Could we have verified Topics?

[AramZS]

I've been thinking about potential solutions to the very likely case of what I previously referred to as Shadow Topics, the idea that other systems will layer on their own ML to build Topic prediction and lookalike models and instead of selling Topics instead sell what is essentially an alternative layer of what looks like Topics but is not.

I don't think that this is a problem exactly. If handled properly then buyers can make decisions against such models and take their chances around less accurately set terms. However, we know that fraud is particular rampant in the ad tech system and it seems very likely that Topics, if mainstreamed, will immediately be accompanied by middlemen adding Topics fraudulently to the ecosystem or using modeling to generate Shadow Topics and treat them similarly to Topics and causing significant confusion and loss of trust in the Topics system.

If buyers have no clear way to understand if a Topics value they are bidding on was generated at the level of the browser or the middleman the ecosystem could act unpredictably. Topics (as I've previously discussed) is likely to be highly trusted by buyers, their presence will be of high value. Topics could then be shared among site and ad tech system. But if the ad tech systems are incentivized to create fraudulent or lookalike Topics then the incentive to keep Topics to themselves as a real or fantasy secret sauce code product is increased. The market for Topics could blow up and then fall apart as declining trust in Topics zeros out value.

Is there a way we could better secure Topics as verifiable by buy-side systems? Some way that a buy-side system could verify that the Topics they are seeing is indeed a Topic generated by the browser? Or, if a Topic is shared with a publisher that the publisher could run that verification?

[dmarti]

In some cases, first-party Topics (generated by the site based on first-party interest data) or third-party Topics could be more desirable to buyers than browser topics.

A Topic verification system would be most useful if also available to be used by non-browser Topics providers.

[michaelkleber]

Would after-the-fact verification be good enough? That is, after you win an ad auction (where the bid request asserted that the user had Topic T), your creative had a way to say "Hey browser, plmk if Topic T was actually a lie"?

I feel like "some middleman lied about targeting signals" is such a pervasive problem that worrying about it for Topics might be slightly ridiculous when it's happening for far more fundamental signals like hostname! But as with hostname, some after-the-fact verification mechanism seems like it should be feasible to provide, and then it's up to the marketplace to push people to actually use it.

[AramZS]

I think this is a case where after-the-fact verification would definitely work!

[AramZS]

A Topic verification system would be most useful if also available to be used by non-browser Topics providers.

I think this is a different use case entirely no @dmarti ?

[dmarti]

@AramZS The Topics received straight from the browser might not be the best available. A site might have a user profile with much better ones, call them first-party Topics. A buyer who needs to check that some intermediary hasn't added an extra Topic would probably want to check both browser Topics and first-party Topics.