evidence·published October 25, 2026·every citation last verified October 25, 2026

can ai be your therapist? what the trials show

every chatbot that sells itself as therapy points at a trial. the trials exist. the question is what they compared the chatbot to, for how long, and who ran them.

this page lays out the randomised trials of mental-health chatbots worth knowing, one row each, then the five meta-analyses that pooled them. the hub page on using chatgpt as a therapist summarises; this one goes trial by trial. every number below was read on the trial record itself.

what did the trials actually test?

mostly, a chatbot against nothing, for a few weeks. of the eight trials below, three used a waitlist or no intervention, two used a government ebook, one used emailed worksheets, one used chatgpt and an assessment-only group, and one used human therapy. the longest ran eight weeks plus a one-month follow-up.

trial	n	who	compared with	length	primary result	who ran or paid
therabot (heinz et al., nejm ai 2025)^[1],[2]	210	us adults with clinically significant depression, generalised anxiety, or high eating-disorder risk	waitlist (no app until the study ended)	4 weeks of use, follow-up at 8	effect sizes d=0.90 depression, 0.84 anxiety, 0.82 eating disorders, per the lead investigator; 6+ hours average use	academic (dartmouth-hitchcock is the registered sponsor); built by the investigators’ own lab
woebot (fitzpatrick, darcy, vierhile, jmir mental health 2017)^[3]	70	people aged 18 to 28 from a university community	nimh ebook, “depression in college students”	2 weeks	phq-9 depression fell in the woebot group, not the ebook group (F=6.47, P=.01); anxiety fell in both; 17% attrition	second author founded the company; it paid participant incentives
tess (fulmer et al., jmir mental health 2018)^[4]	75	students from 15 us universities	nimh ebook	2 or 4 weeks	phq-9 fell in the 2-week daily check-in group vs control (P=.03); gad-7 fell in both tess groups vs control	two authors employed by x2ai, which made tess and paid incentives; a 2026 correction we could not read
wysa (macneill, doucet, luke, jmir formative research 2024)^[5]	68 analysed	adults with arthritis or diabetes	no intervention	4 weeks	depression and anxiety fell in the wysa group (P<.001), not in controls; no change in stress in either	university team, no conflicts declared
woebot for substance use (prochaska et al., 2021)^[6]	180	us adults screening positive for substance misuse	waitlist	8 weeks	past-month use occasions −9.1 vs −3.3 (p=.039); secondary substance-use and mood outcomes did not differ	four authors were woebot health employees
woebot for substance use (prochaska et al., 2026)^[7]	258	us adults with problematic substance use	psychoeducation by email	8 weeks + 1-month follow-up	no significant difference (beta −0.675, p=0.741); both groups cut use by about 13 occasions a month	five authors were woebot health employees
generative therapy app vs chatgpt (kuta et al., jmir mental health 2026)^[8]	147	english-speaking adults recruited online	unstructured chatgpt (gpt-4o-based), and assessment only	3 weeks	both chatbots beat assessment-only on phq-9 (d=−0.47 and −0.44); the purpose-built app did not beat chatgpt on anything	first author employed by the app’s owner; analyses by unaffiliated researchers
friend chatbot vs psychotherapy (spytska, bmc psychology 2025)^[9]	104	women with diagnosed anxiety disorders in combat-affected areas of ukraine	human psychotherapy, 60 minutes three times a week	8 weeks	both improved; therapy more: hamilton −45% vs −30%, beck −50% vs −35%	no funding; single author

what did the therabot trial show, and what was it a trial of?

it showed that a supervised, purpose-built generative chatbot beat a waitlist over four weeks. therabot was given to 106 adults; 104 others got no app until the study ended at eight weeks^[1]. presenting the trial to fda's digital health advisory committee in november 2025, its lead investigator, nicholas jacobson, put the effect sizes at d=0.90 for depression, 0.84 for anxiety and 0.82 for eating-disorder symptoms, with “6+ hours average usage”^[2].

asked by committee members how safety was handled, he said: “we monitored every message sent to or from the chat bot in near real time,” and that when there was “any discussion of suicidal intent and/or even ideation, we did follow up with folks to conduct a risk assessment” with a clinical psychologist or psychiatrist^[2]. that is the thing that was tested: a chatbot with clinicians reading over its shoulder, against a group that got nothing. it is a real result. it is not a result about an unsupervised app, and it is not a result about chatgpt.

why does the comparison group matter so much?

because a waitlist inflates the effect, and the one programme tested both ways shows it. at the same fda meeting, committee member ray dorsey said: “putting someone back on the wait list is not an adequate control. if anything, that's going to overstate the efficacy”^[2]. people on a waitlist know they are getting nothing; people given an app know they are getting something, and they are the ones filling in the questionnaire.

woebot's substance-use programme is the clean illustration. against a waitlist in 2021, it cut past-month use occasions by 9.1 against 3.3 (p=.039)^[6]. against emailed psychoeducation in 2026, with the same company's staff among the authors, there was no significant difference: both groups cut use by about 13 occasions a month, and both held it at follow-up^[7]. the chatbot did not stop working. the control group stopped being nothing.

the woebot and tess trials used a government ebook as the control^[3],[4], which is a step up from a waitlist but still a pdf against an interactive app. in the woebot trial, anxiety fell in both groups; only depression separated^[3].

has any chatbot been tested against a human therapist?

once, in a randomised trial we could find, and the humans did better. spytska, in bmc psychology in 2025, randomised 104 women with diagnosed anxiety disorders in combat-affected areas of ukraine to a chatbot called friend, used daily, or to 60-minute psychotherapy sessions three times a week, for eight weeks^[9]. both groups improved. the therapy group improved more: a 45% reduction on the hamilton anxiety scale against 30%, and 50% against 35% on the beck scale.

it settles less than it seems. it is a single-author study analysed with t-tests, the comparison was intensive in-person therapy rather than ordinary care, and the full text and abstract disagree on which beck inventory was used^[9]. the friend chatbot is described in terms of natural language processing and machine learning, not as a large language model. we searched pubmed and found no randomised trial that put a large language model against a human therapist and measured whether patients got better.

do the newer, chatgpt-style chatbots work better?

not demonstrably, because there are not enough trials to tell. a 2025 meta-analysis that split the two kinds found rule-based chatbots had a small but significant effect on depression (g=0.27, 95% ci 0.02 to 0.51), while llm-based chatbots had a bigger point estimate that was not significant (g=0.41, −0.73 to 1.55)^[14]. a separate 2025 meta-analysis of 14 trials of generative chatbots found an effect of 0.30 on negative mental-health outcomes, only just significant (p=.047), with a prediction interval from −0.85 to 1.67: the next trial could plausibly find harm or a large benefit^[13].

the most direct test is small. in a 2026 feasibility trial, a purpose-built generative therapy app and plain chatgpt both beat assessment-only on the phq-9 over three weeks (d=−0.47 and −0.44), and the purpose-built app did not beat chatgpt on any outcome^[8]. 62% of the chatgpt group completed all nine sessions against 39% of the therapy-app group. the first author worked for the app's maker^[8].

what do the meta-analyses say, together?

something on depression, nothing clear elsewhere, and not for long. the 2023 npj digital medicine review pooled 15 trials: g 0.64 for depression (95% ci 0.17 to 1.12), 0.70 for distress, and no significant effect on overall wellbeing (0.32, −0.13 to 0.78)^[10]. the 2024 journal of affective disorders review pooled 18 trials and 3,477 people: −0.26 for depression and −0.19 for anxiety, clearest at eight weeks, with “no substantial effects” at three months for either^[11].

the 2025 jmir review of young people aged 12 to 25 pooled 15 trials and 1,974 people: after adjusting for publication bias, g 0.61 for depression, but 0.06 for anxiety, 0.002 for stress and 0.04 for wellbeing, none significant^[12]. the authors name the lack of follow-up measures as a key limitation. read together, the reviews agree on direction for depression and disagree 2.5-fold on size, and only one measured what was left after three months.

what about the people who stop using it?

they disappear from the results, and it matters. a 2026 meta-analysis of 39 studies put pooled attrition at 17.04%, lower for generative chatbots (9.25%) than retrieval-based ones (23.12%), and found that in depression trials, lower attrition went with larger effects^[15]. the woebot trial lost 17% before its second measurement^[3]; the 2026 feasibility trial saw only 39% of one arm finish^[8]. the trials measure people who signed up for a study and stayed. they do not measure the person who downloaded an app at a bad moment and never opened it again.

what happened to woebot?

the best-studied consumer chatbot no longer exists as a consumer app. woebot health's own faq says: “the woebot app was retired june 30, 2025. new accounts can no longer be created, and previous accounts can no longer be accessed”^[16]. the company gives no reason on that page. the trials behind it, both directions, are on woebot is gone: what the trials said, and what people moved to is on our woebot alternative page. a treatment whose evidence base outlives the product is a fact worth weighing before you lean on any app.

where talking to people fits

we run a peer-support app, so weigh this paragraph accordingly. peer support is people with lived experience, not clinicians, and it is not therapy either; in trials it reduces depressive symptoms but is not established as equivalent to professional care, as we lay out on is peer support effective. on veisund no ai writes replies; people answer, with one automated exception we name: if a post suggests immediate risk, an automated safety response surfaces crisis resources. you can post any hour, and people can answer, though nobody is promised an answer. the trials above are the honest benchmark for every option, ours included: what was it compared against, and for how long. the wider comparison is on ai vs human support, and the free routes to an actual therapist are on free therapy.

what to ask of any chatbot's trial

questions, not advice. they are the ones the trials above make it reasonable to ask.

was it randomised, and against what: a waitlist, a leaflet, an active treatment, or a person?
how long did it run, and was anything measured after it stopped?
was the outcome a questionnaire the participant filled in, or an assessment by someone blind to the group?
were humans monitoring the conversations during the trial, and will they be when i use it?
who wrote the paper, and do they work for the company?
was the chatbot in the trial the same one, the same model and version, that i would be using now?

this is not medical advice. it is a summary of published research, it is not a diagnosis, and it is not a recommendation for or against any treatment — nobody here has met you. decisions about starting, changing or stopping a medication belong to you and a prescriber who knows your history. do not change a prescribed medication on the strength of a web page, this one included.

last verified October 25, 2026. if a source is updated, corrected or retracted, this page gets changed and re-dated.

how we verify

every study below was fetched live on the access date shown — from PubMed (NCBI E-utilities) or Europe PMC for journal articles, or from the issuing body itself for regulatory documents and guidelines — and its details were checked against that record rather than copied from another article.
every record was checked for retractions, expressions of concern and errata. nothing retracted is used, and corrections are disclosed on the source.
funding is read, never inferred. a study is marked independent, industry, mixed or none only where we could read the funding statement itself. otherwise it says unknown — we do not guess a funder from the journal, the authors or the topic.
numbers here are read from the paper. where a paper does not report a figure, we leave it out instead of estimating.
every source carries its own caveat, including the ones whose findings we like. the evidence tier scores the method, funding and venue — not whether the conclusion is big, current or settled.
we are not clinicians. this page reports published evidence. it cannot account for your history, your diagnoses or your medications.

find an error? tell us and we will fix it and re-date the page. our full scoring method is at how we score evidence.

sources

primary sources only — no news write-ups, no secondary summaries. each was fetched and checked on the access date shown.

[1] Heinz MV, Mackin DM, Trudeau BM, Bhattacharya S, Wang Y, Banta HA, Jewett AD, Salzhauer AJ, Griffin TZ, Jacobson NC. Randomized Trial of a Generative AI Chatbot for Mental Health Treatment. NEJM AI, 2025. doi:10.1056/AIoa2400802.
randomised controlled trial (NCT06013137, sponsor Dartmouth-Hitchcock Medical Center) of a purpose-built generative chatbot, Therabot, for 4 weeks against a waitlist, in adults with clinically significant depression, generalised anxiety, or at high risk for eating disorders; outcomes at 4 and 8 weeks. Published 27 March 2025. Design read from the publisher abstract and the trial registry; the full paper is paywalled. · n = 210 · evidence tier: early signal · funding: academic sponsor (Dartmouth-Hitchcock Medical Center, per the registry); the paper’s funding statement was not read · accessed October 25, 2026
the catch: one trial, waitlist control, self-reported symptom scales, 8-week horizon, every message monitored by humans during the trial, and the chatbot was built by the investigators’ own lab.
[2] US Food and Drug Administration, Center for Devices and Radiological Health; presentation by Nicholas C. Jacobson (Dartmouth). Digital Health Advisory Committee, 6 November 2025 (Generative AI-Enabled Digital Mental Health Medical Devices): meeting transcript, and presentation by Nicholas C. Jacobson, “An Academic Perspective on Developing & Regulating Generative AI for Mental Health”. FDA advisory committee record, 2025.
official meeting transcript (fda.gov/media/190451) and the Therabot lead investigator’s slides (fda.gov/media/189846), in which he reports the trial’s effect sizes and describes its human oversight · evidence tier: moderate · funding: n/a (regulator record); the slides are the investigator’s account of his own trial · accessed October 25, 2026
the catch: the effect sizes on the slides are the developer presenting his own product to a regulator; they match the published trial’s framing but are not an independent re-analysis.
[3] Fitzpatrick KK, Darcy A, Vierhile M. Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial. JMIR Mental Health, 2017. doi:10.2196/mental.7785. PMID 28588005.
unblinded randomised trial, 70 people aged 18 to 28 from a university community, 2 weeks of Woebot against an NIMH ebook on depression in college students · n = 70 · evidence tier: early signal · funding: industry-linked: the second author founded Woebot Labs, which paid the participant incentives (both stated in the paper) · accessed October 25, 2026
the catch: two weeks, 70 people, no follow-up, a PDF as the comparator, 17% attrition, and the company’s founder as an author.
[4] Fulmer R, Joerin A, Gentile B, Lakerink L, Rauws M. Using Psychological Artificial Intelligence (Tess) to Relieve Symptoms of Depression and Anxiety: Randomized Controlled Trial. JMIR Mental Health, 2018. doi:10.2196/mental.9782. PMID 30545815.
randomised trial, 75 students from 15 US universities: Tess for 2 weeks or 4 weeks against the NIMH ebook · n = 75 · evidence tier: early signal · funding: industry-linked: two authors were employees of X2AI, which made Tess and paid the participant incentives · accessed October 25, 2026
the catch: very small arms (24 to 26 people), 2 to 4 weeks, an information-only control, company authors. PubMed lists a correction published in August 2026 that we could not read, so figures here are from the original abstract.
[5] MacNeill AL, Doucet S, Luke A. Effectiveness of a Mental Health Chatbot for People With Chronic Diseases: Randomized Controlled Trial. JMIR Formative Research, 2024. doi:10.2196/50025. PMID 38814681.
randomised trial (NCT04620668), adults with arthritis or diabetes, 4 weeks of the Wysa app against no intervention; 68 analysed · n = 68 · evidence tier: early signal · funding: independent (university authors; no conflicts declared) · accessed October 25, 2026
the catch: small, four weeks, no-treatment control, results reported as within-group changes rather than a between-group effect size.
[6] Prochaska JJ, Vogel EA, Chieng A, Baiocchi M, Maglalang DD, Pajarito S, Weingardt KR, Darcy A, Robinson A. A randomized controlled trial of a therapeutic relational agent for reducing substance misuse during the COVID-19 pandemic. Drug and Alcohol Dependence, 2021. doi:10.1016/j.drugalcdep.2021.108986. PMID 34507061.
randomised trial, 180 US adults screening positive for substance misuse, 8 weeks of Woebot-SUDs against a waitlist · n = 180 · evidence tier: early signal · funding: industry-linked: four authors were Woebot Health employees · accessed October 25, 2026
the catch: waitlist control; secondary substance-use and mood outcomes did not differ between groups.
[7] Prochaska JJ, Wang M, Chieng A, Moore J, Vogel EA, Baiocchi M, Pajarito S, Pirner MC, Bullard L, Darcy A, Robinson A. A relational agent for treating substance use in adults: A randomized controlled trial with a psychoeducational comparator. Journal of Substance Use and Addiction Treatment, 2026. doi:10.1016/j.josat.2026.209993. PMID 42035882.
randomised trial (NCT04925570), 258 US adults with problematic substance use, 8 weeks of Woebot-SUD against psychoeducation by email, 1-month follow-up; 202 analysed · n = 258 · evidence tier: moderate · funding: industry-linked: five authors were Woebot Health employees · accessed October 25, 2026
the catch: one programme, substance use rather than depression; a null result reported by the company’s own staff, which counts in its favour.
[8] Kuta B, Novak L, Zidkova R, Furstova J, Malinakova K, De Winter A, Husek V. Effectiveness of a Fully Automated Mobile Therapeutic Versus a General Chatbot in Reducing Depression and Anxiety and Improving Well-Being: Feasibility Randomized Controlled Trial. JMIR Mental Health, 2026. doi:10.2196/82642. PMID 42018980.
three-arm feasibility randomised trial, 3 weeks: a purpose-built generative therapy chatbot (n=44) vs unstructured ChatGPT (GPT-4o-based, n=60) vs assessment only (n=43) · n = 147 · evidence tier: early signal · funding: mixed: the first author was employed by the company that owns the therapy app; the paper says analyses were run by unaffiliated researchers · accessed October 25, 2026
the catch: a feasibility trial, three weeks, self-reported outcomes, only 39% of the therapy-app group completed all sessions.
[9] Spytska L. The use of artificial intelligence in psychotherapy: development of intelligent therapeutic systems. BMC Psychology, 2025. doi:10.1186/s40359-025-02491-9. PMID 40022267.
randomised trial, 104 women with diagnosed anxiety disorders living in areas of active combat in Ukraine, 8 weeks: the Friend chatbot daily vs 60-minute psychotherapy sessions three times a week · n = 104 · evidence tier: early signal · funding: none (the paper states no funding was received); single author, no competing interests declared · accessed October 25, 2026
the catch: single author, analysed with t-tests, the chatbot is not described as a large language model, and the full text names the second scale as the Beck Depression Inventory where the abstract says Beck Anxiety Inventory.
[10] Li H, Zhang R, Lee YC, Kraut RE, Mohr DC. Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. npj Digital Medicine, 2023. doi:10.1038/s41746-023-00979-5. PMID 38114588.
systematic review of 35 studies, meta-analysis of 15 randomised trials (search to May 2023) · evidence tier: moderate · funding: unknown (authors declare no competing interests) · accessed October 25, 2026
the catch: wide confidence intervals and mostly pre-LLM agents.
[11] Zhong W, Luo J, Zhang H. The therapeutic effectiveness of artificial intelligence-based chatbots in alleviation of depressive and anxiety symptoms in short-course treatments: A systematic review and meta-analysis. Journal of Affective Disorders, 2024. doi:10.1016/j.jad.2024.04.057. PMID 38631422.
meta-analysis of 18 randomised trials, 3,477 participants (search April 2024) · n = 3,477 · evidence tier: moderate · funding: unknown (authors declare no competing interests) · accessed October 25, 2026
the catch: heterogeneous chatbots and therapies; the authors name limited population diversity.
[12] Feng Y, Hang Y, Wu W, Song X, Xiao X, Dong F, Qiao Z. Effectiveness of AI-Driven Conversational Agents in Improving Mental Health Among Young People: Systematic Review and Meta-Analysis. Journal of Medical Internet Research, 2025. doi:10.2196/69639. PMID 40367506.
meta-analysis of 15 randomised trials (14 articles), 1,974 people aged 12 to 25 (search to August 2024) · n = 1,974 · evidence tier: moderate · funding: unknown; one author is affiliated with Lingxin AI, a company, although no conflicts are declared · accessed October 25, 2026
the catch: the authors name heterogeneous therapeutic approaches and a lack of follow-up measures.
[13] Zhang Q, Zhang R, Xiong Y, Sui Y, Tong C, Lin FH. Generative AI Mental Health Chatbots as Therapeutic Tools: Systematic Review and Meta-Analysis of Their Role in Reducing Mental Health Issues. Journal of Medical Internet Research, 2025. doi:10.2196/78238. PMID 41401240.
systematic review of 26 studies of generative or hybrid chatbots; meta-analysis of 14 randomised trials, N=6,314 (search completed March 2025) · n = 6,314 · evidence tier: moderate · funding: unknown (no conflicts declared) · accessed October 25, 2026
the catch: only 14 trials, a prediction interval that crosses zero, protocol registered retrospectively, moderate risk of bias.
[14] Du Q, Ren Y, Meng ZL, He H, Meng S. The Efficacy of Rule-Based Versus Large Language Model-Based Chatbots in Alleviating Symptoms of Depression and Anxiety: Systematic Review and Meta-Analysis. Journal of Medical Internet Research, 2025. doi:10.2196/78186. PMID 41343858.
meta-analysis of 15 studies published 2020 to 2025, rule-based and LLM-based chatbots analysed separately · evidence tier: moderate · funding: unknown; four of five authors are affiliated with a Beijing technology company, although no conflicts are declared · accessed October 25, 2026
the catch: few LLM trials, so the LLM estimates have very wide confidence intervals.
[15] Yang H, Liu J, Li L, Jiang R, Lou Q, Sun W, Zhou C. Ethical implications of high attrition in AI-based mental health interventions: a systematic review and meta-analysis. BMC Medical Ethics, 2026. doi:10.1186/s12910-026-01510-7. PMID 42374513.
systematic review and meta-analysis of attrition across 39 studies (PROSPERO CRD420251275051, search to September 2025) · evidence tier: moderate · funding: unknown (no competing interests declared) · accessed October 25, 2026
the catch: attrition from a study is not the same as stopping using an app in ordinary life; real-world drop-off is not measured here.
[16] Woebot Health. FAQs. company statement (woebothealth.com/faq), 2025.
the company’s own FAQ page, read on the accessed date · evidence tier: contested · funding: n/a (the company whose product is the subject) · accessed October 25, 2026
the catch: cited only for what the company says about its own app; it gives no reason for the retirement on that page.

questions

Can AI be your therapist?

Not on the current evidence. Randomised trials show that purpose-built chatbots reduce self-reported depression symptoms more than a waitlist or an information leaflet over two to eight weeks. Pooled estimates range from g 0.26 to 0.64, and the benefit had faded by three months in the one meta-analysis that looked. The only randomised trial against human therapy (104 women with anxiety disorders in Ukraine, 2025) found larger reductions with the human therapists: 45% versus 30% on the Hamilton anxiety scale.

Has any AI therapy chatbot been tested against a real therapist?

One randomised trial we could find. Spytska (BMC Psychology, 2025) assigned 104 women with diagnosed anxiety disorders in combat-affected areas of Ukraine to a chatbot called Friend or to 60-minute psychotherapy sessions three times a week for eight weeks. Both groups improved; the therapy group improved more (45% vs 30% reduction on the Hamilton scale, 50% vs 35% on the Beck scale). It was a single-author study with simple statistics, so it settles little, but it is the only head-to-head trial on the record.

What did the Therabot trial find?

Therabot, a generative chatbot built at Dartmouth, was tested in 210 adults with depression, anxiety or eating-disorder risk: 106 used it for four weeks, 104 waited. The lead investigator told FDA’s advisory committee the effect sizes were d=0.90 for depression, 0.84 for anxiety and 0.82 for eating disorders, with over six hours of average use. The control group received nothing, outcomes were self-reported, and he testified that his team monitored every message sent to or from the chatbot in near real time.

Did Woebot work?

Its original 2017 trial found that 70 college students using Woebot for two weeks reduced PHQ-9 depression scores while those given an NIMH ebook did not; anxiety improved in both groups. Later Woebot trials in substance use beat a waitlist in 2021 but showed no difference from emailed psychoeducation in a 2026 trial of 258 adults. Woebot Health’s own FAQ says the app was retired on June 30, 2025.

Do ChatGPT-style chatbots work better than the older scripted ones?

Not demonstrably yet. A 2025 meta-analysis found rule-based chatbots had a small significant effect on depression (g=0.27) while LLM-based ones had a higher but non-significant estimate (g=0.41, confidence interval −0.73 to 1.55), because there are so few LLM trials. A 2026 feasibility trial found a purpose-built generative therapy app did no better than unstructured ChatGPT; both beat no intervention on the PHQ-9 over three weeks.

Why do people say the chatbot trials are weak?

Mostly because of the comparator. Most trials compare a chatbot with a waitlist, an ebook or nothing, which tells you it beats doing nothing. At FDA’s November 2025 advisory committee, member Ray Dorsey said putting someone on a waitlist “is not an adequate control” and would overstate efficacy. When Woebot’s substance-use programme was tested against emailed worksheets instead of a waitlist, the difference disappeared. Outcomes are also self-reported, most trials last two to eight weeks, and many are run by the company that makes the chatbot.

if a few weeks of something helped, a person might help for longer

veisund is free peer support, and you are anonymous to everyone you talk to. you post under a handle you pick — no real name, no phone number, nothing that follows you back to your real life 🤍

get veisund — it's free