AI therapy: which startup is ahead?

Which AI therapy startups are we really comparing?

The serious AI therapy field currently has six direct competitors, but only Wysa, Limbic, Slingshot AI and The Path have enough recent momentum or clinical weight to shape the debate.

We include products where the AI itself holds a therapeutic or structured mental health conversation with the user. Wysa and Youper come from the earlier chatbot generation, with guided exercises, mood tracking and cognitive behavioral therapy techniques. Limbic sells assessment and between-session support to healthcare providers. Slingshot AI’s Ash, The Path and Sonia use newer generative models for longer, more personal conversations.

We exclude Spring Health, Lyra Health, Talkspace and Headspace because human care, benefits or content still drive their businesses. We also exclude Replika, Character.AI and general assistants such as ChatGPT because they were not built as mental health products. Woebot would once have made the list, but its consumer app closed in 2025, which is a useful warning that research, funding and a known brand do not guarantee a lasting business.

Funding figures are approximate. Private databases do not always treat grants, small early rounds and extensions the same way. For Wysa, we show both the venture total commonly reported by startup databases and the larger figure reached when recent research grants are included.

Startup	What it does	Publicly disclosed funding
Slingshot AI	Runs Ash, a generative AI built for mental health conversations	About $93 million
Wysa	Combines AI self-help, guided exercises, coaching and healthcare integrations	About $30.5 million in venture funding, roughly $40 million including recent grants
Limbic	Provides AI intake, assessment and between-session support to mental health services	About $14.7 million
The Path	Offers persistent, proactive AI conversations across therapy and coaching frameworks	$14.3 million
Youper	Provides an AI wellbeing chatbot, journaling and CBT-style exercises	About $5 million
Sonia	Offers text and voice-based AI mental health support	At least $0.5 million publicly confirmed; some databases report $3.4 million

Is Wysa actually leading AI therapy today?

Wysa leads AI therapy overall today, with Limbic close behind in clinical care and Slingshot AI emerging as the strongest consumer challenger.

Wysa simply covers more ground. The company currently reports more than six million users, 11 million covered lives, one billion AI conversations and operations across more than 100 countries. Those are company figures rather than audited revenue. Even with that caveat, no direct rival reports the same reach across both consumers and healthcare buyers.

Limbic is smaller but stronger inside healthcare. It says its software has completed more than 650,000 clinical assessments and has been used by 45% of NHS Talking Therapies services. The difference is where the product is used: inside intake, assessment and treatment support, rather than waiting for people to find an app on their own.

Ash is building from the consumer side. The app is newer, free and built much more clearly around modern generative AI. Its research program is already substantial for such a young product. Slingshot still lacks the provider contracts, paid adoption and clinical deployment needed to move ahead of Wysa or Limbic.

The market currently looks like a narrow top two, with Ash as the one credible consumer upset.

If you want more recent data on this point, please see our latest digital health market report.

Which AI therapy startup has reached the most people?

Wysa has reached the most people by a wide margin; Youper is the only older consumer rival that comes remotely close.

Wysa says more than six million people have used its product. Youper has previously reported more than three million users, although it has shared few recent growth figures. Ash’s latest large study covered 102,684 users. The Path is growing quickly too, but it remains far smaller than the older consumer apps. Sonia publishes no comparable reach figure.

Overall, the scale gap is huge. Wysa’s reported user base is roughly 60 times the size of Ash’s published research cohort and more than 100 times The Path’s current membership. These figures say little about which bot gives the best conversation. They reveal who has already handled the messy work of serving many different users: onboarding, repeat use, localization, crisis routing and support.

Limbic’s number means something different. Its 650,000 assessments are fewer than Wysa’s consumer users, but each one happened inside a real care pathway. A provider-linked assessment is also worth more commercially than a download from someone who may never return. For sheer reach, Wysa wins. For people brought into a real care pathway, Limbic is much closer than the app totals suggest.

Who has the strongest clinical evidence in AI therapy?

Limbic now has the strongest high-quality clinical evidence, while Wysa still owns the broadest research record.

The most important recent paper came from Limbic and appeared in Nature Medicine. In a controlled study, 227 participants spoke with either a general language model, the same model enhanced by Limbic’s clinical reasoning layer, or a licensed CBT therapist. Twenty-two clinicians rated the transcripts blindly. The enhanced models scored 43% higher than the general models on the Cognitive Therapy Rating Scale, and clinicians preferred them on core clinical criteria 83% of the time.

The researchers then checked whether the controlled result appeared in real use. They analyzed 19,674 conversations from 8,920 people in North America and NHS services. Users with the highest exposure to the clinical reasoning layer had a 52% recovery rate, compared with 33% among those with lower exposure. The real-world comparison was not randomized, so it cannot prove that heavier exposure to Limbic caused the better recovery rate. Even so, Limbic has clearly moved beyond satisfaction surveys and app-store reviews.

Wysa’s evidence is broader. Its current research page lists eight clinical trials, six service evaluations, 13 real-world studies and more than 36 peer-reviewed publications. The work covers depression, anxiety, chronic pain, maternal health, workplace support and several countries. The quality varies, but the volume and range are hard to dismiss.

Ash has produced unusually large real-world usage datasets for such a young product. One study grouped 102,684 users by engagement pattern and found moderate improvements in depression and anxiety among people who completed follow-up measures. The catch is follow-up: only 298 people completed both the baseline and follow-up depression and anxiety measures. The Path and Sonia currently rely much more heavily on company-reported usage, reviews and early safety work.

Startup	Best available evidence	Our reading
Limbic	Nature Medicine controlled study plus large NHS and real-world datasets	Strongest recent clinical evidence
Wysa	Dozens of peer-reviewed studies across several conditions and settings	Broadest accumulated evidence
Slingshot AI	Large real-world studies of Ash users	Promising, but follow-up samples remain small
Youper	Earlier observational studies on anxiety and depression	Credible history, limited recent progress
The Path	Early safety benchmarking and company-reported usage	Too new for a clinical verdict
Sonia	Early internal outcome claims	Evidence remains thin

Which AI therapy startup is closest to real healthcare?

Limbic is currently the closest to becoming part of normal mental healthcare, and the gap is clear.

Limbic Access holds Class IIa UKCA medical-device status. The product helps collect patient information, supports assessment and connects with clinical systems. The certification does not prove that every Limbic response is correct, but it forces the company to document risk management, testing, usability, security and software changes in a way ordinary wellness apps do not.

Lately, Limbic has moved deeper into the clinician’s working day. Limbic Care added a medical AI scribe built with TORTUS, alongside guided CBT sessions and patient information in the clinician dashboard. The scribe puts documentation, caseload management and between-session support in one place. Limbic is becoming a broader clinical workspace, with the chatbot as only one part of it.

Wysa comes next. It has an FDA Breakthrough Device designation for one specific conversational CBT product aimed at chronic musculoskeletal pain with related anxiety or depression. The FDA is clear that breakthrough status speeds interaction and review without granting marketing clearance. Wysa now states that limitation explicitly.

Regulation is also pushing the category in Limbic’s direction. Illinois law now restricts AI from independently providing therapy or psychotherapy, while Nevada limits systems that present themselves as professional mental health providers. Products sold as supervised tools for clinicians have a cleaner path than apps claiming to replace a therapist. Limbic fits that supervised model better than anyone else in the group, while Wysa is also moving further toward hybrid care.

Who has built the strongest AI therapy business?

Wysa has built the strongest overall business, but Limbic has the best proof that healthcare organizations keep using the product for important work.

Wysa is no longer relying on a consumer subscription alone. It distributes through employers, insurers, health systems and public programs. It also merged with April Health, which brings behavioral care into primary care, and acquired Kins, a US physical therapy provider. Those deals reveal a clear strategy: attach Wysa’s AI to places where patients already receive reimbursed care.

Limbic sells through fewer channels, but the product goes deeper once a service adopts it. Adoption across nearly half of NHS Talking Therapies gives it a base it can keep expanding. In the United States, Rogers Behavioral Health reported that people screened through Limbic were admitted at three times the rate of ordinary web submissions, with 90.5% positive patient satisfaction. The result comes from a customer case study rather than an independent audit, but it is strong enough for a provider to take into a budget meeting.

The money story also looks better for these two companies than for the consumer-only group. A health system can justify spending when software shortens assessments, raises completed referrals or saves clinician time. Ash remains free and Slingshot has disclosed no substantial revenue or enterprise contract base. The Path has strong early engagement, yet its public reviews do not tell us how many people pay or stay for months. Youper has a mature consumer product but has said little lately about revenue, customers or new deals.

Wysa wins commercially because it has more ways to get paid and more places to meet patients. Limbic’s contracts may prove harder to displace.

If you want more recent data on this point, please see our latest digital health market report.

Which AI therapy startup is growing fastest now?

Ash is growing fastest on the consumer side, Wysa has the strongest overall momentum, and The Path is the standout new entrant.

Ash’s research cohort has now passed 100,000 users, up from a 50,000-person beta before launch. The same study found that two-thirds of users opened the app at least once overnight, suggesting that Ash is reaching people when human help is usually unavailable. The overnight pattern is striking, although more than half of users were classified as early dropouts.

Wysa’s growth looks quieter because it is happening through care channels rather than download spikes. It added April Health and Kins, expanded to six more languages and received a £5.3 million Wellcome grant for work with adolescent girls in rural India. Its systems are now handling around 1.5 million requests per quarter. These moves all point in the same direction: Wysa is putting its AI inside more real care channels.

The Path has gone from a small launch to more than 50,000 members, 2.5 million sessions and a 4.9 rating across 2,700 verified reviews. Its $14.3 million seed round gives it enough capital to turn that start into a meaningful consumer business. We still need retention and outcome data before calling it a top-three company.

Limbic’s latest momentum is clinical. It published the strongest new research in the field, expanded its US provider work and added regulated scribing to Limbic Care. Youper and Sonia have not produced a similar run of visible milestones.

Startup	Recent momentum	What it shows
Slingshot AI	Public launch, heavy overnight use and a rapidly expanding research base	Fastest consumer growth
Wysa	Six new languages, major research funding and higher quarterly request volume	Strongest business momentum
The Path	Rapid member growth, heavy session volume and a fresh seed round	Strongest new launch
Limbic	Major clinical publication, US provider traction and a broader care product	Strongest clinical momentum
Youper	Few major recent disclosures	Established, but slowing
Sonia	Limited public milestones	Too early to judge

Which AI therapy product is actually ready to scale?

Wysa has the most complete product for broad deployment, and Limbic has the best product for scaling inside healthcare.

Wysa can serve an individual, sit inside an employer benefit, guide a patient toward care, support coaching and connect with primary care or physical therapy. Its product is less flashy than the newest generative apps, but it already works across several buyer types and countries. The company’s recent engineering update on handling 1.5 million requests per quarter gives us a better view of real operating load than a lifetime download total.

Limbic is built for a narrower job. It handles intake, assessment, clinician decision support, between-session exercises and now clinical documentation. Provider deployments take longer because every organization has its own records, governance and local processes. Once integrated, however, the product becomes part of daily work and is harder to remove.

Ash and The Path can scale conversations quickly because users download them directly. Their problem is keeping enough people engaged. Ash’s large cohort split into 52.2% early dropouts, 25.3% weekly users, 16.8% concentrated users, 4.1% intensive users and 1.6% power users. Some people may only need brief help, but the split still shows why sign-ups cannot be treated as ongoing therapy relationships.

The Path reports 2.5 million sessions across more than 50,000 members. That works out to roughly 50 sessions per member, far deeper than a download count suggests. We still need an independent cohort analysis to see whether a small group of heavy users drives most of those sessions.

Which AI therapy startup gets the best results?

Limbic currently shows the best measurable results in real care, though the evidence supports AI-assisted therapy more strongly than fully autonomous therapy.

A peer-reviewed study of Limbic Care covered 3,961 NHS patients receiving human-led group therapy, but only 35 people used the AI support tool. That small group recorded 94.3% reliable improvement versus 69.8% in the matched control group, while reliable recovery reached 71.4% versus 50.7%. The gaps are large, but the treated group contained only 35 self-selected users. We should not treat that as a final verdict. The study is still useful because it follows people inside real treatment rather than volunteers rating a chatbot.

The Nature Medicine research strengthens the case. Users receiving more of Limbic’s specialized clinical reasoning recovered at a higher rate over roughly ten weeks. The controlled part of the study also showed that the extra clinical layer improved CBT quality across several underlying models. Together, those findings give Limbic the best mix of model testing and large clinical deployment.

Wysa has a wider set of positive studies, but fewer results are as recent or directly comparable. One peer-reviewed study of 1,213 users found that adding AI support between coaching sessions made people nearly three times more likely to complete the coaching program. For Wysa’s hybrid model, this is exactly the behavior the bot is supposed to improve: keeping people engaged around human care.

Ash’s early outcome data is encouraging but easy to overstate. Depression and anxiety scores improved among the 298 users who completed both measurements, with moderate effect sizes. That group represented a tiny fraction of the full cohort. We would need a randomized study with much better follow-up before concluding that Ash itself caused the change.

If you want more recent data on this point, please see our latest digital health market report.

Which AI therapy startup handles safety best?

Limbic and Wysa are the safest bets today because they combine technical controls, clinical rules and real provider oversight.

Limbic’s regulated status gives it the strongest formal safety setup. Its products are also designed to work around clinicians rather than quietly replace them. Wysa uses structured pathways, crisis signposting and human oversight across its care products. An independent JMIR review of ten mental health chatbots found that Wysa was the only one offering all five types of crisis support examined by the researchers.

The generative challengers are taking safety more seriously than ordinary companion apps, but the proof is newer. Ash uses a separate guardrail layer and reported that 76 risky sessions in one real-world study followed its escalation process. A later investigation by The Verge still found problems with crisis resources, including location handling, across several chatbots and mental health apps. A working internal flag can still lead to a poor response for the person asking for help.

The Path says its system scored 95 on the VERA-MH benchmark, compared with about 65 for leading general models. VERA-MH itself has shown strong agreement with clinician ratings, which makes the score worth watching. The first version mainly tests suicide-risk detection and response, so it cannot settle broader questions about delusions, dependency, privacy or poor therapeutic advice.

Several layers of accountability beat one impressive benchmark. Limbic ranks first on safety, Wysa second, and the newer generative apps still have more to prove.

What can the AI therapy leaders do that others cannot easily copy?

Limbic owns the hardest advantage to copy, while Wysa has built the widest collection of smaller advantages.

Most visible consumer features are becoming cheap. A competent team can now add memory, voice, mood tracking, therapist personalities, CBT prompts and proactive notifications using commercial models. Those features may improve the product, but they do not protect the company for long.

Limbic combines proprietary clinical reasoning, hundreds of thousands of assessments, NHS integrations and years of provider trust. Its regulated-device processes add another barrier. A competitor would have to match the technology, then repeat the slow work of validation, procurement, integration and clinical adoption.

Wysa has built its lead piece by piece. It has a large user history, a broad research record, relationships with payers and providers, multilingual delivery and human-care assets through April Health and Kins. A rival could copy one or two of those pieces. Matching the whole route from self-help to supported care would take much longer.

Slingshot’s best chance of becoming hard to copy lies in its psychology-specific model and conversation data. The company has enough money to train and test its own system rather than simply wrap a general model. So far, Ash trails cheaper alternatives on demonstrated long-term outcomes and retention. The Path’s limits on excessive validation and very long sessions are thoughtful, but competitors can copy product ideas faster than they can copy clinical evidence or distribution.

Is Slingshot AI turning its funding into real progress?

Slingshot’s progress is real, yet far too small for a company with a funding lead this large.

The company has raised about $93 million. The total is roughly three times Wysa’s venture funding and more than six times Limbic’s disclosed capital. In return, Slingshot has built a purpose-made model, run a large beta, launched Ash for free and assembled a serious research program.

The work is credible. Commercial progress is still too thin for a company funded this heavily. Wysa built several healthcare and payer channels with much less venture money. Limbic built a regulated clinical product and strong provider relationships with a fraction of Slingshot’s funding.

Training a specialized generative model costs more than building an earlier rule-based chatbot, and Slingshot is much younger, so a direct funding comparison has limits. Still, the next evidence needs to change. Another engagement study will add less than a large paid contract, a strong randomized trial or proof that users keep returning after several months.

More than half of Ash users in the large cohort left early. Slingshot now needs to show that its heavy investment can improve that curve, produce paid demand and create an advantage general AI products cannot erase.

If you want more recent data on this point, please see our latest digital health market report.

Are consumer AI therapists or clinical AI tools winning?

Clinical AI tools are winning the business today, even though consumer AI therapists are growing faster and attracting more attention.

Consumer products have obvious appeal. People can open Ash, Wysa, Youper, The Path or Sonia at night, avoid a waitlist and talk without feeling judged. The American Psychological Association’s recent survey found that more than a third of psychologists already have patients using AI as an additional source of mental health help. Demand is here.

The business model remains awkward. Consumers can compare a paid specialist app with ChatGPT, Claude or Gemini, which many already use. They can also leave after one difficult week, making retention unpredictable. A healthcare provider buys for a different reason: shorter assessments, more referrals, less paperwork or better patient follow-through.

Limbic therefore looks stronger than its user count suggests. Its software helps services move people into care and supports clinicians once treatment starts. Wysa is moving the same way through primary care, insurers and physical therapy. The companies currently building the most durable businesses are connecting AI conversations to human care and buyers that can keep paying for them.

Consumer AI therapy will remain the easiest way to reach people who have nowhere else to go. The likely winners, however, will give that conversation a safe next step rather than treating it as the entire product.

Could ChatGPT kill the AI therapy startups?

General AI will wipe out weak therapy wrappers, but it will struggle to replace companies with clinical evidence, regulated workflows and healthcare distribution.

ChatGPT, Claude and Gemini already offer voice, memory and long conversations. For many users, that is enough for journaling, emotional support or rehearsing a difficult discussion. A separate app cannot defend itself with a friendly tone and a therapy prompt anymore.

The specialist companies need to beat general AI where mistakes carry weight. Limbic’s research gives the clearest example: adding its clinical reasoning layer raised CBT quality across GPT-4, Claude, Gemini and Llama 3. The specialist layer produced the improvement across several foundation models.

Dedicated products still have room to win on safety. The American Psychological Association currently advises users to treat generative chatbots and wellness apps cautiously, especially when symptoms are severe or a crisis is developing. Research published lately has found that general models can acknowledge distress politely while still giving clinically poor guidance when therapy exercises become difficult.

Wysa and Limbic are best protected because their distribution and evidence live outside the chat window. Slingshot can earn protection if its psychology model keeps producing measurable advantages. The Path must turn its strong early engagement and safety positioning into independent outcomes. Small apps with little research, no provider channel and no proprietary data will find it hard to survive.

Which AI therapy startups are actually ahead?

Wysa is ahead overall today, Limbic is the closest challenger, and Slingshot AI ranks third because its momentum has not yet caught up with its funding.

Wysa wins across the whole category. It combines the largest disclosed consumer reach with distribution through healthcare buyers, a long research record, multilingual delivery and human-care pathways. It has also kept moving through higher request volume, new languages and research funding. No rival currently covers as much of the path from anonymous self-help to care backed by people and providers.

Limbic ranks a close second and already wins the clinical version of the question. Its provider integrations and clinical workflow give it the strongest position inside formal mental healthcare. It could take first place if its US deployments expand and its provider revenue begins to match Wysa’s broader footprint.

Slingshot ranks third. Ash is growing quickly and has a serious technical team. The missing pieces are paid adoption, durable retention and controlled evidence that Ash itself caused the reported improvements. The Path enters fourth after a notably strong launch. Its usage and reviews are impressive, but the product is too new for us to place above companies with years of deployment.

Youper ranks fifth. It helped establish the category and retains a large historical user base, yet it has shared little evidence of fresh commercial or clinical acceleration. Sonia remains sixth because public data on scale, funding and outcomes is still too limited.

Wysa’s lead is real but not secure. Limbic is close enough to overtake it as healthcare buyers become more important than consumer downloads. Slingshot could change the order if Ash converts its funding and fast start into paid use, long-term retention and controlled clinical results. For now, Wysa has done the best job of turning AI therapy into a business that works across consumer use and real healthcare.

Rank	Startup	Why it ranks here
1	Wysa	Best overall mix of reach, research, distribution, product breadth and recent business execution
2	Limbic	Strongest clinical position, with regulated workflows, deep NHS adoption and the best recent evidence
3	Slingshot AI	Fastest major consumer challenger, held back by weak monetization disclosure and early retention
4	The Path	Excellent early engagement and a strong launch, but too little independent evidence
5	Youper	Large historical reach, with limited recent momentum
6	Sonia	A credible product with too little public evidence to rank higher

If you want more recent data on this point, please see our latest digital health market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.