Can AI Help Your Mental Health? Apps, Evidence & Risks

Right now there is a ton of anti-AI sentiment on social media and the news. I’m not going to dissect that here, but I want to discuss how AI can play a positive part in mental health.

Before we kick off though, I want to stress that I am not saying any of the below are a replacement for proper therapy, care and meds. However, we need to bear in mind that for many people those things might not be easily accessible, be too costly, or perhaps be something they’re not comfortable asking for.

So I’m not saying people should use AI products in lieu of ‘proper’ care – but that I think they either have a place in someone’s overall mental health strategy, or at least provide something if nothing else is accessible.

Why Do People Use AI Products For Mental Health?

Being flippant of course, the answer is ‘for their mental health’. In the absence of proper data, my assumptions are:

• It’s cheap (or free)
• It’s easily or instantly accessible
• You can use it at any time of the day or night
• It can be anonymous (or pseudo-anonymous)
• It avoids any perceived stigma about talking about mental health
• Loneliness

And again, based on my assumptions, I’d be willing to bet the main use cases for AI for mental health are around anxiety and depression. (Related, but not considered mental health issues, I’m sure ADHD and neurodivergence support are also common use cases).

What Problems Do These Apps Claim To Help With?

The common problems AI mental health apps aim to address are:

• Anxiety
• Stress
• Low moods
• Sleep issues
• Loneliness
• Burnout
• Support between therapy sessions

Just as important, it’s worth stating what they’re not for:

• Severe conditions
• Crisis situations
• Diagnosis
• People aged under 18

Given the huge numbers of people across the world that have anxiety etc., this makes the potential market size for these apps massive.

Related: AI Mental Health – Who’s Winning The SEO Race?

Key Stats About People Using AI For Mental Health

There doesn’t seem to be tons of official data out there about how people use AI for mental health. My assumptions for this are:

1) AI product providers like ChatGPT and Claude have no benefit in disclosing it
2) ‘Mental health’ can be quite broad a term – e.g. someone looking for ’emotional support’ or to vent about a bad day might fall in the umbrella, as would depression
3) People surveyed may not feel OK in disclosing how they use AI, due to perceived shame or stigma

• One in eight adolescents and young adults use AI chatbots for mental health advice in the UK (2025, source)
• Almost a fifth of adolescents and young adults reported using AI chatbot in the US (2025, source)
• (77%) have spoken with patients who have used AI for support, engagement, or other reasons, in the US (2026, source)
• (18%) people used AI chatbots to support their mental health in the last 12 months in the UK (2026, source)
• Four in 10 adults in the UK would be happy to use AI for their mental health support in the UK (2026, source)
• 1 in 3 Israeli adults surveyed who use AI tools reported seeking emotional support, coping advice or mental health assistance (2026, source – albeit very small sample size)

AI-Based Mental Health Products

Like any other industries, there’s been a ton of AI products launched aimed at mental health support.

I’ll cover these below – with the caveat of I haven’t personally tried any of them, and this isn’t aimed at being a ‘Top X’ listicle.

I’ve deliberately left out companion apps like Replika. They’re mainly marketed as companionship rather than as mental health support, which makes them a different conversation.

Types of AI-Based Mental Health Products

For products currently on the market, they fall into these two buckets:

• Consumer self-help chatbots
• Clinician / enterprise side

Wysa

What it is: Wysa is a mental health AI platform with four parts:

• Navigator: Helps people find and connect with support. Can be used by employers and insurers
• Gateway: Clinical screening and referrals. Can be embedded on a website and speeds up the data collation and processing for care professionals and insurers. Wysa claims an impressive 30 minutes saved per patient. Gateway is also designed to flag people at high risk (1.9% of users are flagged as this – and signposted to more urgent mental health support).
• AI Self-Help: Conversational AI to support mental health. It includes ‘safety detection built in’ and highly visible ways of contacting local helplines etc.
• Copilot: Apparently this was previously called ‘Wysa Therapist Companion’ and frankly I think calling it ‘Copilot’ (or, to be fair, ‘Wysa Copilot’) was a mistake, as Copilot is so closely associated with Microsoft (and also as a more general application). Anyway, Copilot is a service to help in-between sessions. This includes features like interactive AI-guided tools, secure messaging and crisis escalation. Apparently, each year ‘5.2% of Wysa users trigger the SOS feature.’

Where they’re based: Boston

Founded: 2015

Notable customers: NHS, Vitality

Ash (by Slingshot AI)

What it is: Ash is a free iOS and Android app pitched as ‘the first AI designed for therapy’. It went public in July 2025 after an 18-month beta with around 50,000 users (source).

The big difference from most of the products here is that Slingshot didn’t just put a wrapper around ChatGPT or Claude. They built their own foundation model for psychology. It’s trained on behavioural health data that covers CBT, DBT, ACT, psychodynamic therapy and motivational interviewing. Their clinical team fine-tuned it for things like when to challenge someone, when to stay quiet, and when to end a conversation (source).

But not in the UK: You can’t actually use Ash in the UK anymore. In January 2026 Slingshot pulled it from the UK due to worries that it fell foul of UK medical device regulations (source). It’s a prime real-world example of the ‘therapy vs wellness’ line I cover in the regulation section below.

Where they’re based: New York.

Founded: 2023

Notable customers: None public – it’s direct to consumer. Backers include a16z, Forerunner and Radical Ventures, with $93m raised.

Earkick

Note: I’d urge some caution with this company – here are my reasons (not product related):

• If you try to open EarKick.com, you’re redirected to the Blog.Earkick.com subdomain
• This is a bunch of articles, which seem legit (albeit all the images are AI generated)
• I clicked ‘About’ to research them for this article, and that redirects back to the blog homepage
• If I click on ‘Chat with Panda’ it opens Chat.Earkick.com which gives an error
• Their Facebook and Instagram hasn’t been posted on since October 2025, Twitter/X not since October 2024, YouTube & LinkedIn not since February 2026
• There is no contact page or email
• The most recent Google review was eight months ago which says ‘Requires sign up, isn’t free, and has a hard limit of an hour before having to pay. Isn’t even available on play store, shameful‘

As a consumer, based on all of that, there’s no way I would trust the product, especially given the seriousness of mental health and wellbeing.

With all that said, here is what I could find about them, mainly using the third party site WhitefordResearch.com.

Where they’re based: San Francisco

Founded: 2021

What They Do: An AI mental health app (from what I can see, only Apple – there is a Google Play page but no download) but supposedly also a web-chat experience (I assume ‘Chat With Panda’ if/when it works).
As per WhitefordResearch, ‘Earkick says its technology combines inputs from chat, audio, video/text memos, typing behavior, sleep, exercise, meditation, heart-rate data, contextual tags, symptoms, cycle data, location/weather data when permitted, and panic-attack logs‘

I’m at a loss as to what’s going on with the product and company, but based on their current presence (or lack thereof), it comes across as rather odd to say the least.

Headspace Ebb

What it is: Ebb is the AI companion built into the Headspace app (yep, the very cool meditation one). You tell it how you’re feeling and why, and it talks things through with you, then points you to relevant meditations and exercises from Headspace’s library (source).

• Built by clinical psychologists and data scientists, and based on motivational interviewing
• Launched in the US in October 2024 and came to the UK in April 2025. It’s part of the paid subscription
• 64% of US users surveyed said they felt heard and understood by it (source)
• From May 2025 it’s also been offered to employers, and Headspace says it reached a massive 8 million+ employees across 2,000+ companies (source)
• Headspace is clear that it isn’t meant to diagnose or treat. It’s clearly positioned for self-reflection and as support between human care sessions.

Where they’re based: San Francisco (which, as an aside, is, in my opinion the coolest city in the world)

Founded: 2010 (Headspace); Ebb launched 2024

Notable customers: Consumer subscribers plus 2,000+ employers via Headspace for Organisations

Sonia

Where they’re based: San Francisco

Founded: 2023

Notable customers: We Are Enough, Piramidal, MoovLab, Tekton Dynamics, CrowdVolt

The below text all comes from Lukas Wolf at Sonia Health. Prior to writing this article I reached out to each company – Lukas was the only one who replied. The below is all verbatim from him. Thanks Lukas!

What Sonia is, and who it’s for

Sonia AI is a chatbot general wellness companion product that we have live in the app store for consumers to access. Our technology allows our users to converse back and forth with our AI the way they would with someone who knows them. They typically spend about 10-15 minutes talking with Sonia and leave each session with a concrete skill to practice in their daily lives.

It’s voice-first, available any time, and built for adults 18 and older dealing with everyday stress or burnout, especially people who’d rather talk through something anonymously or simply don’t have easy access to support otherwise. In contrast, it is not a crisis service, it doesn’t diagnose, treat, or cure any condition, and it’s not intended for people who need a higher level of care (those conversations get routed out).

We also are about to launch a second product, Sonia Care, which pairs unlimited Sonia AI with a licensed clinician for people who want or need human or more deeper clinical support. The therapist and AI share information back and forth which we hypothesize will improve clinical care as well as the AI models themselves. Our products, Sonia AI and Sonia Care, will share the same “front door”, with two levels of support behind it.

Why it helps

Two things drive benefit, and our rigorous research (currently under review by a peer-reviewed journal) has shown that it’s NOT the number of times or overall minutes users are in our app.  The first is around access: appointment-based care can’t cover the hours people actually struggle.

More than half of all Sonia AI sessions start outside typical clinic hours (5pm–8am), and a meaningful portion of conversations happen overnight. The second thing that we’ve found drives benefit is something called forming a working alliance with our AI. In our research, how understood and supported someone felt by Sonia AI in their first week, and not how often or how long they used it, was the strongest predictor of improvement.

We build for that first-week connection deliberately, rather than optimizing for engagement metrics the way a lot of digital health products do that operate without conducting research on their users.

Statistics and outcomes

• Efficacy: Earlier this year, we conducted a 4-week RCT, a randomized controlled trial that enrolled 400 adults to Sonia AI or to an active control (a psychoeducation tool used in UK stepped care). We found significant differences favoring Sonia AI users on worry and mood in as little as two weeks. This is currently preprinted and under peer review, not yet in final published form, so I’d frame it as “early trial results” rather than “published study.”
• Engagement and alliance: a related analysis of that same trial data (briefly mentioned above) found no difference in outcomes between heavy and light users of the app (p > 0.50), but first-week working alliance predicted improvement (p < 0.001) and mediated 51% of the engagement-outcome relationship. 
• Usage: our study participants averaged 3.1 hours of conversation across roughly 9 sessions over four weeks. Our native users average about 14 minutes per session. 
• In-app outcomes (preliminary, self-reported, not from the controlled trial): a median 12-point improvement on the WHO-5 well-being scale at two weeks (a 10-point change is considered clinically meaningful), 64.5% of users report feeling improved since starting, and an average Net Promoter Score of 55.6.
• User feedback: 4.7 average rating across more than 1,000 App Store ratings. One reviewer wrote that Sonia “helped me slow down, organize my thoughts, and reframe situations more clearly.”
• Trust: internal analysis found that a large majority of conversations touch on genuinely sensitive ground (grief, relationship difficulties, things people are struggling with), and a meaningful share involve something the person had never told anyone before. We read that as a sign people feel safe with it, not as something to over-index on.

Anything else worth knowing

We think that two worries in our space, safety of AI systems and their associations with loneliness, deserve mention here as well. Safety is built in and a critical feature we build towards. Every one of our sessions is monitored in real time, with a human Clinical AI Safety Monitor overseeing our internal safety system, which we found to work really well in another one of our studies that is currently preprinted and also under review at a peer-review journal.

We’d also point to findings from the RCT internally analyzed that show that Sonia AI users report more social connection and less loneliness over time, not more isolation, which speaks to a common worry about AI in this space.

Limbic

What it is: Limbic is comprised of four products all of which are aimed at mental health providers rather than direct to consumer:

• Limbic Access: Clinical AI assistant to help referrals. It’s a tool that a health provider can embed on their site and provides a full clinical report. The idea being that a clinician can then kick off the first contact with a patient with a decent assessment and possible diagnosis.
• Limbic Care: An app for patients which helps with support between sessions and improved support post-care. Features include tailored courses and content, and access to an AI-powered chat assistant. Notably, the content doesn’t have to be ‘out of the box’ – the care provider can make it bespoke
• Phone Therapy: 24/7 access to AI-based voice therapy which is powered by a CBT method (so there’s clinical structure to it rather than being a generic chatbot). A key benefit is there’s no time limit to the calls.
• AI Therapy: From what I can see on their website, this is the same product as ‘Phone Therapy’.

Where they’re based: HQ in London, team is global and remote-based

Founded: 2017

Notable customers: NHS, Everyturn Mental Health, Rogers Behavioral Health

Lyra Health

What it is: Lyra is a big US workplace mental health provider, so you get it through your employer, not the app store. In October 2025 it launched Lyra AI, which it calls the ‘first clinical-grade AI’ for mental health. It’s a 24/7 conversational guide for mild to moderate issues like sleep issues, burnout and stress, and it’s backed by Lyra’s network of human providers (source). It became generally available to members globally in May 2026.

• Designed to help people use the skills they learn with their human therapist in day-to-day life, not to replace the therapist
• Lyra uses AI to match people to the right provider and level of care
• Lyra published the ‘Polaris Principles’, its own framework for safe, clinically rigorous mental health AI

Where they’re based: Burlingame, California

Founded: 2015

Notable customers: Employers (business-to-business only). It’s been named Workday Wellness’s preferred mental health partner (super impressive as Workday has over 20,000 employees).

Spring Health

What it is: Another employer and health plan provider, and a Lyra competitor. Spring says it was founded as an AI company, originally using machine learning to match people to the right care. In April 2026 it launched Guide, an AI that carries memory and context through someone’s whole mental health journey, so care doesn’t reset every time they change therapist, job or health plan (source).

• Spring Health co-developed VERA-MH, which it describes as the first open-source safety benchmark for mental health AI. It tests how chatbots handle high-risk conversations like suicidal ideation (source).
• Spring acquired Alma (a network of independent therapists) in 2026, adding 20,000+ therapists
• Interestingly, Spring explicitly asks not to be described as an ‘AI chatbot’ or ‘AI companion’ (source)

Where they’re based: New York

Founded: 2016

Notable customers: Target, The Coca-Cola Company, BlackRock, Microsoft, Pfizer

Talkspace

What it is: Talkspace is a well-known US online therapy platform. In June 2026 it launched Tee, an AI mental health guide built on its own fine-tuned LLM, which it markets as a safer alternative to using general-purpose chatbots for mental health (source).

• It watches for 10 areas of risk in real time and can if needed escalate users to licensed clinicians (source)
• Talkspace is clear that Tee isn’t therapy and doesn’t diagnose. It’s intended to give support between sessions, or as a standalone companion
• In August 2026 Talkspace was bought by Universal Health Services (one of the biggest US hospital operators) for around $835m (source)

Where they’re based: New York

Founded: 2012

Notable customers: US health insurers, employers, schools and government bodies. So, available to a massive 200m+ people through those.

Mental Health AI Apps That Didn’t Make it

Youper

Youper was one of the better-known AI mental health apps. It mixed chat with mood tracking and CBT-style techniques, and had over 3 million users.

It shut down on 30 September 2026 and deleted all user data from 1 October (source). As far as I can tell, it simply ran out of money. Which I guess is a good reminder that an app is also a company, and companies can disappear along with your data.

Woebot

Woebot was one of the pioneers of CBT chatbots. It retired its consumer app in June 2025, and the company now only works with businesses (source).

What About ChatGPT, Claude and Other General Chatbots?

So far we’ve explored apps and companies specifically aiming for the mental health market. In reality, many people will also be using general LLM products like ChatGPT for support.

This is important to bear in mind when we look at statistics which are negative, as they are often about general chatbots rather than mental health specific.

So, if there’s mental health specific AI products available, why would anyone use a general one? I’m guessing here in the absence of data that the reasons would be:

• Convenience (don’t need to sign up for another product, don’t need to download yet another app)
• The products are already free or cheap
• Familiarity (and theerfore perceived trust) in what they already use

However, as ChatGPT etc. are general chatbot apps, they simply weren’t built for mental health support. They are designed to keep you engaged (e.g. ChatGPT isn’t going to tell you “you’ve been talking to me for four hours, go and get some sleep”).
They also lack clinical oversight and human escalation.

It’s because of this that I think mental health AI products are getting a bad reputation – i.e. by association.

And, for OpenAI and other big players, negative news stories means bad press and therefore impact to brand trust.

What Open AI Has Said

In 2025, OpenAI wrote about their principles and plans for how ChatGPT handles ‘sensitive conversations’ (source). Summarising what this said:

• They improved the model so it will help guide the user to getting help
• ChatGPT should “avoid affirming ungrounded beliefs that potentially relate to mental or emotional distress”
• And a rather vague “a model that more reliably responds well to users showing signs of psychosis, mania, thoughts of suicide and self-harm, or unhealthy emotional attachment to the model.” I’m cynical about this one as ‘more reliably responds’ doesn’t feel definitive or confident
• “These conversations are difficult to detect and measure given how rare they are..” – I don’t buy this. Whilst I can assume they are rare, surely a conversation that might involve subjects like self harm or suicidal ideation should be fairly easy to detect?
• OpenAI is playing on the idea that because the conversations are so ‘rare’ – 0.07% (source), it can be swept under the carpet. Now, I’m biased as I really really can’t stand Sam Altman, but, if we take 1 billion global monthly active app users (source) (and admittedly I don’t trust their stats or accounting) , that means 700,000 people are having those types of conversations.
• OpenAI has a ‘Global Physician Network’ which helps then inform their research and views.

Open AI Related Lawsuits

Meanwhile, OpenAI have had these to deal with:

• 2026, US: Eight states suing them, alleging ChatGPT is ‘harming children en masse’ (source)
• 2025, US: A couple in California suing over their son who took his life, allegedly after ChatGPT encouraging him to do so (source)
  • Pardon my language, but what fucking sickens me is that OpenAI “OpenAI requested from the family footage from Adam’s memorial services, a list of attendees at the services and a list of everyone who had supervised him in the past five years” (source)
• 2026, US: A man in Washington alleges ChatGPT use led to him being hospitalized (source)

What Anthropic Has Said

Anthropic also covered the topic in 2025 (source).

Give Anthropic their dues here – this article is a lot easier to digest than OpenAI’s, with some nice table and graph breakdowns. However, it’s more broadly about how people use Claude for ‘support’ rather than mental health specific. My takeaways are:

• ‘Interpersonal advice’ topic dominates over counselling or psychotherapy
• 2.7% is attributed to ‘Coping with workplace stress and professional mental health matters’
• Claude will rarely ‘push back’, and will only intervene if it thinks there is a genuine risk

How Mental Health AI Companies Position Themselves

OK, so given all of the above, Mental Health AI companies are positioned in a great way to sell their products as the ‘safe alternative’ to general LLM chatbots.

The proposition is – an app that’s specific to mental health support, which means less risk of the situations we saw with OpenAI.

Regulation/Legislation For AI Mental Health Products

What you’ll see on any of these apps and sites is that the companies make it explicility clear that their products isn’t therapy and doesn’t diagnose. The reason for that isn’t to be negative about the product, it’s mainly due to regulation.

In the US and UK the main principle is the same: if an app makes a claim that it can treat, manage or diagnose a health condition, it’s then considered a ‘medical device’. It then has to be regulated as one.

But, if the product is positioned as emotional support, wellbeing or stress management, it usually sits outside of that rule (even, in reality, people use it for those things).

In the UK

• Apps that make medical claims are regulated by the MHRA (the Medicines and Healthcare products Regulatory Agency). They need a UKCA or CE mark and should be on the MHRA public register (source)
• If a regulated app causes harm or distress, you (or a clinician) can report it through the MHRA’s Yellow Card scheme
• Mental health data falls under ‘special category’ data under UK GDPR, therefore any company handling it faces stricter rules
• The government has accepted all 44 recommendations from the National Commission into the Regulation of AI in Healthcare, so we can expect ongoing, tighter oversight (source)

It’s because of regulation that Ash left the UK in January 2026, as Slingshot was concerned it fell foul of those UK medical device rules.

In the US

The FDA uses the same diagnose/treat vs wellness split, and has historically left ‘general wellness’ apps alone (source). Some states have gone further than that. Illinois and Nevada now ban companies from offering AI-only therapy, so AI can help, but a licensed professional has to be in charge (source).

So what does that mean for you and I as potential users? A ‘wellness’ label doesn’t mean an app is unsafe. But it does mean no regulator has independently checked what it claims to do.

The Case In Support Of AI To Support Mental Health

I think we can all agree that in an ideal world, every person in any country would be able to easily and freely access mental health support.

However, stating the obvious, that isn’t the case – and it varies dramatically country to country, This can involve long waiting times, full hospital wards, costs, along with cultural issues like stigma. Just as one example, a report in 2026 stated there 500,000 children in the UK on mental health waiting lists. Half a fucking million – a figure that breaks my heart. (Source)

And – a stat I found during research that blows my mind – we have a global median of 13 mental-health workers per 100,000 people. (Source)

Similarly, in the US there is about 1,600 US patients with depression or anxiety for every available provider. (Source)

As someone who is in a blessed position and getting medication cheaply from the NHS here in the UK, along with financing ongoing therapy myself, it’s not something I take for granted. We also have services like the Samaritans for emergency support.

Something else we need to consider is that even in areas where mental health support is available, people aren’t always comfortable or confident asking for it. This might be due to worry about stigma, what will be involved, or feeling a lack of worth.

With all that in mind I think it’s only natural that people will look at alternatives – enter AI.

As I’m writing this (November 2026), AI products are still free or very cheap (it still blows my mind what something like Claude offers for 18 quid per month!). Whether that will stay like that in the future remains to be seen.
But, right now, most AI products are super cheap, or even free.

And of course, instantly accessible (minimal friction being creating an account, perhaps installing an app).

On top of all that is the fact you don’t have to talk to or communicate to another person.

So – at least in principle, AI products, and especially AI products for mental health, offer: ease of access, cheap or zero pricing, convenience, and avoidance of stigma.

Now, you might agree with what I said earlier about an ideal world being one where everyone could get support. So I ask you this – and I’m not sure of the answer myself – is it a good thing that people can at least get some support?

Now let’s look at some evidence to support the bullish view:

The application of artificial intelligence in the field of mental health: a systematic review (2025)

This study was small scale but highlighted these positive takeaways:

• In the Wysa app study, users who engaged more saw bigger improvements in self-reported depression, and 67.7% said it helped
• Young people were generally positive about AI-informed mental health apps, and mHealth use during COVID was linked to better wellbeing
• Deep-learning models (LSTM and CNN) recognised emotions and assessed mental state from forum or social media text
• AI shows promise for early detection, personalised interventions, remote monitoring and wider access, especially for underserved groups

Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being (2023)

In my opinion this study should be taken with caution given it was done back in 2023. Whilst that seems recent, given how AI technology has advanced since then, the products now available are very different. With that said:

• The chatbots significantly reduced symptoms (moderately)
• Moderate to large reduction in psychological distress
• Anxiety, stress, positive mood and negative mood showed no significant effect
• Voice or multimodal chatbots beat text-only ones
• People with clinical or near-clinical symptoms gained much more than the general population

First Therapy Chatbot Trial Yields Mental Health Benefits

This research from 2025 was the first clinical trial of an AI chatbot for therapy. (Interestingly, Therabot, the product the study was on, actually started development in 2019 – I’d always assumed most were post-ChatGPT.) The main points to consider in terms of symptom reduction:

• Depression: −51%, with clinically significant gains in mood and wellbeing
• Anxiety: −31%. Many moved from moderate to mild anxiety, or dropped below the diagnostic threshold.
• Eating-disorder risk: −19% in body image and weight concerns, significantly better than the control group

The Case Against AI To Support Mental Health

I know from conversations that many people see AI for mental health as quite dystopian, and I can get why. The thought of someone in mental and emotional distress, perhaps alone at night typing or talking into a glowing screen which is just a front for an LLM………well, yeah, it’s not nice.

We also need to consider the inherent flaws of LLMs – hallucinations, the inability to think and feel, and also how LLM conversations never end (unless the user ends it). It’s so easy to ‘rabbit hole’ with any LLM conversation and I can totally understand why someone suffering mentally would spend way too long in a ‘conversation’.

As a wider point around health in general, there have been various concerns raised about LLMs misdiagnosing, and exacerbating existing issues.

“AI models reflect the knowledge and beliefs on which they are trained, biases and all,” said Dr. Schabacker. “If healthcare stakeholders are not careful, AI could further entrench the disparities that many have worked for decades to eliminate from health systems.” – Dr. Marcus Schabacker

More worryingly have been times like in 2026 where a boy in England died after ChatGPT gave him advice about how to take his life.

And, I think if we look at how the sentiment to AI in general has changed during 2026, I think there is more general negativity to it from both the public and mainstream press. Even the throwaway insult ‘AI slop’ has become more popular which shows a distrust and dislike of some of its abilities.

So the argument against AI for mental health is the possible perfect storm of bad advice, bad information, isolation and lack of human contact.

Now let’s look at some of the evidence:

• Spring Health’s VERA-MH findings on gaps in how chatbots respond to suicidal ideation – a mental health company flagging the risk itself (source)
• In the ECRI Top 10 Health Technology Hazards For 2026, the ‘Misuse of AI Chatbots in Healthcare’ was #1 (source)
• The quality of the evidence itself: lots of it is small, short-term, self-reported, or produced by the company selling the product (Sonia’s trials are still preprints; STAT questioned how much clinical proof Ash’s own safety study offers – find that STAT piece, it’s not the one linked in the Ash profile)
• Business risk as a harm in itself: For example, Youper shutting down and deleting data, and Earkick’s ghost-town presence. Imagine you’re a user and you’ve become used to the product and suddenly it no longer exists.

How To Choose An AI Mental Health App: A Quick Checklist

If you’re wanting to choose an AI mental health app, my advice would be to consider these factors:

• Who is it for? Is it for a specific age group or level of need? Conversely, who is it not for?
• Your data: Is your data going to be used to train models? Where is it stored? Can you delete it or export it?
• Evidence: Does it have studies which are peer-reviewed, or just numbers the company has provider?
• Is It Legit: Does the company website have a contact page, recent updates, and is it clear who’s behind it?

What’s Next For AI In Mental Health?

The next year or so is going to be pivotal for this space. There is ever more negative sentiment in the press about AI generally, and the talk of the ‘AI bubble’ bursting.

If that were to happen then there’s potential to impact pricing for these products, as well as more risk of buyer scepticism.

I’d expect to see more consolidation, where big players in health buy out high performing AI mental health companies, in conjunction with smaller ones closing down (which could be due to money, regulation or both).

I expect to see tighter and more up to date regulation, especially in the UK and US. I think this is good for consumers but will make things harder for smaller players in this niche.

My Own Views On AI For Mental Health

Over the last couple of years I’ve really flip-flopped with my opinions about this. If I had to sum up how I feel now it’s:

• In an ideal world, no-one would need AI for mental health support
• We don’t live in an ideal world, and there are SO many people lacking even basic mental health support
• General LLM chatbots are OK for superficial support but are never going to be best placed for mental health support
• There are some excellent AI based mental health solutions available now, and I expect as models improve, for all of them to get better and better
• I respect how all are very clear about what they are or aren’t for
• People behind these products seem genuinely passionate about what they’re doing
• Whilst no app will be perfect, I think AI mental health ones are a great solution when appropriate
• As a society, we should be doing whatever we can do reduce things like self harm and suicide – I believe AI products play a part in that

How I’ve Used AI To Help My Mental Health

I mentioned earlier how I’m fortunate to be able to access counselling and coaching. Outside of that I’ve found various uses for AI to improve my mental health, even if indirectly:

• Managing workload: I use Claude a lot to ‘voice dump’ everything that’s on my mind or need to do. AI is incredibly good at synthesising and structuring it
• Rationalising: AI is very good at helping you ‘talk through’ things. So let’s say something is bothering me, I can get Claude to offer alternative opinions, or to talk through whether something I’m worried about is even realistic
• Automation: Indirectly, automating things at work or in my personal life reduces the cognitive load I have – in turn, reducing stress and anxiety.

Disclosure – Use Of AI For This Article

• Claude was used for:
  • Some of the research
  • Summarising regulation
  • Proof reading
• Everything else – me