AI mental health tools after Woebot: what to look for next

What happened to Woebot

Woebot Health retired the Woebot app on 30 June 2025. Its own FAQ says so in one line: new accounts can no longer be created, and previous accounts can no longer be accessed. The same page says all account data was anonymized as of 31 July 2025.

The company told users by email on 28 April 2025, according to HLTH's report, which adds that users could download their conversation history until 30 June. That is roughly nine weeks of notice for a product STAT says about 1.5 million people had used.

On why, the most fully sourced account is STAT's July 2025 report. It quotes founder Alison Darcy attributing the shutdown largely to the cost and challenge of meeting the FDA's marketing-authorization requirements, and says the company wanted to use large language models, which the FDA had not yet worked out how to regulate. The article is behind a paywall; we read only the part that was open. The company's homepage now describes products for providers and payers, so the work continues through organisations, not through an app store.

What happened to Earkick and Youper

The source for both is the Common Sense Media Youth AI Safety Institute's risk assessment of AI mental health apps (page updated 5 May 2026; press release dated 18 May 2026). It says both apps "vanished from the App Store and/or Play Store" during the evaluation, without notice to users, without referrals to alternative care and without public explanation. It dates Earkick's App Store removal to April 2026 and Youper's disappearance to after 25 April 2026. It reports more than 3 million users between them. The assessment is about teen safety, so neither app received a score.

What we confirmed ourselves on 30 October 2026:

Two things follow. A report that an app left the stores is not the same as a company announcing it is closing. And the details of the three exits differ: one with about nine weeks of notice and an export window, one that the assessment describes as sudden, one that is unclear.

What the closures say about the category

Woebot's exit is the one with a stated reason, and the reason is regulatory cost. A product that makes clinical claims carries the expense of proving them to a regulator. A product that avoids those claims avoids the expense, and also the accountability that comes with it. Neither route says whether an app helped anyone. What a closure does show is how much of a person's inner life sat on a company's balance sheet. The Common Sense Media authors make a related point: users were left without notice and without a referral to other care. A tool that expects to end someday should say what it will do for you when it does.

What a person who used them should do now

Start with your data. If you still have an export, keep it somewhere you control. If you do not, a company that has closed may still owe you an answer: under the GDPR you can ask a controller for a copy of your data (Article 15) and, on stated grounds, for erasure (Article 17), and you can complain to your national data protection authority if it does not answer. Do not assume a closed app has deleted anything; ask in writing and keep the reply.

Then notice what the loss was. Many people used these tools between difficult moments: a place to put a bad evening. If something was holding you, tell a person, a clinician if you have one, that it has gone. A replacement app is optional. Support is not. If a tool taught you something that helped, such as a way of naming a feeling or a question you now ask yourself, write it down in your own words. That outlasts any app.

What to look for next

The closures suggest four checks that apply to any replacement. Our ten-question guide goes through them and six more.

1. Who answers for what it says. In the Common Sense assessment, the apps that did well had licensed or trained humans in the loop. Two school-based apps put a human in touch with a guardian within 15 minutes in simulated crises; the consumer apps had no comparable oversight. That finding concerns teens and a small sample, so read it as a design question to ask, not a verdict.
2. What happens to your words if the company stops. Look for a stated export, a stated deletion period and a named controller.
3. How it handles a crisis. The same assessment describes crisis pathways that let a user dismiss a suicidal statement with one tap. Test it yourself only in a calm moment, and read the vendor's description.
4. What the evidence is, and whose it is. Moore and colleagues, in a peer-reviewed 2025 study, found that language models expressed stigma toward some conditions and responded inappropriately in therapy-like settings, and concluded they should not replace therapists. Treat a product's own testimonials as marketing.

Where TwentyThird fits

TwentyThird is one option among several designs, not a successor to any of these apps and not a CBT tool. It is a psychodynamic AI platform in which a member talks with Oneli, an AI agent, and a licensed psychologist is meant to design and supervise that agent. Be clear about its state: psychologist supervision is in early access, not every member has a psychologist yet, and until one does the agent works on its own. No human is alerted today if a member writes something alarming; the agent answers inside the conversation and shows crisis lines. Members can export their data and delete their account in Settings, and deletion is carried out 30 days after the request. The privacy policy lists each processor. None of this is a trial result: no study has examined TwentyThird.

What this does not replace

A closed app was never a clinician, and neither is a new one. If you were relying on a tool to get through a hard stretch, a psychologist, doctor or counsellor can do what software cannot: take responsibility for you. If you are in danger, contact local emergency services now. In the EU, call 112.