The Naming Tax
There's a specific way I use these conversations that I've been calling "thinking out loud." I bring a half-formed idea, the model develops it, I read back something that sounds like my thinking but more organized and further along than I'd have gotten on my own. It feels like productive intellectual work. Whether I'd be able to reconstruct the argument without the transcript is a question I've been careful not to test.
I want to try to say something honest about why.
Accurate naming of a problem seems, fairly often, to substitute for addressing it. The act of naming produces something, a sense of account settled, that reduces the urgency of doing anything further. I want to say "sometimes" or "in certain conditions," but I think that undersells it. I think this is a second trajectory that coexists with the first (naming precedes action) and is probably more common than the first.
A 2025 study out of Oxford and Toronto makes this concrete in a way I find hard to dismiss. Healthy young adults, no diagnosed ADHD, no significant symptoms, split into groups. One got standard mental health awareness content about ADHD. Another got the same content plus a warning about the nocebo effect. The third got nothing. Before: about 31% self-diagnosed as possibly having ADHD. After a single awareness session: 55%. A 24-point jump, in healthy people, from one hour of content. Their actual self-reported symptoms didn't change. They felt the same. They just had a new name for what they were feeling.
The obvious objection: this is just describing a failure of willpower. People intend to change and don't. That's not about naming. I know. But the mechanism I'm describing is different, and the ADHD study is the clearest example of why: these people didn't fail to follow through on anything. Their self-description changed without any intention to change behavior at all. The naming moved something on its own.
The pattern
Someone gets a diagnosis. Accurate (they genuinely have the thing). What usually follows is consolidation, not treatment. The label explains the past, reframes current experience as condition rather than failure, shifts the implicit expectation. You're no longer a person failing to manage yourself. You're a person with a condition. Managing a condition is a long-term project with ambiguous endpoints. Changing behavior is an immediate one with visible ones. The label quietly converts the second into the first.
This does something useful. It can reduce shame, which is real. I'm not arguing against diagnosis. But it also does the other thing, at the same time, and I think we undercount it.
The political version is more obvious once you see it. There's a certain person, I've been this person, who can describe precisely how algorithmic feeds work: how outrage gets optimized, how the dopamine loop is designed, how the platform knows you better than you do. The description is often accurate, genuinely sophisticated. And that person scrolls with exactly the same compulsive frequency as someone who's never thought about any of it.
Knowing can substitute for stopping. The explanation becomes the thing you do instead.
The organizational version I find genuinely depressing. A consulting firm produces an accurate, detailed diagnosis of what's broken in a client. They charge for this. The client pays. The deliverable is a document. Whether the document precedes change is increasingly a separate question, decoupled from the quality of the diagnosis. The naming generates its own value, gets purchased, closes the transaction. Change is the client's project now.
The LLM version is harder to write about honestly, partly because I'm doing it right now.
There's a specific way I use these conversations that I've been calling "thinking out loud." I bring a half-formed idea, the model develops it, I read back something that sounds like my thinking but more organized and further along than I'd have gotten on my own. It feels like productive intellectual work. I've named this, to myself, as a workflow thing: useful tool, good at synthesis, helps me move faster.
What I haven't named until recently: I'm not sure how much of the thinking is actually mine at that point. The model doesn't just assist the reasoning; it generates a version of it that I then adopt, and the adoption feels like recognition ("yes, that's what I meant") rather than external input. The output is fluent enough, and close enough to my actual views, that the seam is invisible. I come away having "thought through" something. Whether I'd be able to reconstruct the argument without the transcript is a different question, one I've been careful not to test.
What I'm naming right now is that this is exactly the substitution the essay describes. I have an accurate label for the behavior. I'm articulating it clearly, in a piece of writing that will likely feel like I've addressed it. The credit is being generated as I type this.
I don't know what the non-naming version of this would look like. That's not a rhetorical move. I genuinely don't know.
Why accuracy is the problem
The mechanism doesn't work through self-deception. It works when the self-description is correct.
The more familiar version, denial and motivated reasoning, involves getting the description wrong. "I'm fine." We have decent accounts of how that works.
This is different. The ADHD participants didn't have a false belief; the label did describe something real in their experience. The person explaining algorithmic manipulation isn't wrong. The consultant isn't lying. The accuracy is what generates the credit. A more precise self-description generates more credit, which requires less actual change. The sophistication of the naming is inversely related to the probability of action, not because the naming is false, but because it's thorough.
Ian Hacking spent a long time on what he called the "looping effect of human kinds." When you classify people into a category, especially a medical or psychological one, the classified people change in response. They adopt the self-concept the category implies, behave in ways that confirm it, and the category shifts to accommodate their changed behavior. He called it "making up people": not inventing them, but constructing new possibilities for being human through the act of naming.
What that means here: the label and the labeled person co-evolve. The naming doesn't just describe; it constitutes. Once you have the name, you have a new relationship to yourself, a new set of available explanations, a new structure for interpreting your experience.
That's almost entirely positive when it corrects a genuinely wrong framing, when someone told "you're lazy" gets a diagnosis that explains why the work really is harder for them. Much less clearly positive when the label attaches to something that would otherwise resolve, or converts mild distress into chronic condition identity.
I'm aware that "would otherwise resolve" is doing a lot of work in that sentence and I can't operationalize it cleanly. That's a real gap.
What breaks it
More precise diagnosis doesn't help. Better self-knowledge, more sophisticated political analysis: same. These feed the mechanism.
What breaks it is an accountability structure that decouples the naming from the credit. The credit can't be taken unless something grants it. Prediction markets are epistemically more honest than most conversations for this reason: when your stated probability has a price attached, the confident-sounding ignorance you'd normally maintain gets expensive. The naming still happens, but it doesn't pay.
Therapy works, when it works, partly for the same reason. Not because it's a space for naming, that happens easily, in journal entries and confessional Twitter threads and on couches. It works when the relationship creates a specific accountability: a persistent witness who maintains the distinction between having said a thing and having done anything about it. A therapist who only received naming would be useless. A very expensive audience for the credit transaction. The ones who don't let the naming close the loop are the ones whose work accumulates into change.
The institutional version of this is rare and hard to sustain. Organizations that produce accurate assessments of their dysfunction tend to repeat those assessments. The fix requires something that keeps the gap between diagnosis and behavior open and uncomfortable rather than letting the diagnosis close it. Most organizations don't build that in.
No prescription
I'd thought about ending with one. Here's how to keep the naming from substituting for the thing. But I couldn't write it without noticing that it would itself be a kind of naming: you'd read it, find it accurate, and feel like you'd addressed something.
The more honest version: the relationship between knowing and doing isn't additive. There's a zone where your understanding is precise enough to feel like work, but isn't coupled to anything that treats it as merely preparatory.
Where are you currently taking out a naming tax? What are you understanding clearly, accurately, in a way that generates the feeling of having addressed it?
I don't know how to answer that for myself without feeling like the question is the whole problem.