checked on 2026-10-13 · every company fact read on woebothealth.com that day, every trial fact from its journal record · 8 min read
Woebot, the best-known CBT chatbot, is gone. Its maker's FAQ says the app was retired on June 30, 2025 and that old accounts cannot be opened. The trials behind it were smaller and shorter than its reputation: the original one was 70 students for two weeks against a PDF, and the most recent one, with a real comparator, found no significant difference. If you came here looking for it, the things that still exist are below: free peer support from real people, the free therapy routes, and an honest read on whether a general chatbot is a reasonable stand-in at all.
Until this week, this page said Woebot was free, available and clinically studied, and compared it to talking with people. One of those three was still true. We rewrote it rather than leave a comparison with a product you cannot download.
In the company's own words, from its about page read on 2026-10-13: “We designed Woebot to be your personal ally, always available to have a conversation that can help you understand yourself and ease your mind.” Its home page describes the company as “Pioneering chat-based AI wellness solutions since 2017.” The 2017 trial paper describes the product more plainly: a “fully automated conversational agent” delivering “self-help content derived from CBT principles in a conversational format”. It was a scripted CBT course in a chat window, with a cartoon robot, mood check-ins, and short lessons on thought challenging and behavioural activation.
That is not a small thing. Woebot had a randomised trial attached from 2017, and the last of the trials below was still being published after the app was gone.
From woebothealth.com/faq, read on 2026-10-13:
“The Woebot app was retired June 30, 2025. New accounts can no longer be created, and previous accounts can no longer be accessed.”
The same page says the deadline to request a transcript of your conversations was July 15, 2025, and that “all account data is anonymized as of July 31, 2025”. If you had an account and wanted your history, that window has closed. The site's footer now lists “Woebot, the behavioral health copilot for Providers” and a version for payers, with a sign-up form for health systems. The company has not gone away; the thing you could download has.
We do not state a reason for the shutdown because the company's own page does not give one, and we do not cite news coverage for company facts.
Five published studies, each read from its PubMed record on 2026-10-13, with the DOI checked on Crossref the same day. Design and comparator first, then the finding, then what the finding cannot carry. Every one of these papers lists company employees among its authors, which the papers disclose and which does not make them wrong; it is a reason to read the comparator line closely.
Fitzpatrick, Darcy and Vierhile, JMIR Mental Health, 2017 · doi 10.2196/mental.7785 · PMID 28588005
design: randomised, unblinded, 70 college students aged 18 to 28, two weeks. comparator: an NIMH ebook, "Depression in College Students".
found: the Woebot group reduced depression scores (PHQ-9) over the two weeks and the ebook group did not (F=6.47, P=.01). On anxiety (GAD-7) both groups improved and the difference between them was not significant. 17% of participants dropped out before the second measurement.
what it cannot carry: two weeks, 70 people, no follow-up, and the comparator was a PDF. The second author founded the company that made Woebot, and the company paid the participant incentives; both are stated in the paper.
Prochaska and colleagues, Drug and Alcohol Dependence, 2021 · doi 10.1016/j.drugalcdep.2021.108986 · PMID 34507061
design: randomised, 180 US adults who screened positive for substance misuse, eight weeks. comparator: a waitlist.
found: past-month substance use occasions fell by 9.1 in the Woebot group against 3.3 on the waitlist (p=.039). 88% said they would recommend it.
what it cannot carry: the secondary substance use and mood outcomes did not differ between the groups. A waitlist control measures "something versus nothing", not Woebot versus another form of help. Four authors were Woebot Health employees.
Prochaska and colleagues, Journal of Substance Use and Addiction Treatment, 2026 · doi 10.1016/j.josat.2026.209993 · PMID 42035882
design: randomised, 258 US adults with problematic substance use, eight weeks plus a one-month follow-up (NCT04925570). comparator: psychoeducation delivered by email.
found: no significant treatment effect on the primary or secondary outcome (beta −0.675, p=0.741). Both groups cut their past-month substance use occasions by about 13, a large within-group change that held at follow-up.
what it cannot carry: this is the trial that put a real comparator in: the same company, the same program, and once the control group got emailed worksheets instead of a waitlist, the difference disappeared. It was published after the app was retired. Five authors were Woebot Health employees.
Darcy and colleagues, JMIR Formative Research, 2021 · doi 10.2196/27868 · PMID 33973854
design: cross-sectional, retrospective, 36,070 adult app users, no control group. comparator: bond scores reported in the literature for face-to-face and group CBT.
found: within five days of first use the mean "bond" score on the Working Alliance Inventory was 3.8, which the authors compare to 4 for outpatient individual CBT and 3.8 for group CBT.
what it cannot carry: it measured what people said about the bot, not whether they got better, and it compared one app's users to numbers abstracted from other studies of other people. Every author was a Woebot Health employee or paid by the company.
Durden and colleagues, Internet Interventions, 2023 · doi 10.1016/j.invent.2023.100637 · PMID 37635948
design: single-arm, open-label, 256 adults recruited through social media ads, eight weeks. comparator: none.
found: perceived stress and burnout fell and resilience rose over the eight weeks.
what it cannot carry: with no control group, an eight-week drop in self-reported stress is what you would expect from time passing and from having signed up for a study. Every author was a Woebot Health employee.
Read in order, the pattern is the one that keeps turning up in chatbot trials: against nothing (an ebook, a waitlist), the chatbot wins; against a cheap active comparator (emailed worksheets), it does not. Not one of these studies compared Woebot to talking with a person, so this page cannot tell you that people beat the bot. It can tell you that the bot did not beat a PDF by much, and did not beat an email at all.
Nothing below costs money, and nothing below is therapy. We are not going to tell you a chatbot and a person are the same thing, in either direction.
veisund is ours, so read this knowing that. You post what is going on, in text or a short video, under a handle you choose, any hour. People who have been through something similar can reply. It is not a chatbot: replies are written by people, with one automated exception, a safety response that surfaces crisis resources when a post suggests immediate risk. There is no premium tier and nothing to upgrade to. It is also not live chat: nobody is waiting on the other end, so we do not promise a reply or how fast one comes. The community is small and we say so. What it has that Woebot never did is the sentence “I have been exactly where you are” from someone who means it. What the trials say about peer support in general is on is peer support effective.
If what you wanted from Woebot was CBT, the techniques are not proprietary; a person can teach them too. free therapy: every real option in the US lists the routes to actual free or sliding-scale therapy, with how to qualify for each.
free alternatives to 7 cups compares the peer-support services that were still running when we last read their sites, with who answers (people or AI), age limits and what is actually free. It also keeps a list of services that have changed or gone.
A lot of people who used Woebot have moved to a general chatbot. is it ok to use chatgpt as a therapist goes through the one randomised trial of a purpose-built chatbot, the evaluation study in which gpt-4o answered a prompt from someone who had just lost their job by listing tall bridges, the 2025 state laws, and what the FDA's own committee said. Both directions, with the records.
Both directions, as always. A scripted CBT chatbot is the better tool if you want structured exercises on a schedule and will actually do them when a bot asks; if you are not ready to say the thing to a person, even under a handle; if you want a mood log more than a conversation; or if you are using it alongside a therapist and want homework between sessions. The AI companions still running, and whether any of them has a trial behind it, are covered on the 7 cups page above and on our Wysa page. What a chatbot cannot do is make you less alone. The 2017 trial did not measure loneliness, and no Woebot trial ever did.
No. Woebot Health's own FAQ, read on 2026-10-13, says: "The Woebot app was retired June 30, 2025. New accounts can no longer be created, and previous accounts can no longer be accessed." The company's site now describes Woebot as a product for providers and payers, not something an individual can download.
The company's FAQ states the retirement date and what happened to user data (anonymised as of July 31, 2025) but does not give a reason on that page, so we do not state one. What its site does say, read 2026-10-13, is that it now offers "Woebot, the behavioral health copilot for Providers" and a version for payers.
Partly, and less than the headlines said. In the 2017 trial, 70 students who used it for two weeks reduced depression scores where a group given an NIMH ebook did not; anxiety improved equally in both groups. In a 2021 trial, adults with substance misuse cut their use more than a waitlist. In the 2026 trial, with an emailed psychoeducation comparator instead of a waitlist, the difference between Woebot and the control was not significant. No published trial compared it to a human.
It depends on what you used it for. If it was the CBT exercises, the techniques are not proprietary and the routes to free or sliding-scale therapy with a person are on our free therapy page. If it was having somewhere to put a thought at 2am, veisund is free peer support from real people: you post under a handle any hour and people who have been through something similar can reply. It is not a chatbot and does not promise a reply. In a crisis, none of these is the tool: call or text 988.
No. Replies on veisund are written by people. The one automated exception is a safety response that surfaces crisis resources when a post suggests immediate risk. There is no AI in the conversation, no premium tier, and nothing to upgrade to.
Tools teach you skills. People make you feel seen. If Woebot was the tool, the skills are still out there, and the people are too.
Download veisund and talk to people who have been there.
adam
Related: free BetterHelp alternatives · can't afford therapy · mental health apps for men
if you're in crisis: please reach out to the 988 Suicide & Crisis Lifeline by calling or texting 988. you can also text HOME to 741741 for the Crisis Text Line. you are not alone, and help is available 24/7.