Digital health · Youth mental health
Koko
A nonprofit that embeds brief mental health support inside the social platforms young people already use, routing those who signal distress toward tested self guided courses. Its research record is unusually deep for its size. So is the public argument about how it runs tests.
Debugging the brain
Rob Morris arrived at the MIT Media Lab with a Princeton psychology degree and no engineering background, and it showed. He dropped most of his first year classes. The depression he had carried for years got worse. He spent his nights teaching himself to code, much of that time on Stack Overflow, the site where programmers fix each other’s bugs for free.
“I struggled with depression on and off for much of my life, but my early time at MIT was especially difficult,” he told The Verge in 2017.
The dissertation came from noticing what Stack Overflow actually was: a crowd of strangers, trained in nothing in particular, reliably producing useful help within minutes. He built the equivalent for unhelpful thinking, called it Panoply, and ran a randomized controlled trial. 166 people with symptoms of depression were assigned to Panoply or to an active control, an online expressive writing tool. The results ran in the Journal of Medical Internet Research in March 2015, with MIT advisor Rosalind Picard and Stephen Schueller as coauthors.
Both arms reduced depression scores. Panoply did better on cognitive reappraisal, the skill of re-reading a bad situation in a less punishing way, and better for people who started out worst off. It also held attention: Panoply users logged in about twice as often as controls.
The finding Morris kept returning to was not about receiving help. It was about giving it. A 2017 paper in Personality and Social Psychology Bulletin reported that people who engaged more by helping others showed greater decreases in depression.
Three founders, two companies
A week after Morris defended, Fraser Kelton emailed him. Kelton and Kareem Kouddous had been building the same thing by hand, recruiting people off Twitter into group chats and teaching them the basics of cognitive behavioral therapy. Within weeks the three had a wireframe.
Union Square Ventures, an investor in Twitter and Tumblr, put in $1 million alongside MIT’s Joi Ito in January 2015, per The Verge. In August 2016 VentureBeat reported a $2.5 million round led by Omidyar Network and Union Square Ventures. Morris’s own site also names Spero Ventures.
The consumer app did not become a business. Koko turned its crisis classifiers toward general content moderation, and in November 2018 Airbnb bought the company on undisclosed terms. Business Insider reported the deal more than a year later and described it as an acqui-hire. The founders joined Airbnb and kept the Koko name.
“It was depressing,” Morris told Fast Company in 2022. “I’d spent years of my life devoted to digital health, and we had this system that was providing a ton of impact, and we had to drop it.”
The second Koko
In late 2020 Morris and Kouddous restarted Koko as a nonprofit. It filed with the IRS as Koko AI Inc and went through Y Combinator in the Winter 2022 batch. The product changed shape. Rather than an app to download, Koko became a layer other platforms install: when a user searches or posts terms a classifier associates with distress, the platform offers a route to Koko and its short self guided courses. Koko names Tumblr, Pinterest and Discord among its partners.
The research followed. A 2022 study in JMIR Formative Research counted 6,179 completions of three brief courses embedded in Tumblr over a year. A 2023 randomized test in the Journal of Behavioral and Cognitive Therapy compared the standard platform response, a crisis line number alone, against a one minute enhanced version. Among 374 users, the share reporting they used the resources offered rose from 38.6 percent to 78.1 percent.
January 2023
On January 6, 2023, Morris posted a thread describing an experiment run the previous autumn in which Koko peer supporters could have GPT-3 draft their replies before sending them. About 4,000 people received messages at least partly written by the model, across roughly 30,000 messages. Morris wrote that the co-written messages were rated higher than messages written by people alone, that response times halved, and that Koko pulled the feature anyway.
The thread drew more than 8 million views and a severe reaction. Within a day Morris posted corrections: recipients had seen a note saying the reply was written in collaboration with Koko Bot, and the feature was described at onboarding. Business Insider, Vice, Ars Technica, New Scientist, NBC News and Gizmodo all covered it. The experiment had no institutional review board approval, which Morris said it did not require. Two months later Vice reported on a separate, already published Koko study of crisis resource uptake, which its authors recorded as non human subjects research after consulting the Stony Brook University review board. That determination, and the use of a privacy policy as the consent step, became the second front.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| Panoply RCT | JMIR, Mar 30, 2015. N=166. Depression improved in both arms; reappraisal improved only with Panoply. Registered as NCT02302248. | Public record |
| Crisis service uptake | J Consult Clin Psychol, Apr 2019, with Harvard’s Matthew Nock. A brief barrier reduction message was associated with a 23% increase in use of crisis services. NCT03633825. | Public record |
| Conversational agent | JMIR, Jun 26, 2018. Ratings from 37,169 users; 79.20% of agent responses were deemed acceptable, but users significantly preferred peers, an effect that held in a controlled study of 1,284 people. | Public record |
| Courses on Tumblr | JMIR Formative Research, Jul 25, 2022. 6,179 completions of three courses with pre to post improvements on hopelessness and self-hate. | Public record |
| Crisis resources, 2023 | J Behav Cogn Ther. 374 randomized users; uptake 38.6% to 78.1%. Funded by a Hopelab grant. Recorded as not human subjects research. | Public record |
| The GPT-3 feature | About 4,000 recipients and roughly 30,000 messages, per Morris. No effect sizes were published and Morris told NBC News he had no official data to share. No review board. | Company-stated |
| Reach and papers | Koko’s home page says 6M+ people in 199 countries and over eight peer reviewed publications. One of its own support landing pages says over 5 million. A December 2025 partner release says over nine publications. | Figures differ |
| Commercial ties | A September 2026 competing interest statement in Child and Adolescent Mental Health states that Koko has paid partnerships with OpenAI and Character.AI aimed at supporting users’ mental health. | Public record |
| Device review | No FDA 510(k) or PMA record, and no trials registered under this Koko. | Not found |
Read plainly: the routing and the brief courses have been tested in real settings and published, with coauthors at Harvard, Northwestern and Stony Brook. The 2022 GPT-3 feature is the one piece of work with no published data attached, and it is the piece everyone remembers. Koko’s own reach and publication figures drift between its pages.
What to watch
- Whether the third party review process Morris described to NBC News in 2023 is documented publicly. Koko now names an external ethics advisory board on its site.
- Whether the paid OpenAI and Character.AI partnerships appear as program service revenue in the FY2026 Form 990.
- Results from the Closegap school integration announced in December 2025.
In their words
“It was just this intuition that, just as we can get people on Stack Overflow to help us identify and fix bugs in code, perhaps we can harness a crowd to help us fix bugs in our thinking.”
Rob Morris, founder, MIT News, 2015 · Independent
“By helping others, they were helping themselves, and that seemed like a great synergy.”
Brad Burnham, Union Square Ventures, The Verge, 2017 · Independent
“the consensus in the academic community is there is great potential promise here, but the jury is still out”
Matthew Nock, professor of psychology, Harvard, The Verge, 2017 · Independent
“I’d spent years of my life devoted to digital health, and we had this system that was providing a ton of impact, and we had to drop it.”
Rob Morris, on the sale to Airbnb, Fast Company, 2022 · Independent
“This feature was opt-in. Everyone knew about the feature when it was live for a few days.”
Rob Morris, clarification of January 7, 2023, quoted by Business Insider · Company-stated
“No psychiatric or psychological group has verified its efficacy or laid out potential risks.”
Arthur Caplan, bioethics, NYU Grossman School of Medicine, Business Insider, 2023 · Independent
“It’s a cornerstone of ethical practices, but when you don’t have the requirement to do that, the public could be at risk.”
Camille Nebeker, UC San Diego, on informed consent, NBC News, 2023 · Independent
“No participant recruitment or data collection occurred at Stony Brook University, and the study/project design was determined by staff at Koko.”
Jessica Schleider, then of Stony Brook, coauthor, Vice, 2023 · Independent
Related companies
Sources
- Public recordEfficacy of a crowdsourced peer-to-peer cognitive reappraisal platform
- Public recordTowards an artificially empathic conversational agent
- Public recordRCT of an online risk assessment and intervention platform
- Public recordSingle-session interventions embedded within Tumblr
- Public recordImproving uptake of mental health crisis resources
- Public recordDebate: conversational AI and young people’s mental health
- Public recordIRS Form 990, Koko AI Inc, FY2021 to FY2025
- IndependentCrowdsourced tool for depression
- IndependentThe empathy layer
- IndependentKoko raises $2.5 million
- IndependentAirbnb quietly acquired Koko
- IndependentHe sold his mental-health startup to Airbnb
- IndependentMental health company uses ChatGPT in experiment
- IndependentAI chat used in experiment on real users
- IndependentVice investigation of the crisis uptake study
- CompanyKoko home page, about page and ethical commitments
- PartnerClosegap and Koko school partnership
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
