Engagement Metrics

What it is

The origin is the most useful fact about the concept, and James Williams supplies it. Analytics platforms built to measure advertising established clicks, impressions and time on site as the default operational metrics for websites themselves. That extended the design logic of advertising, and specifically of attention-oriented advertising as opposed to advertising that serves a user's existing intentions, to the design of the whole user experience. So a firm grades itself on a number that fails to describe value to the user because the number was built to describe something else entirely.

The mechanism by which that number then governs everything is the part worth keeping. Tristan Harris, describing Google's internal practice from the inside, says that whenever he proposed products that interrupted people less he was told it conflicted with the bottom line, and that a change improving a user's attention shows up on a dashboard two to four weeks later as reduced time on site and gets rolled back. That second claim explains why individual good intentions inside such a company do not survive: the measurement loop is shorter than the argument, and the number wins by default.

It remains one person's recollection of internal practice, and this publication grades it that way.

In effect

The figures usually attached to this account come from Harris's 2013 presentation at Google, which has circulated publicly since 2016 and can be checked. Its wording is narrower than the paraphrases that travel. The deck says teenagers aged thirteen to seventeen send four thousand texts a month, once every six minutes awake; that the team shapes more than eleven billion interruptions to people's lives every day; and that it sets the notification standards on more than half the world's mobile phones, which describes Android standards rather than control of half of all notifications. A twenty-minute figure often quoted for the cost of clicking a photograph is a rhetorical question on a slide, not a measurement. None of these carries a source in the deck either. They are an employee's advocacy figures and this publication does not cite them as data.

The independence problem matters more than any individual number. Harris and Aza Raskin later founded an advocacy organisation on this thesis and were the central figures of a widely watched documentary. Williams is a third voice from the same circle, declaring an ongoing alliance with the same community in his acknowledgements and calling Harris a friend in the text. That is not an accusation of dishonesty. It is a statement about how much corroboration an account containing several named sources actually holds, which is less than the count suggests. Williams is useful because his contribution is conceptual rather than statistical, and he is not independent confirmation of Harris.

Williams also supplies the answer to the usual defence, which is that people freely chose to stay. Borrowing from Harry Frankfurt, he points out that a system can be perfectly accurate about what a person wants and wholly indifferent to what they want to want, with no contradiction between the two. His design consequence is that a product should serve a user's intentions, meaning their reflectively endorsed goals, rather than their moment-to-moment pull.

What it does not say

It does not say that measuring attention is dishonest. It says the metric measures something the organisation can sell rather than something the user values, and that the two diverge.

It does not say Tristan Harris invented the practice. He did not, and he does not claim to. He is a witness to one company's version of it.

It does not come with checkable internal data. The dashboards, the rollback and the bottom-line objection are all recollection, and nothing in the deck or in any of these accounts has a source behind it.

It does not have independent corroboration in the popular literature. The best-known accounts come from three former colleagues who went on to campaign together, and they should be counted as one source rather than three.

And it does not establish harm. What the concept explains is why a product develops in a particular direction. Whether that direction damages the people using it is a separate question with its own contested evidence.

---