Why TikTok's algorithm keeps recommending you a mental health diagnosis

I don't get surprised by much any more. Cancellation cycles, manufactured outrage, the speed at which someone's entire character gets reduced to a fifteen second clip, I've made a career out of not blinking at any of it. But there's something I’ve seen recently that has unsettled me, and it isn't a client's scandal. It's what happens before someone ever becomes my client. It's what's sitting in their pocket at 11pm, quietly rewriting how they understand their own mind.

I watched two videos about a mental health condition on TikTok. Two. Not a deep dive, not hours of doomscrolling, just two videos out of ordinary curiosity. It was instant. Less than 5 minutes, easily 80% of my FYP was that one disorder, on repeat. There was a point where I flicked through ten videos in a row and every single one of them was about it. I'll be honest, it was fucked up. Not just the volume of it, but the comment sections underneath. Comment after comment: "wow, oh my god, I finally realised what's wrong with me." "I've never related to anything so much in my life." "It's OCD I have." "It's autism I have." Hundreds of strangers diagnosing themselves in real time under a video made by someone they'd never met, based on a feeling that could describe almost anyone on a bad week. And then, right there in the replies, the account that posted it: "I'm so glad you feel seen, click the link." It's fucked up. That's not concern, and it isn't a public service. It's a sales funnel wearing the costume of validation.

My feed wasn't a fluke either. Independent audits of TikTok's recommendation system have tracked exactly this pattern. One recent study found accounts that engaged with mental health content saw it grow to roughly 45% of their entire feed within a week, regardless of what they'd actually searched for, while other reports have put the figure as high as 60 to 70%. Mine felt like eighty. Whatever the precise number, the direction is always the same: watch a couple of videos and the platform stops giving you a feed, it gives you a diagnosis loop. Even accounts that actively tried to avoid the topic still couldn't get exposure below roughly 11 to 13%. You cannot fully opt out.

That's not a glitch. It's the system working exactly as intended.

The business model is the symptom

Engagement optimised platforms don't reward accuracy, they reward relatability, and relatability is cheapest when it's vague. "Do you ever feel like everyone can see through you?" "Do you replay conversations in your head for hours?" "Do you feel exhausted by small talk?" These aren't diagnostic criteria. They're the emotional equivalent of a horoscope, broad enough that almost anyone watching will nod, which is exactly why they perform so well. A feeling everyone has had, repackaged as a symptom only you apparently have, is very good content. It's also, quite often, nonsense.

The research backs this up with uncomfortable precision. A large scale audit of over 5,000 posts across platforms found TikTok was the worst offender for mental health misinformation of any major platform, with 52% of ADHD videos and 41% of autism videos containing inaccurate information, against roughly 22% on YouTube and under 15% on Facebook (UEA / EurekAlert, 2026). The same study found the misinformation wasn't spread evenly. It clustered almost entirely around who was talking. Videos from actual healthcare professionals were inaccurate about 3% of the time. Videos from creators with no clinical background were wrong 55% of the time. That's not a subtle gap. It's the difference between a diagnosis and a personality quiz wearing a diagnosis's clothes.

The second group reaches vastly more people, because the algorithm was never built to sort for credentials. It was built to sort for watch time.

Real conditions, real consequences

I want to be careful here, because I think both the "it's all overblown" crowd and the "everything on TikTok is fake" crowd are wrong in ways that matter. Autism, ADHD, OCD and dissociative disorders are real, serious, and historically under-diagnosed, particularly in women and adults who were failed by a system that only recognised these conditions in a narrow, often male, presentation. Some of the visibility social media has given these conditions has genuinely helped people recognise themselves for the first time, in videos made by people who do have lived experience and do cite real research. That's worth saying plainly, because nuance tends to be the first casualty of this kind of conversation.

But visibility and diagnosis aren't the same thing, and the platforms have every commercial incentive to blur that line for as long as possible.

The clearest evidence of actual harm, not just misinformation but people's minds and bodies changing in response to content, comes from the sudden, well documented rise in functional tic like behaviours among teenagers, heavily correlated with consumption of "TikTok Tourette's" content. Several peer reviewed studies from 2021 onwards describe clinics reporting sharp increases in tic presentations that don't match typical Tourette's patterns but do closely mirror the specific tics shown in popular videos, something researchers have described as a mass social media induced illness (PubMed, 2021; Frontiers in Psychiatry, 2022). This isn't people lying. Functional neurological symptoms are genuinely experienced. It's a well established psychological phenomenon being triggered and shaped by algorithmic exposure at population scale. If that doesn't tell you the feed is doing something to people, I'm not sure what would.

The gender problem makes this even more dangerous

There's a layer to all this that makes the self-diagnosis problem worse specifically for women and girls, and it starts long before TikTok existed. The clinical picture of autism, ADHD and OCD that most people carry around in their heads, including a fair few clinicians, was built almost entirely by studying boys and men. Autism's core diagnostic tools were validated on predominantly male samples, and a systematic review found the diagnostic gap runs at roughly four boys diagnosed for every girl overall, widening to ten to one once intellectual disability is removed from the picture. Autistic girls and women tend to show more of what researchers describe as cognitive and behavioural differences rather than the obvious social-interaction difficulties the diagnostic criteria were originally built around, and a lot of them learn to mask those differences so well that clinicians miss them entirely, sometimes for decades. Higher-IQ women are, remarkably, less likely to be diagnosed even when their underlying symptoms are just as severe as a diagnosed man's.

ADHD follows a similar pattern. Boys tend to be hyperactive and impulsive in ways a teacher notices within about thirty seconds. Girls tend to be inattentive, forgetful and quietly overwhelmed, easily mistaken for anxious or "a bit scattered", which is exactly why women were historically diagnosed roughly ten times less often than men, and why the average woman with ADHD is often diagnosed a decade or more after her symptoms started, frequently after years of being treated for anxiety or depression instead. OCD has its own gendered split: men more commonly report harm-related, sexual or symmetry-focused obsessions, women more commonly report contamination and cleaning-focused ones, and postpartum OCD affects women at roughly twice the rate it affects men.

Now put that history next to an algorithm that has no concept of any of it. It doesn't know your history, your hormones, your masking, or which version of a condition is statistically more likely to apply to you. It knows what you paused on. If the two videos that hooked you happened to describe the loudest, most commonly filmed version of a condition, usually whichever presentation performs best on camera, you can end up either convinced you have something you don't, because you related to a stereotype rather than your actual symptoms, or quietly reassured you're fine when what you were shown was never going to look like your version in the first place. A woman scrolling contamination-focused OCD content might conclude she doesn't have OCD at all, because nobody showed her the harm-based intrusive thoughts that would have made her go "oh". A woman scrolling hyperactive-ADHD content might rule herself out entirely, because her ADHD has never once looked hyperactive. The algorithm isn't correcting for any of that bias, it's amplifying whatever version of the story already gets the most watch time, which is very often the version the research says we've been under-diagnosing for decades in the first place.

This is exactly why a proper assessment matters so much, and why it can't be replaced by a video, however accurate that video happens to be for the person who made it. A good clinician is trained to ask about the version of a condition that doesn't perform well on camera, the quiet, internal, masked, atypical version, precisely because that's the version most likely to have been missed by every system before it, including the algorithm. If you take one thing from this piece, let it be that: go and see someone properly qualified, in person or genuinely credentialed online, and let them ask the boring, thorough, individual questions that no fifteen second video ever will.

Self-diagnosis is not the same thing as diagnosis

I'm not a mental health expert, and I would never hand out advice on anyone's mental health. I'm simply not qualified to, and I take that seriously. What I am is someone who works alongside therapists and clinicians every week because of the clients I represent, and who has put a fair amount of my own time into reading the research behind this piece. So I'd happily bet a fair amount of money that I'm still, technically, more qualified to talk about this than a lot of the accounts that many thousands of young people are watching every single day. That's not a boast. It's genuinely the problem.

Here's the part that tends to get lost when people argue the fix is simply "get a qualified therapist on there instead": even a properly qualified therapist cannot diagnose you from a video. That isn't caution on their part, it's because diagnosis doesn't actually work that way. A real diagnosis relies on a clinical interview, a full personal history, and enough time to rule out the other conditions that can produce almost identical symptoms. A piece in the journal Psychological Medicine points to a specialist autism clinic where only around half of the people referred with a suspected diagnosis actually met the criteria on proper assessment. The rest had something else entirely, often OCD, social phobia or depression, conditions that can look similar on the surface but need a completely different course of treatment (Psychological Medicine, Cambridge). Fatigue and low mood can look identical whether the cause is depression, a vitamin deficiency, a sleep disorder or plain burnout. Taking ADHD medication while an underlying bipolar disorder goes unrecognised can trigger a manic episode rather than help anything at all (Cleveland Clinic). None of that is visible from a symptom checklist in a sixty second video, however well-meaning the person making it is.

It's worth saying too that the therapists I work with regularly, on behalf of clients, aren't exactly cheering the influx of qualified professionals joining TikTok either. More than one has said to me directly that it's a genuinely difficult position for a properly trained therapist to be on these platforms at all, because so little of what they could responsibly post actually fits into a short form clip. Done properly, it looks like general psychoeducation: what OCD is as a category, roughly how a diagnosis actually gets made, why it takes more than a checklist, when it's worth seeing a GP. What a responsible account can't ethically do is anything that resembles diagnosing an individual, because they haven't assessed the person watching and never will. The moment a qualified account starts doing "does this sound like you" content instead of that, the qualification stops being the safeguard everyone assumes it is.

This is exactly why mental health has to be treated individually rather than by category. It has to be personal to be accurate. The same word, OCD, ADHD or autism, can sit on top of ten completely different lived experiences, ten different histories and ten different appropriate treatment plans. That's what therapy and proper clinical assessment actually exist to do, not to confirm the label someone already suspected, but to work out what's genuinely true for that specific person, which a comments section can never do.

The self-diagnosis that follows isn't a harmless shortcut either. It can quietly stop people getting the help they actually need, because once someone believes they already know what's wrong with them, they often stop looking any further, and whatever is actually going on underneath, or alongside it, goes unaddressed. It can also start to run someone's life from the inside, reshaping how they explain their own behaviour, their relationships and their choices around a label that was never properly assessed in the first place. That's a heavier thing to carry than most of these videos ever let on.

When "unqualified" becomes a business

Then there's the layer above the videos themselves: the people monetising the audience the algorithm builds for them.

City AM recently reported on the near total absence of regulation in this space. In the UK, "therapist" and "counsellor" aren't protected titles, which means anyone can use them regardless of training. Their reporting found that of 500 mental health related TikTok videos analysed, 91% came from creators with no medical qualifications at all (City AM, 2026). The piece is worth reading for its case study alone: a woman named Gabriella who spent thousands of pounds on OCD "recovery" coaching from Robert Bray, an unaccredited provider who built his following on TikTok, before experiencing poor therapeutic boundaries and a relapse. She isn't an isolated case either, similar concerns have been raised independently in OCD support communities discussing his services. I'd encourage anyone curious to read that reporting directly rather than take my word for it, I'm not in a position to judge individual practitioners, and I don't think that's really the point. The point is the ecosystem that makes someone like this possible: a job title with no legal meaning, an algorithm that rewards confident delivery over credentials, and an audience primed by months of "does this sound like you?" content to trust whoever says yes to them next.

Only around 12% of the British public actually know that anyone can legally call themselves a counsellor (City AM, 2026). That gap between what people assume is regulated and what actually is regulated is where the guides, the £200 "recovery frameworks", the branded workbooks and the paid group chats all live. It's a genuinely clever business model, if you can stomach calling it that: build the audience by convincing them they have the problem, then sell them the solution.

Why this is a crisis PR problem, not just a mental health one

I spend my working life inside this exact mechanism, just aimed at a different target. When a client of mine is being cancelled, the algorithm doesn't show them a balanced picture of what's actually happening. It shows them the pile-on, on repeat, because that's what they've just engaged with, whether by refreshing, reading the replies, or simply existing while the story trends. Their entire feed becomes them, the accusation, the pile-on, again and again, and it wrecks their mental health in a way that has nothing to do with whether any of it was even fair. That's not a side effect. It's the same recommendation engine doing precisely what it's built to do: take whatever you've just shown interest in, however briefly or involuntarily, and give you more of it until it's the only thing left in the room.

Now picture that same mechanism pointed at someone who has done nothing wrong at all, who simply feels a bit different lately and clicks on one video wondering, out of curiosity, whether it might be autism. Within days they're not looking at a feed any more, they're drowned in it. Video after video, then comment sections stacked with strangers saying "yeah, I feel the same" over and over, in a tone that stops feeling like community and starts feeling like something closer to a cult. That repetition isn't informing anyone. It's manufacturing consensus out of thin air, the same way a pile-on manufactures certainty about a person nobody in the mob has actually met. A person doesn't get accused of something by someone else here, they get quietly convinced of something about themselves, at 2am, by an algorithm with no clinical training and no accountability for being wrong. There's no retraction. No correction ever reaches as far as the original video did. And unlike a reputational crisis, where at least the person living through it knows something is happening to them, this kind of harm is often invisible from the inside. It just feels like self-knowledge.

I've sat with clients whose genuine, diagnosed conditions were dismissed by people around them as "just a TikTok thing", because the platform has saturated the conversation so thoroughly that real conditions now compete with content-farmed imitations for credibility. That's the quieter tragedy in all this. The flood of low quality content doesn't just mislead the people watching it, it corrodes trust in the people who actually have these conditions and are simply trying to be believed.

What accountability would actually require

The comparison I keep coming back to is the Molly Russell case. When a UK coroner concluded in 2022 that algorithmically pushed content had contributed to a 14 year old's death, the platforms' defence wasn't that the content didn't exist, it was that they hadn't intended for any single user to see quite so much of it. The coroner didn't accept that as an excuse (Open Rights Group, 2022). Her father put it more plainly than any regulator has: the platforms were "monetising misery". I don't think mental health misinformation is a lesser version of that same story. I think it's the same story, running at a lower simmer, at a much greater scale, mostly because it doesn't produce one tragedy dramatic enough to force a public reckoning, just a slow, diffuse erosion across millions of feeds that never quite makes the news.

Real accountability here isn't complicated to describe, even if it's politically difficult to enforce. Platforms would need to stop treating "engagement with mental health content" as one undifferentiated signal to chase, and start distinguishing between content from qualified sources and content from someone performing symptoms for reach. They would need friction on health claims in the feed, the way certain misinformation categories already get. And they would need to stop letting an unregulated job title do the work that actual verification should be doing. None of that is technically difficult. It's commercially inconvenient, which in my experience is a much harder problem to solve.

Concretely, that means legislation, not just community guidelines a platform can quietly rewrite on a Tuesday. If an account is going to publish anything that could lead someone to self-diagnose a mental health condition, autism, ADHD, OCD, whatever it is, it should have to meet roughly the same bar the NHS website holds itself to: written or checked by someone actually qualified to say it, and accountable to a real professional body if they get it wrong. Anything short of that shouldn't be eligible for algorithmic promotion at all. Not banned outright, people are entitled to talk about their own experiences, but not boosted, recommended, or allowed to functionally operate as a diagnostic tool for thousands of strangers a day. Right now there's no such bar. Anyone can post it, and the platform will hand it the exact same reach as a genuine clinician, sometimes more, because outrage and relatability outperform accuracy every time. That's the part that's actually irresponsible, not that people are talking about mental health online, but that platforms treat a stranger's guess and a clinician's assessment as interchangeable content, and let the algorithm pick the winner based on watch time.

This is also, for what it's worth, something I'd like to be more than an opinion piece about. I'd happily work with charities in this space if it would help, and I've already started reaching out to my own contacts at TikTok, Meta and Google directly, trying to get some of the websites attached to these accounts, the ones selling the guides, workbooks and "recovery frameworks", taken down.

Until then, my advice is the same as what I'd tell any client mid-crisis: don't take the internet's word for who you are. Not the people accusing you of something you didn't do, and not the algorithm quietly suggesting you might have something you don't. Both are optimised for your attention, not your accuracy. If a video makes you wonder whether you might have a genuine, diagnosable condition, that instinct is worth taking seriously, by taking it to an actual clinician, not a comment section.

If any of this has struck a nerve

Please don't self-diagnose from a video, and please don't sit with it alone. Speak to an actual GP or clinician first. In the meantime, these are legitimate, accountable places to start.

UK
Samaritans: 116 123, 24/7, free to call, samaritans.org
Mind: 0300 123 3393, Mon to Fri 9am to 6pm, mind.org.uk
Shout: text SHOUT to 85258, 24/7 crisis text service, giveusashout.org
OCD Action: 0300 636 5478, ocdaction.org.uk
OCD UK: ocduk.org
National Autistic Society: autism.org.uk
ADHD UK: adhduk.co.uk
YoungMinds (for under 25s and parents): youngminds.org.uk

USA
988 Suicide and Crisis Lifeline: call or text 988, 24/7, 988lifeline.org
Crisis Text Line: text HOME to 741741, 24/7
NAMI HelpLine: 1-800-950-6264, Mon to Fri 10am to 10pm ET, nami.org
International OCD Foundation: (617) 973-5801, iocdf.org
CHADD (ADHD support): 1-866-200-8098, Mon to Fri 1pm to 5pm ET, chadd.org
Autism Society: 1-800-328-8476, autismsociety.org