- Guided digital CBT beat unguided programmes at the end of treatment, but the gap closed by 6 and 12 months. An individual-patient analysis of 39 trials (9,751 people) found both guided and unguided internet CBT beat control conditions for depression. Guided programmes did better at the end of treatment (PHQ-9 difference −0.8 points, 95% CI −1.4 to −0.2), mainly for people with moderate or worse symptoms (Karyotaki et al. 2021).
- Smartphone apps help, but only a little. Across 176 trials, apps reduced depression (g=0.28; number needed to treat 11.5) and generalised anxiety (g=0.26). Trials that compared apps with inactive controls found larger effects than trials using placebo apps (Linardon et al. 2024).
- In the real world, most people stop. In a study of 93 popular Android mental health apps (10,000+ installs), the median 30-day retention was 3.3% (Baumel et al. 2019).
- In England, NICE allows specific guided digital therapies in NHS Talking Therapies, with practitioner support, while more evidence is gathered. It said the chatbot Wysa should be used only in research (NICE HTG675). NICE recommends the insomnia programme Sleepio as cost saving at £45 per person (NICE HTG624).
- Privacy is a documented risk, not a theory. Under a proposed FTC order, BetterHelp must pay $7.8 million for sharing health questionnaire data with Facebook, Snapchat and others for advertising (FTC 2023), and Cerebral under a proposed order requiring more than $7 million in payments, covering data sharing and cancellation practices (FTC 2024).
- Money trail: the large Sleepio trial on daytime functioning was funded by Sleepio's maker, Big Health, whose co-founder was first author (Espie et al. 2019). Big Health raised $75 million in 2022 in a round led by SoftBank Vision Fund 2 (Lowenstein). The pivotal trial of the depression app Rejoyn was reported by its developers, Otsuka and Click Therapeutics, and its primary analysis was not statistically significant (Rothman et al. 2025).
- Evidence grade: Strong for guided internet CBT for depression and for digital CBT for insomnia. Moderate for CBT-based apps for anxiety and low mood (small effects). Weak for chatbots. Risk for data sharing and poor crisis handling.
Independent evidence review · Digital therapy
Digital CBT is a real treatment, not a gimmick, but its benefit depends heavily on human support, on what it is compared with, and on whether people keep using it. Guided internet CBT for depression and anxiety, and digital CBT for insomnia, have large, consistent trial evidence, much of it publicly funded. Stand-alone apps and chatbots show small average effects that shrink against placebo apps, and real-world use drops off fast. Several flagship products rest on trials paid for or written by their makers. Our confidence: high that guided digital CBT and digital CBT for insomnia help; moderate that self-help apps give a small benefit; low for chatbots.
Table of contents
- Evidence summary
- What digital CBT is, and how it is regulated
- How it works
- What works and what does not
- Benefits by claim
- Risks and all side effects
- Interactions with other treatments
- Who should not rely on an app alone
- Programme lengths studied and product quality
- Follow the money: who makes it and who funded the evidence
- Related research
- Frequently asked questions
- Sources and funding notes
Evidence summary
| Claim | Evidence | Source | Funding / conflict | Strength |
|---|---|---|---|---|
| Internet CBT treats depression; guidance adds benefit | Individual data from 39 trials (n=9,751). Guided and unguided both beat controls short and long term. Guided beat unguided at post-treatment (PHQ-9 −0.8; −1.4 to −0.2), not at 6 or 12 months. Gap largest when baseline PHQ-9 was above 9. | Karyotaki 2021 | Netherlands ZonMw and Swiss National Science Foundation; funders had no role. Several co-authors declared ties to digital-therapy firms. | Strong |
| Smartphone apps reduce depression and anxiety symptoms | 176 trials. Depression g=0.28 (0.23 to 0.33), n=33,567; generalised anxiety g=0.26. CBT-based apps had larger effects on depression. | Linardon 2024 | UK Medical Research Council grant MR/T021780/1 listed on PubMed; other funders not verified | Moderate |
| Waitlist controls inflate effects | Effects are larger against inactive controls than against placebo apps. In a network meta-analysis of CBT trials for depression, no-treatment controls did better than waitlist controls. The authors suggested a nocebo effect, and noted the evidence was of less than ideal quality, with the differences not persisting in post hoc sensitivity analyses adjusting for publication bias. | Linardon 2024; Furukawa 2014 | Furukawa: funder not verified | Moderate |
| People keep using apps | Android apps with 10,000+ installs: median 30-day retention 3.3%. Trial dropout pooled 26.2%, rising to 47.8% after adjusting for publication bias. | Baumel 2019; Torous 2020 | Baumel and Torous: funding not verified in full text. | Weak (against the claim) |
| Digital CBT treats insomnia | 11 trials, n=1,460: improvements in insomnia severity and sleep measures (g 0.21 to 1.09), comparable to face-to-face CBT for insomnia. Largest trial (n=3,755): insomnia d=1.11 at 10 weeks. | Zachariae 2016; Freeman 2017 | Zachariae: co-author developed SHUTi; funding not accessed. Freeman: Wellcome Trust and NIHR grants; Sleepio supplied free by Big Health. | Strong |
| Chatbots reduce depression | 15 RCTs pooled: depression g=0.64 (0.17 to 1.12), wide interval. | Li 2023; Abd-Alrazaq 2020 | Li and Abd-Alrazaq: funding and competing-interest statements not verified in full text. | Weak |
| Rejoyn (prescription app) treats depression alongside antidepressants | n=386 randomised (354 in the primary analysis). Primary analysis: MADRS −1.78 vs sham app (p=0.0568), not significant. Supportive ITT analysis −2.12 (p=0.0211). FDA cleared it in March 2024. | FDA K231209; Rothman 2025 | Developed by Otsuka and Click Therapeutics; author employment not checked in full text | Weak |
| Mental health platforms keep your data private | FTC orders against BetterHelp (2023) and Cerebral (2024) for sharing sensitive data for advertising. Mozilla: 19 of 32 mental health apps carried its privacy warning label in 2023. | FTC; Mozilla 2023 | US regulator; non-profit consumer review | Risk |
| Apps respond safely to crises | An FDA advisory committee discussed hallucinations, context failures and model drift as risks of generative-AI therapy devices. We could not verify the 2018 BBC chatbot tests, so they are not relied on. | FDA DHAC 2025 | Public broadcaster; US regulator's advisory committee | Risk |
Independent evidence and credibility scorecard
Independence tiers: 1 = regulator or government body; 2 = academic or non-profit team with no declared commercial funding for the work; 3 = publicly funded work with in-kind support or minor declared ties to the product's maker; 4 = manufacturer-funded or manufacturer-authored. Credibility runs from A (highest) to D.
| Source | Who funds it | Country | Independence tier (1–4) | Credibility (A–D) | Why it would tell the truth / residual bias |
|---|---|---|---|---|---|
| NICE early value assessments HTG675 (depression) and HTG676 (anxiety) | UK public body | UK | 1 | A− | No product to sell; put several products into research-only status. "Early value" means evidence gaps were accepted on purpose. |
| NICE HTG624: Sleepio | UK public body | UK | 1 | B+ | Independent committee, but much of the evidence it weighed comes from developer-linked trials. NICE itself flagged no direct comparison with face-to-face CBT for insomnia. |
| FDA 510(k) records: Rejoyn, Somryst, SleepioRx, Daylight | US government | US | 1 | B+ | Official clearance decisions. The 510(k) summaries are written by the applicant, and clearance means "substantially equivalent" to an earlier device, not proof of large benefit. |
| FTC orders: BetterHelp (2023), Cerebral (2024) | US government | US | 1 | A | Enforcement findings agreed in settlements. They cover therapy and telehealth platforms, not every app. |
| Karyotaki et al. 2021, IPD network meta-analysis | Netherlands Organisation for Health Research and Development; Swiss National Science Foundation | Netherlands / Switzerland (international data) | 2 | A | Used raw patient data from 39 of 42 eligible trials. Some trialist co-authors declared ties to digital-therapy firms (e.g. HelloBetter shares, SilverCloud salary, MoodGYM royalties). |
| Linardon et al. 2024, 176-trial app meta-analysis | UK Medical Research Council grant (MR/T021780/1); other funders not verified | Australia / US / UK | 2 | A− | Reports its own weak spots, including how effects shrink against placebo controls. Follow-up was 12 weeks or less in 96% of trials. |
| Li et al. 2023, chatbot meta-analysis | Not verified in full text | Singapore / US | 2 | B | No competing interests declared. Pools only 15 RCTs, mostly small, with wide confidence intervals. |
| Baumel et al. 2019, real-world engagement | Funding not verified in full text | Israel / US | 2 | B+ | Used independently collected usage data rather than developer reports. Android only, and only apps with 10,000+ installs. |
| Freeman et al. 2017, OASIS trial | Wellcome Trust and NIHR grants; Sleepio provided free by Big Health; one author's Oxford post funded by Big Health | UK | 3 | A− | Very large (3,755 students) and charity-funded. Compared with usual practice, not a placebo programme, and the Sleepio developer was senior author. |
| Big Health-funded trials: Espie 2019 (Sleepio) | Big Health, maker of Sleepio; first author a Big Health co-founder | UK / US | 4 | B− | Large. Big Health was involved in design, conduct, interpretation and manuscript development. |
| Rothman et al. 2025, Mirai trial (Rejoyn) | Otsuka; authors employed by Otsuka and Click Therapeutics | US | 4 | B | Sham-controlled, which is rigorous. Reported that the primary analysis missed significance, though the paper's conclusion is favourable. |
| Fitzpatrick et al. 2017, Woebot trial | Woebot Labs paid participant incentives; second author is the company's founder | US | 4 | C | Randomised but unblinded, 70 students for 2 weeks, compared with an e-book. Widely cited as evidence for chatbots. |
Note on clustering: most independent syntheses come from a small international network of academic groups (Amsterdam, Deakin, Harvard-affiliated and Oxford teams), several of whose members also develop or advise on digital products. Regulatory evidence is split between the UK (NICE, MHRA) and the US (FDA, FTC). Developer-funded trials are concentrated in the US and UK.
What digital CBT is, and how it is regulated
Digital CBT delivers the techniques of cognitive behavioural therapy (identifying unhelpful thoughts, scheduling activity, exposure, sleep scheduling) through a website, an app or a chatbot. The core ideas are the same as therapist-led CBT, which we review separately in our CBT evidence review and our CBT for insomnia guide. What differs is how much human contact there is, who regulates the product, and who pays for the evidence.
Five types that matter
- Guided internet CBT: a structured online course with regular check-ins from a practitioner. In England these are called "digitally enabled therapies" and are offered through NHS Talking Therapies for anxiety and depression (formerly IAPT) (NICE HTG675).
- Unguided self-help apps: CBT, mindfulness or mood-tracking apps used alone. Some products offer no human contact at all.
- Chatbots: apps that deliver CBT-style content through conversation, such as Woebot and Wysa. Older chatbots used scripted, rule-based replies; newer products may use large language models (Li 2023).
- Prescription digital therapeutics: apps cleared by the FDA as medical devices and available only on prescription, such as Rejoyn for depression and Somryst for insomnia (FDA K231209; FDA K191716).
- Digital CBT for insomnia: programmes such as Sleepio, SHUTi/Somryst and SleepioRx (NICE HTG624; AASM).
Regulatory status by country
| Place / body | Status | Source |
|---|---|---|
| England: NICE (depression) | Space from Depression can be used in NHS Talking Therapies with trained practitioner support, while more evidence is gathered. Iona Mind, Minddistrict and Wysa: research only. Deprexis was removed because it is no longer available in the UK. | HTG675 (formerly HTE8, May 2023) |
| England: NICE (anxiety) | Five therapies allowed with support: iCT-PTSD and iCT-SAD (OxCADAT), Space from Anxiety (SilverCloud), Perspectives (Koa Health) and Spring (Cardiff University). Cerina, Iona Mind, Minddistrict, Resony and Wysa: research only for generalised anxiety. | HTG676 (formerly HTE9) |
| England: NICE (insomnia) | Sleepio recommended as cost saving for people who would otherwise get sleep hygiene advice or sleeping pills; medical assessment first for higher-risk groups such as pregnancy. | HTG624 (formerly MTG70, May 2022) |
| England: NHS England | Products must also pass the Digital Technology Assessment Criteria (DTAC) and an NHS Talking Therapies digitally enabled therapies assessment before use. | NICE HTG675 |
| UK: MHRA | February 2025 guidance explains when a mental health app counts as a medical device, based on its intended purpose and how it works, and how risk class is set. | MHRA press release; guidance |
| US: FDA (cleared devices) | Somryst (insomnia, cleared 2020); Rejoyn (depression, adjunct to antidepressants, March 2024); Sleepio and Daylight (insomnia and anxiety, cleared 2024). | K191716; K231209; K233577; K233872 |
| US: FDA (unregulated apps) | FDA says it intends to use "enforcement discretion" for lower-risk software, giving as an example apps that help people with diagnosed depression, anxiety, PTSD or OCD keep up coping skills. Most consumer apps are therefore not reviewed. | FDA software policy (2022) |
| US: generative AI | An FDA advisory committee met on 6 November 2025 to discuss risks and evidence needs for AI "therapist" devices, using a hypothetical prescription chatbot for depression. | FDA DHAC summary |
How it works
Digital CBT is assumed to work the same way as therapist CBT: by teaching skills people then practise. What the trials actually show:
- Human support matters, at least early on. Guided programmes outperformed unguided ones at the end of treatment, but the gap closed by 6 and 12 months (Karyotaki 2021). In the 2019 meta-analysis of 66 trials, CBT-based apps that offered professional guidance and reminders produced larger effects (Linardon 2019).
- CBT content matters. CBT-based apps had larger effects on depression than apps without CBT content (Linardon 2024).
- Better sleep may drive better mood. In the OASIS trial, reductions in insomnia statistically mediated reductions in paranoia and hallucinations (Freeman 2017). In the Espie 2019 trial, the authors describe insomnia improvement as a mediator of gains in functioning and well-being (Espie 2019, Big Health-funded).
- Rejoyn uses a different idea. It combines brief CBT lessons with an "emotional faces memory task" meant to train cognitive control of emotion; the sham app used a shapes memory task instead (FDA K231209). We found no independent study showing this mechanism works in humans.
What works and what does not
| Use | Verdict | Evidence | Key caveat |
|---|---|---|---|
| Guided internet CBT for depression | WORKS | 39-trial IPD analysis (Karyotaki 2021); NICE conditional approval (HTG675) | Best evidence is for programmes with practitioner support |
| Unguided internet CBT for mild depression | WORKS | Similar to guided when PHQ-9 is 5–9 (Karyotaki 2021) | Less effective than guided for moderate or worse symptoms |
| Self-help smartphone apps for depression or anxiety | SMALL EFFECT | g=0.28 depression, 0.26 anxiety (Linardon 2024) | About 1 in 11–12 users benefit beyond control; shrinks vs placebo apps |
| Digital CBT for insomnia | WORKS | Zachariae 2016; Ritterband 2017; NICE HTG624 | No head-to-head trial of Sleepio vs face-to-face CBT for insomnia |
| Chatbots for depression | MIXED | g=0.64, wide CI (Li 2023) | Small, short, often developer-linked trials; NICE: research only for Wysa |
| Chatbots for overall well-being | NO CLEAR EFFECT | g=0.32 (−0.13 to 0.78) (Li 2023) | Not significant |
| Rejoyn added to antidepressants | UNPROVEN | Primary endpoint missed (FDA K231209) | One manufacturer-run trial |
| Apps for panic symptoms | INSUFFICIENT | g=−0.12, not significant (Linardon 2024) | Few, small, high risk-of-bias trials |
Benefits by claim
Does online CBT work for depression?
Yes, with the strongest evidence for guided programmes. The Karyotaki analysis 39 trials contributed raw data on 9,751 people, and 8,107 were included in the main synthesis and found both guided and unguided internet CBT better than control conditions on the PHQ-9 depression questionnaire, at the end of treatment and in the long term. Guided beat unguided by 0.8 points on the 27-point PHQ-9 right after treatment. That is a modest gap, and it disappeared by 6 and 12 months. The key finding was personalisation: for people with mild or below-threshold symptoms (PHQ-9 of 5 to 9), unguided programmes did about as well; above 9, guided programmes did better (Karyotaki 2021).
In England, NICE allowed Space from Depression to be used with practitioner support, noting that digitally enabled therapies for depression need on average 90 minutes of therapist time, compared with 8 hours for standard care (NICE news, May 2023). NICE asked for more data on recovery, deterioration, dropout and relapse (HTG675).
Do mental health apps help anxiety?
A little, on average. Apps reduced generalised anxiety symptoms by g=0.26 (Linardon 2024). NICE allows five digitally enabled therapies for specific anxiety disorders while more evidence is gathered, with practitioner support; for generalised anxiety, Wysa is research-only (HTG676).
Daylight, a smartphone programme for generalised anxiety, showed a large effect in a 256-person trial (GAD-7 adjusted difference 3.22 points; d=1.08) (Carl 2020). Funding and author affiliations were not re-checked in full text (trial protocol). It also compared Daylight with a waitlist, the comparison most likely to exaggerate benefit (see below).
Do digital CBT apps work for insomnia?
Yes. This is the best-supported use. A meta-analysis of 11 trials (1,460 people) found internet CBT for insomnia improved insomnia severity and sleep measures, with effects comparable to face-to-face CBT for insomnia and generally maintained at follow-up (Zachariae 2016). A publicly funded US trial of SHUTi (the program behind Somryst) in 303 adults found that the insomnia benefits lasted a year (Ritterband 2017). We could not verify a 12-month remission rate, so we do not quote one.
For Sleepio, the largest trial (OASIS, 3,755 UK students, Wellcome-funded) found a large drop in insomnia (d=1.11) and small drops in paranoia (d=0.19) and hallucinations (d=0.24) versus usual practice (Freeman 2017). The Espie 2019 trial (1,711 adults) found small gains in functional health and well-being and a large gain in sleep-related quality of life versus sleep hygiene education. It was funded by Big Health, with NIHR Oxford Biomedical Research Centre support. Big Health was involved in design, conduct, interpretation and manuscript development (Espie 2019).
Can digital insomnia treatment prevent depression?
Possibly reduce symptoms, not proven to prevent diagnosis. The Australian GoodNight trial (1,149 people with insomnia and depressive symptoms, publicly funded) found SHUTi lowered PHQ-9 scores at 6 months versus an attention-matched online control, but new major depression was similar in both groups (9 vs 13 cases; p=0.32) (Christensen 2016).
Do AI chatbots like Woebot and Wysa work?
The evidence is thin. A meta-analysis of 15 RCTs found chatbots reduced depression (g=0.64; 95% CI 0.17 to 1.12) and distress, but not overall well-being. Pooled trials are few (15 RCTs) (Li 2023). The most-cited Woebot trial randomised 70 students for 2 weeks against an e-book; the company's founder was an author and the company paid participant incentives (Fitzpatrick 2017). An early Wysa study had no control group and compared heavy with light users; two of its three authors are listed with Wysa affiliations, and a third, Becky Inkster, was also named a Wysa advisor (Inkster 2018). NICE placed Wysa in research-only status for depression and generalised anxiety (HTG675; HTG676).
Are prescription apps like Rejoyn better than ordinary apps?
Not demonstrably. FDA clearance means the device was judged "substantially equivalent" to a legally marketed predicate device (FDA K231209). In its pivotal trial, the primary analysis showed a 1.78-point MADRS advantage over a sham app that was not statistically significant against the pre-set threshold (p=0.0568) (FDA K231209). The MADRS runs from 0 to 60, so this is a small difference.
Risks and all side effects
Privacy across the app market
Mozilla's 2023 review gave 19 of 32 mental health apps its "Privacy Not Included" warning, and found 40% had got worse than the previous year's review. Some improved: Woebot updated its policy on data access and deletion, and Wysa and the US government's PTSD Coach were praised (Mozilla 2023). Because most consumer apps fall outside FDA review (FDA), privacy protection in the US often depends on consumer-protection law rather than medical regulation.
Dropout and false reassurance
Dropout in depression-app trials averaged 26.2%, and 47.8% after adjusting for missing negative studies (Torous 2020). Outside trials, 30-day retention was a median 3.3% (Baumel 2019). The practical risk is that someone uses an app briefly, it does not help, and they delay effective care.
Trial design that flatters products
Many app trials use an inactive control such as a waitlist rather than a placebo app (Linardon 2024). Waitlists may act as a "nocebo": in CBT trials for depression, people given no treatment did better than people told to wait, although this did not hold in every sensitivity analysis (Furukawa 2014). When apps were compared with active treatments such as face-to-face therapy, there was no significant difference for depression, but there were only 12 such comparisons (Linardon 2019).
Reported side effects in trials
| Effect | What the evidence shows | Source |
|---|---|---|
| Adverse events from Rejoyn | No treatment-emergent adverse events or discontinuations judged related to the app. | Rothman 2025 |
| Adverse events in large Sleepio and SHUTi trials | None reported in OASIS or GoodNight. | Freeman 2017; Christensen 2016 |
| Harm monitoring in chatbot trials | Harm reporting is limited in chatbot trials. | Abd-Alrazaq 2020 |
| Deterioration | NICE asked for data on rates of "reliable deterioration" and stepping up of care, because it is not well measured. | NICE HTG675 |
Interactions with other treatments
| Combined with | What is known | Seriousness | Source |
|---|---|---|---|
| Antidepressants | Rejoyn is cleared only as an add-on for adults already on antidepressants, under a clinician's care. It is not a replacement for medication. | Use as labelled | FDA K231209 |
| Sleeping pills | NICE recommends Sleepio for people who would otherwise be offered sleep hygiene or sleeping pills, and found fewer sleeping pill prescriptions after it was introduced in 9 GP practices. | Discuss with prescriber | NICE HTG624 |
| Usual care and clinician oversight | SleepioRx is cleared as an adjunct to clinician-supervised care; NHS digitally enabled therapies must include practitioner monitoring of safety and progress. | Keep the clinician informed | AASM; NICE HTG675 |
| Other sleep disorders (e.g. sleep apnoea) | NICE advises medical assessment before Sleepio for people at higher risk of other sleep disorders, in pregnancy, or with comorbidities. | Assess first | NICE HTG624 |
| Social media and advertising platforms | Tracking tools inside apps and websites have passed health data to advertisers. | Check privacy settings | FTC 2024 |
Who should not rely on an app alone
- Anyone in crisis or with thoughts of self-harm: apps are not crisis services. See our guide to mental health crisis warning signs.
- People with moderate or severe depression: guided programmes did better than unguided ones when PHQ-9 was above 9 (Karyotaki 2021). NHS digitally enabled therapies are offered after an assessment that includes risk assessment (NICE).
- Children and teenagers using adult products: FDA-cleared products such as Rejoyn and Somryst are indicated for ages 22 and over (K231209; K191716). NICE assessed separate guided programmes for young people (NICE HTE3).
- Pregnancy, possible sleep apnoea or other medical conditions (insomnia apps): get a medical assessment first (NICE HTG624).
- Panic disorder: app trials showed no benefit on panic symptoms (Linardon 2024); see our panic disorder review.
- People who need strong privacy guarantees (for example in custody disputes, immigration or sensitive jobs): check whether the app shares data for advertising (Mozilla).
Programme lengths studied and product quality
These are programme lengths used in trials or stated in regulatory files. They are not recommendations.
| Programme | Length / format studied | Source |
|---|---|---|
| NHS digitally enabled therapies (depression) | Average 90 minutes of therapist time, vs 8 hours for standard care; anxiety programmes about 4 hours vs 10 | NICE news |
| Rejoyn | 6-week treatment plus 4-week extension; CBT lessons, memory-task exercises and text reminders | FDA K231209 |
| Somryst / SHUTi | 9-week prescription programme; six "cores" (SHUTi trial: 6 weeks) | FDA K191716; Ritterband 2017 |
| SleepioRx | 90-day treatment, adjunct to clinician-supervised care | AASM |
| Sleepio (UK) | NICE costing at £45 per person | NICE HTG624 |
| Daylight trial | Smartphone programme; primary endpoint at week 6, follow-up week 10 | Gu 2020 protocol |
| Woebot trial | Up to 20 sessions over 2 weeks | Fitzpatrick 2017 |
Product quality problems
- Products disappear. Deprexis was removed from NICE's recommendation because it was no longer available in the UK (NICE HTG675). Woebot announced it would retire its consumer app on 30 June 2025 (HLTH). Somryst's original maker went bankrupt (see below).
- App stores do not check clinical claims. FDA does not review most consumer mental health apps (FDA); the MHRA says apps that diagnose, prevent or treat conditions must meet medical device standards (MHRA).
- Useful questions before starting: is it based on CBT; has it been tested against a placebo app or active treatment rather than a waitlist; is a human involved; does it share data with advertisers; and what happens if symptoms worsen.
Follow the money: who makes it and who funded the evidence
Who makes and sells it
- Big Health (US/UK; Sleepio, SleepioRx, Daylight): raised $75 million in January 2022 in a Series C led by SoftBank Vision Fund 2, with ArrowMark Partners, Octopus Ventures, Gilde Healthcare, Kaiser Permanente Ventures and Morningside Ventures (Big Health; Lowenstein). Its US address in FDA filings is San Francisco (FDA K233577).
- Otsuka and Click Therapeutics (Rejoyn): the 510(k) applicant was Otsuka America Pharmaceutical (FDA K231209).
- Pear Therapeutics → Nox Health (Somryst): Pear filed for Chapter 11 bankruptcy in April 2023 (Fierce Biotech). Nox Health bought the Somryst-related assets for $3.9 million, within an auction that raised about $6 million in total (Sleep Review, June 2023).
- Woebot Health (US): raised about $90 million in 2021 and $9.5 million in 2022 after earlier rounds, then retired its consumer app in June 2025 (HLTH).
- Wysa (founded in India): raised a $20 million Series B in 2022 led by India's HealthQuad, with British International Investment, the UK's development finance institution; it received FDA Breakthrough Device Designation in May 2022 for adults with chronic musculoskeletal pain and associated depression or anxiety, which is not a clearance (TechCrunch).
- SilverCloud (Space from Depression, Space from Anxiety): agreed in July 2021 to be bought by US telehealth firm Amwell; the announced combined price for SilverCloud and Conversa was about $320 million (MedCity News).
- BetterHelp (Teladoc Health): BetterHelp segment revenue was about $950 million in 2025, as rounded in earnings coverage (Teladoc results). BetterHelp offers human counselling online rather than an app-only CBT programme, but it is the largest privacy enforcement case in this market.
Who funded the trials
- Sleepio: the trial was funded by Big Health, with NIHR support; its first author was Big Health's co-founder, chief medical officer and a shareholder (Espie 2019). OASIS was Wellcome-funded, but Big Health supplied Sleepio free and funded one author's university post (Freeman 2017).
- Daylight: funding not re-checked in full text; the published protocol is cited for design (protocol).
- Rejoyn: developed by Otsuka and Click Therapeutics; author employment not checked in full text (Rothman 2025).
- SHUTi/Somryst: the trial funders were not re-checked in full text (Ritterband 2017; Christensen 2016). Nox Health acquired the Somryst assets from Pear (see above). Developer disclosures in the 2025 older-adults trial were not re-checked (Ritterband 2025).
- Woebot and Wysa: early studies were authored by company founders, employees or paid advisers (Fitzpatrick 2017; Inkster 2018).
- Independent syntheses: funders listed in the source notes (ZonMw, Swiss National Science Foundation, NHMRC, UKRI, NIMH, Singapore government) are not all verified in full text, though some trialist co-authors of the Karyotaki analysis declared links to Pear, SilverCloud, HelloBetter and MoodGYM (Karyotaki 2021).
- The regulator's own money: the MHRA guidance came from a project funded by Wellcome, a charity, at about £2 million (MHRA).
Documented integrity events
- FTC enforcement: BetterHelp (2023, $7.8 million in partial refunds) and Cerebral (2024 proposed order, more than $7 million, with limits on sharing health information for most advertising, subject to court approval) (FTC 2023; FTC 2024).
- Bankruptcy of a prescription-app pioneer: Pear Therapeutics, maker of the prescription insomnia app Somryst, filed for Chapter 11 in April 2023 (Fierce Biotech).
- Product withdrawn from the UK market: Deprexis was removed from NICE's depression recommendation because it is no longer available in the UK (NICE).
Research funding at a glance
12 sources graded. The counts below are the independence tiers assigned in the scorecard above. They count sources, not funding amounts or evidence quality.
Public funding alone does not prove independence, and commercial funding alone does not prove a claim false. Named funders, countries and relationships are in the scorecard and source notes.
What this means: The general case for digital CBT, especially guided programmes and insomnia programmes, rests mainly on publicly funded trials and independent meta-analyses. The case for specific branded products, especially Sleepio's daytime benefits, Daylight, Rejoyn and the chatbots, rests heavily on trials funded or written by the companies that sell them. Venture capital has flowed into the sector, products have been withdrawn or sold cheaply, and the main documented harms so far are about data, not symptoms.
Related research
For wider, evidence-graded context, see these Pure City Research guides:
- Cognitive behavioural therapy (CBT): assessment, treatment evidence and safety
- Cognitive behavioural therapy for insomnia
- Insomnia: assessment, treatments and safety
- Major depressive disorder
- Generalised anxiety disorder
- Behavioural activation for depression
- Social anxiety disorder
- Health anxiety
- Panic disorder
- Mental health crisis warning signs
- Sleep problems: assessment, healthy habits and treatment evidence
- Stress, anxiety and depression: evidence and care options
Frequently asked questions
Is online CBT as good as seeing a therapist?
For some people, roughly. Guided internet CBT clearly beats no treatment for depression (Karyotaki 2021), and internet CBT for insomnia had effects comparable to face-to-face CBT for insomnia (Zachariae 2016). Direct comparisons of apps with face-to-face therapy are few (Linardon 2019), and NICE found no direct comparison for Sleepio (NICE).
Do mental health apps actually work?
On average, a little. Across 176 trials the effect on depression was g=0.28, meaning about 11 to 12 people need to use an app for one extra person to benefit, and it was smaller against placebo apps (Linardon 2024).
Is Sleepio available on the NHS?
NICE recommends Sleepio for insomnia in primary care for people who would otherwise get sleep hygiene advice or sleeping pills, at £45 per person (NICE HTG624). Local availability depends on commissioning, which we did not verify area by area.
Can you get digital CBT through NHS Talking Therapies?
Yes. NICE allows specific digitally enabled therapies in NHS Talking Therapies with practitioner or therapist support, after an assessment (HTG675; HTG676).
Are AI therapy chatbots safe?
Not proven. Trials are small and few measure harms (Abd-Alrazaq 2020), an FDA advisory committee has discussed risks for AI therapy devices (FDA DHAC). NICE says Wysa should be used only in research (NICE).
What is Rejoyn and does it work?
Rejoyn is a 6-week prescription app for adults with depression already taking antidepressants, cleared by the FDA in March 2024. In its manufacturer-run trial, the primary analysis did not reach statistical significance (FDA K231209).
Do mental health apps sell your data?
Some have shared it. The FTC took action against BetterHelp and Cerebral for disclosing sensitive data for advertising (FTC 2023; FTC 2024), and Mozilla flagged 19 of 32 apps in 2023 (Mozilla).
Why do so many people stop using mental health apps?
Independent usage data show a median 30-day retention of 3.3% (Baumel 2019). In trials, dropout was lower when human feedback and mood monitoring were offered (Torous 2020).
Sources and funding notes
Checks on 9 October 2026: The central numbers were checked against PubMed or Europe PMC abstracts, PubMed Central full texts where open, Europe PMC grant records, NICE guidance pages, FTC press releases and company or bankruptcy reporting. Verified: Karyotaki 2021 (39 of 42 RCTs contributed data, 9,751 participants, 8,107 in the synthesis, PHQ-9 difference −0.8, 95% CI −1.4 to −0.2; ZonMw and Swiss National Science Foundation funding with a funder-no-role statement; co-author ties to digital firms); Linardon 2024 (176 RCTs, depression g=0.28, NNT 11.5, N=33,567; UK MRC grant; placebo and inactive control subgroup finding); Baumel 2019 (93 apps, median 30-day retention 3.3%); Torous 2020 (dropout 26.2%, 47.8% after adjusting for publication bias; NIMH grants); Zachariae 2016 (11 RCTs, 1,460 people); Ritterband 2017 (303 adults, effects maintained at 1 year; NIMH grant); OASIS (3,755 participants, insomnia d=1.11; Wellcome Trust and NIHR grants; Sleepio supplied free by Big Health; an author's Oxford post funded by Big Health); GoodNight (1,149 participants; 9 vs 13 major depression cases, p=0.32; NHMRC grants); Espie 2019 (1,711 participants; funded by Big Health, which was involved in design, conduct, interpretation and manuscript; Espie is a Big Health cofounder); Li 2023 (15 RCTs, g=0.64, 95% CI 0.17 to 1.12; no competing interests declared); Rothman 2025 (386 randomised, 354 in the primary analysis, −1.78, p=0.0568; supportive ITT −2.12, p=0.0211; Otsuka and Click employees declared); NICE HTG624 (Sleepio £45, cost saving in primary care, no direct comparison with face-to-face CBT for insomnia) and HTG675 (Space from Depression with support; Wysa research-only); FTC BetterHelp (2 March 2023) and Cerebral (15 April 2024); Big Health $75 million Series C (January 2022); Amwell and SilverCloud (announced July 2021, about $320 million with Conversa, completion not confirmed); Pear Chapter 11 (7 April 2023); Wysa $20 million Series B (July 2022) and its May 2022 Breakthrough designation; Woebot 2021 round of about $90 million; Woebot app retirement (30 June 2025); Mozilla 2023 (19 of 32 apps). Not confirmed to the figure: the 2018 Woebot round (sources give $8 million or $22.8 million, so it is omitted) and the exact BetterHelp 2025 revenue (about $950 million, as rounded).
Not verified and not relied on: the funding statements of Baumel 2019, Li 2023, Abd-Alrazaq 2020, Zachariae 2016, and Furukawa 2014; Daylight (Carl 2020) funding and author affiliations; the 95% confidence interval for Rothman's primary analysis (removed); the Inkster 2018 competing-interests wording beyond the points stated above; any 12-month remission rate for SHUTi. Changed in this review: the Rothman primary-analysis confidence interval; the Furukawa sensitivity-analysis wording (now the authors' own statement, not a commentary); the NICE HTG676 wording (research-only status for generalised anxiety); the Linardon 2019 description; the Linardon 2024 sample (33,567); the BetterHelp wording (a proposed order in March 2023); and the Freeman funding details, the Woebot 2018 amount (omitted) and the exact BetterHelp 2025 revenue (rounded).
- Karyotaki E et al. Internet-based CBT for depression: IPD network meta-analysis. JAMA Psychiatry 2021 (full text) — key synthesis; Netherlands ZonMw and Swiss National Science Foundation; some co-authors declared digital-company ties.
- Linardon J et al. Mental health smartphone apps: meta-analysis of 176 RCTs. World Psychiatry 2024 (full text) — UK Medical Research Council grant MR/T021780/1 listed on PubMed; other funders not verified.
- Linardon J et al. App-supported smartphone interventions: meta-analysis. World Psychiatry 2019 (full text) — no funding statement found.
- Linardon J, Fuller-Tyszkiewicz M. Attrition and adherence in smartphone interventions. J Consult Clin Psychol 2020 — academic; funding not accessed.
- Torous J et al. Dropout rates in trials of apps for depressive symptoms. J Affect Disord 2020 — funding not verified.
- Baumel A et al. Objective user engagement with mental health apps. J Med Internet Res 2019 — funding not verified.
- Furukawa TA et al. Waiting list may be a nocebo condition. Acta Psychiatr Scand 2014 — funder not verified.
- Li H et al. AI-based conversational agents for mental health: meta-analysis. NPJ Digit Med 2023 (full text) — funding not verified in full text.
- Abd-Alrazaq AA et al. Effectiveness and safety of chatbots for mental health. J Med Internet Res 2020 — not verified in full text.
- Fitzpatrick KK et al. Woebot RCT. JMIR Ment Health 2017 — company founder co-author; Woebot Labs paid incentives.
- Inkster B et al. Wysa real-world data evaluation. JMIR Mhealth Uhealth 2018 — all authors linked to Wysa.
- Zachariae R et al. Internet CBT for insomnia: meta-analysis. Sleep Med Rev 2016 — co-author is SHUTi developer; funding not accessed.
- Ritterband LM et al. Web-based CBT for insomnia with 1-year follow-up. JAMA Psychiatry 2017 — funding not verified.
- Ritterband LM et al. Digital CBT for insomnia for older adults. NPJ Digit Med 2025 (full text) — National Institute on Aging; developer disclosures not re-checked.
- Christensen H et al. SHUTi for prevention of depression (GoodNight). Lancet Psychiatry 2016 — Australian NHMRC.
- Freeman D et al. OASIS trial. Lancet Psychiatry 2017 (full text) — Wellcome Trust and NIHR grants; Sleepio supplied free by Big Health.
- Espie CA et al. Digital CBT for insomnia on health and well-being. JAMA Psychiatry 2019 (full text) — funded by Big Health Ltd.
- Carl JR et al. Digital CBT (Daylight) for GAD. Depress Anxiety 2020; Gu J et al. protocol, Trials 2020 — supported by Big Health Inc.
- Rothman B et al. CT-152 (Rejoyn) Mirai trial. J Affect Disord 2025 — developer-run trial; author employment not checked in full text.
- FDA 510(k) K231209: Rejoyn — US regulator; summary written by applicant.
- FDA 510(k) K191716: Somryst — US regulator; summary written by Pear Therapeutics.
- FDA 510(k) K233577: Sleepio and K233872: Daylight — US regulator databases.
- FDA. Policy for Device Software Functions and Mobile Medical Applications (September 2022) — US regulator.
- FDA Digital Health Advisory Committee 24-hour summary, 6 November 2025 — US regulator's advisory committee.
- NICE HTG675 (formerly HTE8): digitally enabled therapies for adults with depression; HTE8 redirect page — UK public body.
- NICE HTG676 (formerly HTE9): digitally enabled therapies for adults with anxiety disorders — UK public body.
- NICE HTG624 (formerly MTG70): Sleepio to treat insomnia — UK public body.
- NICE HTE3: guided self-help digital CBT for children and young people — UK public body.
- NICE news: nine treatment options (16 May 2023) — UK public body.
- MHRA press release (3 February 2025) and guidance — UK regulator; project funded by Wellcome.
- FTC: BetterHelp (2 March 2023) — US regulator.
- FTC: Cerebral (15 April 2024) — US regulator.
- Mozilla Foundation: *Privacy Not Included, mental health apps (2 May 2023) — non-profit consumer review.
- American Academy of Sleep Medicine: FDA clears SleepioRx (2024) — professional society news.
- Big Health: $75 million Series C (19 January 2022); Lowenstein Sandler — company and law-firm announcements.
- Pear Therapeutics Chapter 11 filing, reported April 2023 — trade press.
- Sleep Review: Nox Health closes acquisition of Somryst (June 2023) — trade press.
- HLTH: Woebot Health is shutting down its app (28 April 2025) — trade press.
- TechCrunch: Wysa raises $20 million (14 July 2022) — trade press.
- Amwell: acquisitions of SilverCloud Health and Conversa Health (28 July 2021), press release — company filing.
- Teladoc Health fourth quarter and full year 2025 results — company filing.
Re-checked 9 October 2026. This article is not medical advice; talk to a qualified clinician or pharmacist before starting, stopping or combining treatments.
Have a question — or want us to cover something?
Ask about anything on this page, or request the next deep dive: an ingredient, a supplement, or a health concern. We use published research, evidence syntheses, and regulatory guidance, with clear source links.
One daily research roundup
Get the topics, key findings and links from our new articles in one email. At most one digest a day, only when there is something new.