2025 hindsight: Quite pleased with spotting, in 2016, and before my time at Google, that search keywords needed clustering into higher order intents, and that selling those intents rather than individual words was where the search product would need to go. Mildly amusing that even in the present days of Performance Max (son of UAC) we still have search ’experts’ complaining about the death of the keyword. Google presumably wants to get rid of them all - better to have the machines choosing the words than people. (Although ‘better for whom?’ is a fun question…)

Similar people, similar things

A lot of the online data we’re working with these days depends on making links between things that are similar. Similar products, similar people, similar needs.

Amazon says ‘this book is like these other books because the same people browse or buy them’. Facebook says ‘these people are like these other people because they share the same interests’. Google says ‘these searches are like these other searches because people found the same results page helpful’.

One important practical difference though, is in how resemblance is managed by the platform.

Facebook clusters related signals

In Facebook land, the grouping of likes and pages into interests, and their ascription to people, is handled by Facebook. They have put algorithms (and some sensible questions that they ask when you’re setting up a Page) to work, to enable you to target people on the basis of their interest in say, charitable giving, without having to worry whether someone’s ‘like’ of the Kitty Heaven page is actually endorsement of a massage parlour rather than a cat charity.

Furthermore, this is done across countries using the same list of headings. I can target people on their Facebook interest in ‘Charity and causes’ - which is real by the way - and I don’t need to worry how to define charities in Brazil versus Spain.

Google leaves it to the user

In Google’s case however, for search advertising, management of resemblance is done manually by the advertiser. Search marketers have to decide word-by-word what counts as important vocabulary for the brands and campaigns in question, so they can bid on those words. And if I want to cover lots of countries, I need to manage large lists of words for each language in turn.

This is not very scalable, and adds a lot of complexity to managing search compared to managing Facebook campaigns, but at the same time affords a lot of control. In some ways there is a parallel with how Google approaches Android - giving the user the ability to mess with things, even to the point where they mess them up.

By contrast, Facebook is much more Apple-esque here, giving buyers of media on its properties a higher-order set of constructs to work with, but excluding users from fine-grained control.

Intent ≠ Keyword

In the world of user-facing technology Google’s stance is understandable, but one wonders how long this will carry on in a search context. After all, agencies make their money on search by managing lists of words and transacting them on the Google’s platform. A lot of legwork goes into managing these lists especially across languages and country borders, but with the ever present drive for growth, and the ever shrinking scope of places to get it from, it wouldn’t surprise me if Google had ambitions to expand along the search value chain.

What if, for example, they had an automated, higher-order grouping of words around key brands and topics. Rather than making me buy all the words separately {cat charity, feline foundation, cat protection league, caridad del gato} I could just buy {cat charity} and Google would just take care of it?

They already see all the words all the advertisers buy, as well as, of course, all the related things that people search for and find as a result, and for a machine learning company, one can’t imagine it would be all that difficult to produce a Facebook-like set of unified buying primitives that worked across countries and langauges.

Arguably they have already done it. In some ways the Knowledge Vault that Google wrote about in this paper in 2014 is exactly this sort of high level “structured repository of knowledge that is language independent”.

To the best of my knowledge it isn’t possible to do search marketing against this construct, but one wonders how long that will remain the case.