion rate, reporting on conversions per click rather than conversions per impression. The latter an industry standard allows for a holistic measure of performance, accounting, for instance, for those who might search for a product following the delivery of an ad. This can only be viewed as an act of misdirection.
4. Google does not show whether budget shifted within Google’s ad buying platforms from behavioural to search channels. If you say budget shifted by a single percent from Google purchased Display ads, it remains unclear whether the supposedly unaffected 99% retained the same proportional split between these two channels, or whether more budget shifted internally to Search ads.
5. Google’s use of Mantel-Haenszel test to frame its results, very possibly obscuring the extent to which Topics performs in comparison to current technologies.

In summary, the issues raised in our submissions queried whether Topics’s would ever be able to replace the functionality of current technologies as required by the Privacy Sandbox Commitments.

Besides functional limitations, Topics also poses a glaring privacy problem in that it can be quite easily used to identify people online. Google having failed to find a technical solution to the issue, have announced an honour policy instead, whereby “developers enrol to use the API and attest that they won’t abuse the API”.

Mozilla and Apple have, however, signalled a lack of support for the proposal, with a Mozilla engineer commenting “fundamentally, we just can’t see a way to make this work from a privacy standpoint”.

So as it stands, Topics offers a less effective technology and adds nothing by way of privacy enhancements. In fact, it does quite the opposite. Currently, we have efficient solutions, in which abuses of user privacy are restricted by data protection law;  Google is proposing a structurally unsound API with regards to both privacy and technical adequacy, regulated by a flimsy honour code.