On this page
- Who built it and what it counts
- Products that replace the therapist
- The other findings
- What the figures support
- Questions to ask a vendor
- Frequently Asked Questions
- Do mental health products that replace therapists fail more often?
- Who compiled the 542-company dataset?
- Why did Woebot shut down?
- Does raising a lot of money protect a mental health startup?
- Is Citt designed to replace therapists?
A 2026 dataset of 542 digital mental health companies that left the market shows that products built to treat people without a clinician shut down more often than any other kind. It cannot show that keeping a clinician involved is what makes a company last. This post sets out what the dataset contains, who built it, and how far its figures go for a therapist deciding which software to trust.
In 2021 Pear Therapeutics went public through a SPAC deal worth $1.6 billion. Its substance use program, reSET, had in 2017 become the first digital therapeutic to receive FDA clearance. In April 2023 Pear filed for Chapter 11 bankruptcy, and in May its assets were sold at auction to four buyers for just over $6 million. Its founder, Corey McCann, wrote that the company had shown clinicians would prescribe its products and people would use them, and that this was not enough because payers could deny payment.1
Pear is one row in a dataset of 542 digital mental health companies that left the market between 2000 and 2026, published in July 2026 and revised in August.2
Who built it and what it counts
The dataset was compiled by Pavel Trubetskov, founder of Mentalium, a voice CBT diary that clients fill in between sessions and export for their therapist. He writes that he built it before committing years to a product in this field, to understand why so many companies in it had failed. He has a stake in the answer. His own coding places Mentalium in "augmentation", the category of products that support existing therapy, and the headline finding favors products with a clinician involved. He is open about this, and he reports one result that cuts against his own team: having a medical co-founder made no difference to outcomes.
The author started from 754 organizations with a recorded exit and removed brick-and-mortar providers and companies outside mental health, which left 542. Of those, 242 (45%) were acquired, 201 (37%) died, and 99 (18%) pivoted or were absorbed in consolidation. "Died" is the report's term for companies that ran out of money (93), found no product-market fit (79), were stopped by a regulator (14), were outcompeted (9) or ended in lawsuits (6). So most of the "542 dead startups" in this post's web address did not shut down. They were sold, merged or changed course.
Facts such as funding, dates, country and outcome come from Crunchbase, CB Insights, Tracxn, public deadpool databases, app store removals, domain data and trade press, with each company confirmed against at least two sources. Labels such as product role were assigned by language-model agents against a fixed taxonomy, which the author calls the weakest part of the method. The rows and the written reasoning for every label are published under a CC BY 4.0 license, so anyone can check them.3
The author lists the limits himself. Every company in the sample has already left the market, so the shares compare groups with each other and are not probabilities of failure. Calm, Headspace and BetterHelp, for example, are not in it. About 67% of the companies are from the US or the UK. Funding is known for 322 of the 542. He treats any group of fewer than 25 companies as showing a direction and not a precise value.
Products that replace the therapist
The report sorts each product into one of five roles. Clinical replacement treats a condition with no human clinician involved: autonomous chatbots, prescription software, passive monitoring presented as treatment. In care delivery a clinician treats through the product, as in teletherapy. Augmentation supports therapy already under way, for example with between-session work.
| Product role | Companies | Died | Acquired |
|---|---|---|---|
| Clinical replacement | 59 | 31 (53%) | 23 (39%) |
| Wellness self-help | 176 | 86 (49%) | 58 (33%) |
| Augmentation | 56 | 20 (36%) | 29 (52%) |
| Care delivery | 140 | 38 (27%) | 75 (54%) |
| Infrastructure | 111 | 26 (23%) | 57 (51%) |
Replacement products were sold to consumers more often than the other roles, and consumer-paid products died more often, so the author compared roles within each type of payer. Among consumer-paid companies, 56% of replacement products (20 of 36) died, against 43% of care delivery products (30 of 69). Among institution-paid companies, the figures were 50% (11 of 22) and 9% (6 of 69). Within each role, care delivery companies paid by an institution died far less often than consumer-paid ones (9% against 43%), while for replacement companies the two shares were close (50% against 56%). Both comparisons rest on the institution-paid replacement group of 22 companies, which is below the author's 25-company threshold, so they show a direction.
Of the 59 replacement companies, 13 died for lack of product-market fit and 5 because of a regulator. Of their 23 acquisitions, 83% had no disclosed price. The disclosed ones included Akili at $34 million, after a SPAC listing that valued it near $1 billion, and myStrength at $30 million.
Woebot is the best-known recent case. STAT reported on July 2, 2025 that Woebot Health had shut down its app that week. Roughly 1.5 million people had used it, and its responses were pre-scripted and did not come from a large language model.
Founder Alison Darcy told STAT the closure was largely due to the cost and challenge of meeting FDA requirements for marketing authorization, made more pressing by large language models that the company wanted to use and the FDA had not yet worked out how to regulate.4 The report records that Woebot had raised $124 million and that free general-purpose models had begun to do much the same thing. The clinical trial evidence on chatbots is a separate question, covered in what the trials of AI mental health chatbots show.
The other findings
The report has seven findings in all. These are the remaining figures, as published.
| Cut | Group | Companies | Died |
|---|---|---|---|
| Who pays | Consumer | 258 | 137 (53%) |
| Institution (all) | 254 | 53 (21%) | |
| Employer | 83 | 17 (20%) | |
| Health plan | 50 | 6 (12%) | |
| Clinic | 80 | 25 (31%) | |
| Hospital | 18 | 2 (11%) | |
| School or university | 23 | 3 (13%) | |
| Business model | B2C | 210 | 112 (53%) |
| B2B | 214 | 52 (24%) | |
| B2B2C | 57 | 5 (9%) | |
| Revenue model | One-time purchase | 26 | 22 (85%) |
| Freemium | 53 | 33 (62%) | |
| Subscription | 174 | 37 (21%) | |
| Per employee per month | 63 | 9 (14%) | |
| Funding raised (disclosed) | Under $1M | 114 | 79 (69%) |
| $1M to $10M | 77 | 16 (21%) | |
| $10M to $50M | 80 | 15 (19%) | |
| $50M to $100M | 11 | 0 (0%) | |
| $100M or more | 40 | 10 (25%) | |
| Clinical evidence | Published data | 128 | 43 (34%) |
| None | 306 | 134 (44%) | |
| Medical co-founder | Yes | 156 | 59 (38%) |
| No | 254 | 82 (32%) |
The author treats the payer as the strongest signal in the dataset. On funding, the $1 million to $10 million group had the highest share of acquisitions (70%), and 17 of the 40 companies that raised $100 million or more were acquired. On evidence, he reads SilverCloud Health (bought by Amwell in 2021) and MindBeacon (acquired in 2022) as bought largely for their clinical credibility. On medical co-founders, the acquisition rate was 47% in both groups. Evidence status could not be established for 108 companies, which sit outside that comparison.
What the figures support
Among digital mental health companies that left the market, those built to treat without a clinician had the highest share of deaths of any role. For wellness and care delivery products, the share that died was more than 25 points lower among institution-paid companies than among consumer-paid ones; for replacement products the gap was 6 points. There are four reasons to hold even that loosely: the sample contains no surviving companies, the role labels came from a language model, several groups are small, and the shares describe association with no account of cause.
The figures also do not show that keeping a clinician involved is what makes a company last. Augmentation, the group closest to between-session tools, died at 36%, close to the sample average of 37%. The author's example of where his own category breaks is Therachat, a between-session homework and mood-tracking tool that grew past 7,500 therapists, raised under $1 million, and was sold in 2020. He attributes its trouble to therapists rarely paying for software. He also calls the line between care delivery and augmentation the blurriest part of his coding.
By the report's definitions, Citt would fall in the augmentation group. The therapist sets up each client's between-session support and chooses its approach, and session notes stay drafts until the therapist approves them. The Trust page sets out where our evidence stops.
Questions to ask a vendor
The dataset suggests four things to check before you build your practice around a tool, alongside the clinical questions in how to know if mental health AI is safe:
- Who pays the vendor, and on what model. In this sample, consumer-paid, one-time-purchase and freemium products died most often.
- Whether the product claims to treat on its own or supports treatment you deliver.
- Whether the vendor has published evidence of its own, as distinct from testimonials.
- What happens to your clients' records if the company closes or is sold. Will your practice software still exist in five years? covers that question in more depth.
Footnotes
-
Park, A. (19 May 2023). Pear pulped: Digital therapeutics maker's assets sold for $6M at auction after bankruptcy filing. Fierce Biotech. Reports the $1.6 billion SPAC deal, the 2017 reSET clearance, the April 2023 Chapter 11 filing, the auction to four bidders, and Corey McCann's statement. ↩
-
Trubetskov, P. (2026). Mental health startup graveyard v1.2: 542 dead companies, seven patterns. Mentalium. Published 29 July 2026, updated 10 August 2026; version 1.2 changes the text but not the figures. All dataset figures in this post come from this report. The author founded Mentalium, a between-session CBT diary that his coding places in the augmentation category. The figures are descriptive shares within a sample of companies that had already exited and are not failure probabilities. ↩
-
Trubetskov, P. (2026). Dataset: 542 dead mental health startups, 2000-2026, coded on 18 fields. Mentalium. Open data under CC BY 4.0, with the reasoning behind each label. States that product-role labels were assigned by language-model agents. ↩
-
STAT. (2 July 2025). Why Woebot, a pioneering therapy chatbot, shut down. Most of the article is behind a paywall; this post uses only the opening section, which is public. ↩
Frequently Asked Questions
Do mental health products that replace therapists fail more often?
In a 2026 dataset of 542 digital mental health companies that left the market, 53% of products built to treat without a clinician died, against 27% of products through which a clinician treats. Because the dataset holds only companies that exited, these figures compare groups within it and do not give a probability of failure.
Who compiled the 542-company dataset?
Pavel Trubetskov, founder of Mentalium, a between-session CBT diary, published it in July 2026 with the underlying rows under an open license. His product falls in the augmentation category his own coding describes, which is worth keeping in mind when reading findings about clinician involvement.
Why did Woebot shut down?
Founder Alison Darcy told STAT in July 2025 that the closure was largely due to the cost and challenge of meeting FDA requirements for marketing authorization. Large language models, which the company wanted to use and the FDA had not yet worked out how to regulate, added to the pressure.
Does raising a lot of money protect a mental health startup?
This dataset cannot answer that, because it contains only companies that left the market and no survivors to compare with. Within it, 10 of the 40 companies that raised $100 million or more died (25%), a higher share than among companies that raised $1 million to $10 million (21%) or $10 million to $50 million (19%).
Is Citt designed to replace therapists?
No. The therapist sets up and directs each client's between-session support, and Citt's session notes stay drafts until the therapist approves them.
Stop rewriting notes from scratch
Citt.ai drafts session notes under your review, keeps between-session context in one place, and stays therapist-led. Start free. No card required.
Related Articles
- What the Trials of AI Mental Health Chatbots Show, and What They Leave Open
The published trials of AI mental health chatbots, including Therabot, Limbic Care and Ash: what each compared, what it found and what is still open.
- How Do You Know If a Mental Health AI Is Safe?
What mental health AI safety evals can and cannot show, the questions to ask a vendor, and how Citt.ai answers those questions itself.
- What Gets Lost Between Therapy Sessions, and What a Therapist Gains From Seeing It
What clients forget between sessions, why practice between sessions matters, and what a therapist gains from seeing more of a client's week.