The Algorithm Eats Diagnosis

There's a genre now. You've seen it. A face fills the frame, the captions bounce along the bottom, and the voice says something like: five signs you might have ADHD. Or: things I didn't know were autism. Or the compound form, the one I see most lately: AuDHD, autism plus ADHD, delivered as a single identity with its own aesthetic, its own in-jokes, its own merch. Thirty to ninety seconds. Confident eye contact. A symptom list that lands like a horoscope, which is to say it lands on everyone.

This genre bothers me, and I've spent a while trying to take the bother apart honestly, because some of the obvious objections to it are objections I don't get to make. I share my mental health experience publicly, in detail, on the internet; this entire website is testimony. I can't be against people talking about their minds. And I'm genuinely not against people finding each other through a label; I just wrote an essay arguing that the communities who forked their diagnoses and maintain their own meaning have the healthiest possible relationship to the manual. The clustering isn't the problem. Something happens after the clustering, and the algorithm is what makes it happen.

My credentials, again, sort of

Here's the personal stake, and it's an embarrassing one.

For three years I explained Bipolar I from the inside. Sincerely, publicly, in essays I still stand behind. I described what it felt like, what it meant, what others should understand about it. Then in 2019 the label changed underneath me: schizoaffective disorder, bipolar type. The professionals had been watching me for years, with charts and credentials and every incentive to get it right, and the first answer still needed revising. I didn't change that day. The diagnosis did.

So when I watch a twenty-something narrate their disorder to two million strangers with total fluency, I'm not sneering from above. I recognize the feeling from inside. The lived experience is absolutely real, and the certainty feels earned, because nothing on earth feels more authoritative than your own interior. But I had the same certainty and the wrong noun. Diagnosis is hard. It's hard for clinicians with a decade of your history in front of them. The format hands out the certainty for free and skips the part where it can be wrong.

That's the first thing the genre launders: one person's experience of a label that may or may not be correctly attached, presented in the grammar of fact. Not "my ADHD looks like this" but "this is what ADHD looks like." The slide from witness to expert takes about one sentence of caption text, and the viewer can't see it happen.

What the algorithm selects for

None of this would matter much if the content were ranked by accuracy. It's ranked by watch-through.

The algorithm cannot see whether a symptom list is diagnostic. It can see whether you finished the video, and you finish videos that feel like they're about you. So the selection pressure runs directly against precision: the symptoms that survive the edit are the relatable ones, which means the universal ones, which means the least diagnostic ones. Everyone interrupts people. Everyone loses keys, rehearses conversations, gets tired in a way naps don't fix. A video listing the experiences that would actually distinguish a disorder from a hard week is specific, unflattering, and boring, and the feed buries it.Psychologists call this the Barnum effect: statements vague enough to feel personally, uncannily true to nearly anyone. It's the engine of astrology, cold reading, and personality quizzes. Now it has a recommendation system behind it, running A/B tests at planetary scale to find the phrasings that feel most like you.

The same pressure strips the hedges. Watch what survives: "I" becomes "we," "my experience" becomes "people with ADHD," "sometimes" disappears entirely. Qualifiers read as weakness, and weakness loses the retention graph. But the hedge was the medicine. The hedge was the entire difference between testimony and diagnosis. When I write about my own treatment, the most important words are the boring ones, the for me, the your body is different, and those are precisely the words the format cannot afford. Sixty seconds selects for the version of an illness optimized to be recognized by the largest possible audience. That is the exact opposite of what a diagnosis is for, which is to distinguish.

And the creator usually isn't lying. That's what I keep having to remind myself. Sincerity is orthogonal to all of this. The algorithm doesn't reward her because she's right or wrong; it rewards her because she's confident and relatable, and it would reward me identically for my wrong label of 2017, narrated with conviction in good lighting.

Where it turns

So far this is a critique of information quality, and if that were all of it, it wouldn't really bother me. Wrong information about minds is older than the manual. Here's the part that actually gets me, the part I find, and I've tried to find a fairer word and failed, gross.

The label gets celebrated.

Not the person. Not the survival. The diagnosis itself, worn the way people wear a star sign. The condition becomes an aesthetic: pastel carousel slides, that's-so-AuDHD-of-me, symptoms reframed as quirks, the disorder as a personality with fan content. Somewhere in the last few years the destigmatization project quietly overshot. The opposite of shame was supposed to be neutrality, a diagnosis as unremarkable as a blood type. We sailed past neutrality into pride, and pride needs content, and the algorithm was right there.

I want to be careful, because the people doing this are mostly not cynical. If a label finally explains thirty years of unexplained friction, euphoria is a reasonable first response, and posting through your euphoria is just what this era does. But watch what the celebration does structurally, because it does two things, and they're both expensive.

First: a celebrated label can't update. The genre turns a diagnosis into a niche, the niche into an audience, the audience into income and identity at once. Now run my 2019 experience through that machine. A re-diagnosis stops being a re-index and becomes a rebrand, with followers to lose. I wrote in Breaking Changes that building your identity on the manual is building on someone else's API; the creator economy pours concrete on it. The committee can at least revise its categories. A brand can't. And the viewer inherits the same trap at smaller scale: once the label is your bio, your community, your content diet, and your sense of finally making sense, every incentive you have points away from the possibility that it's the wrong label. The first question of diagnosis, is this actually what's happening to me, becomes a threat to the whole stack built on top of it.

Second, and this is the one I feel in my body: the celebration is selective, and the selection is brutal. Only the marketable disorders get the treatment. There is no schizoaffective aesthetic. Nobody's doing pastel carousels about the angel I watched descend, and "five signs you might be psychotic" does not trend as a celebration; when my side of the manual shows up in the feed at all, it's horror content, true-crime adjacent, the thing the quirky labels get contrasted against.There's a word for what the feed is running: a marketability sort on the DSM. Anxiety, ADHD, autism-as-superpower on one side; schizophrenia, psychosis, personality disorders on the other. The sort criterion isn't severity or prevalence. It's whether the symptom list can be made cute. The algorithm took the manual and sorted it into adorable and frightening, and the rebranding of the adorable half makes the frightening half lonelier. Every video that renders a diagnosis as a personality makes it a little harder for the person whose diagnosis is not a personality but a fight.

And the words themselves lose currency. When ADHD circulates as a synonym for quirky and distractible, the person who cannot hold a job because of it walks into rooms where the word has already been spent. The label was supposed to be load-bearing, a claim on accommodation and care. Celebration inflates it like any other currency, and the people who need its full purchasing power are the ones who can least afford the devaluation.

What's actually worth celebrating

I keep asking myself what the healthy version looks like, since I've ruled out the easy answer of nobody talks about their mind online, on the grounds that it would delete my website.

I think it looks like the distinction the format can't hold. Celebrate the person; the diagnosis is a map of where it hurts. Celebrate finding your people; the thing you have in common is a wound, and you can love your people without making the wound the flag. Celebrate survival, loudly, mine included. And keep the grammar of witness, the grammar this genre strips first: my autism, my mania, my experience, one data point, offered to anyone it helps, with the version number showing and the hedges left in.

Maybe some of this is me being precious about words that cost me something to carry. I've sat with that possibility and I don't think it's the whole of it, because the pattern here is the same one running through this entire series: the algorithm finds whatever humans use to hold something sacred, and renders it as content, because engagement is the only value it can see. It ate language, love, reality, time. Of course it eats diagnosis. Suffering that explains you is the most relatable content there is.

A diagnosis is a hard-won, provisional, revisable name for a way a human being suffers. The feed turned it into a zodiac. The difference between those two things is everything the word was for.