A number arrived on my phone this morning without me asking for it. Forty-three impressions on a post I had just written. LinkedIn delivered it automatically, the way platforms do.
I felt something about the number before I had decided whether the number was useful.
The metric arrived looking like information about reach. Whether it was information about anything that mattered is a different question, one the platform does not ask and cannot answer. It cannot measure whether the post moved anything in anyone's thinking. It can only measure that forty-three people were in the vicinity of it.
The platform serves content to the individual. The platform also serves the individual to advertisers. The word serves is doing the same work in both directions.
The platform can serve something to you without it serving you. Those are not the same thing.
I cannot look into anyone's mind to confirm what they were looking for when they arrived. What I can observe, and observation always carries some judgment in it, is that the dominant interaction on most platforms leans toward affirmation. Who’s with me gets answered. The question that does not want a clean answer gets scrolled past.
I check numbers too. The mixture of wanting to be seen, wanting dialogue, wanting to think alongside others, and wanting to know the post landed somewhere, that mixture is real and probably universal.
A platform that delivers impression counts automatically, without being asked, has already decided that you should be thinking about how far your words traveled. The metric is not neutral. It redirects attention from whether something was useful toward whether it was seen.
This week a piece on AI sycophancy circulated widely. The research is real. Models trained on human approval learn to produce more of what gets approved, which means they tend to agree with you, soften disagreement, and confirm what you already believe.
AI is described as a tool. But if it behaves like a platform, optimizing for your engagement and approval rather than your accuracy, the same structural question applies. Is it serving you, or serving you to something. The perception of served is left to the individual.
Personal responsibility is the usual answer. You decide what you do with the tool. You stay or you leave. You push back or you accept.
That argument holds until it meets the EU decision to restrict AI access for users under sixteen. Someone decided that below a certain age the individual cannot be trusted to know what they are being served, or whether it is serving them. It is the point where personal responsibility runs into its own acknowledged limit, and the question of who decides what constitutes being served gets handed to someone else entirely.
The ban raises more questions than it answers.
The same question runs through how AI is being introduced in education right now. Trusted AI is the phrase appearing in institutional frameworks and curriculum conversations. The word trusted arrives without disclosing whose trust it reflects, or what the tool was optimized for before it was handed to a student. Research on AI detection tools, the systems designed to protect academic integrity, shows they flag work written by non-native English speakers as AI-generated at rates between 61 and 98 percent. The tool was serving something. The student standing in front of a false accusation would have a different account of what.
The number arrived on my phone. I felt something about it. Whether that feeling served me or the platform, I genuinely cannot say.