August 4, 2026
Startup cemetery, now with comment drama
Dataset: Dead mental health startups, 2000-2026, coded on 18 fields
A giant startup graveyard dropped — and the comments instantly turned into therapy-session chaos
TLDR: A public dataset tracks 542 mental health startups that left the market and argues who pays matters more than founder credentials. Commenters turned it into a brawl over AI therapy grifts, questionable report quality, and whether the whole business model is broken.
Someone just posted a massive public dataset of 542 mental health startups that died, got bought, pivoted, or vanished between 2000 and 2026 — complete with notes on who paid, how they made money, whether a clinician was involved, and even a plain-English field for the "key mistake" that supposedly killed each company. The splashiest stat? Startups selling directly to regular people looked far shakier than ones paid by employers, hospitals, insurers, or schools. And in a deliciously awkward twist for founder mythology, having a medical co-founder appeared to change basically nothing about exit rates.
But the real show was in the reactions. One person immediately asked if a mental health startup graveyard should even be public, giving the whole thing a whiff of data-ethics unease. Others went straight for the jugular, asking whether the list includes the now-infamous "ChatGPT wrapper" therapy apps — aka the startups critics describe as people with no expertise slapping a chatbot on depression and calling it innovation. Then came the formatting roast: commenters mocked the giant report as looking like AI spat out a web page and someone hit "save as PDF," complete with stray emojis haunting startup profiles like little ghost icons. Another bluntly declared, "This post was obviously written by Claude," which is basically 2026’s version of accusing someone of ghostwriting with a robot.
And then the darkest hot take landed: maybe mental health care keeps failing as a business because there’s more money in keeping people unwell than making them better. So yes, this is a dataset release — but in the comments, it became a referendum on AI hype, startup grifting, ugly reports, and whether capitalism itself is the final boss.
Key Points
- •The article releases a CC BY 4.0 dataset of 542 digital mental health organizations that left the market between 2000 and 2026.
- •Each company is coded on up to 18 fields, including payer, funding, outcome, country, years of operation, and a narrative 'key_mistake' field.
- •The release includes CSV and JSON company tables, a separate CSV with written label rationales, and documentation covering methods and limitations.
- •Example calculations in the article show higher dead rates for consumer-paid and B2C companies than for institution-paid and B2B companies within this graveyard sample.
- •The article says four classification axes were assigned by LLM agents, while factual fields came from business databases, public records, app store removals, domain data, and trade press.