Quick summary: 1 in 6 US adults has already discussed mental health with an AI chatbot and 35% would consider doing so instead of seeing a professional, showing that people turn to these tools because they are immediate and carry less social discomfort. Clinicians therefore cannot assume a consultation starts from a blank page, as patients may arrive with chatbot advice that feels coherent and tailored yet falls short of a proper clinical assessment and sits outside usual privacy protections. Mental health practice needs to ask about that AI use, treat it as information to examine rather than an invisible influence, and help people move towards qualified care rather than away from it.
About 1 in 6 US adults has used an artificial intelligence chatbot to discuss their mental health, according to the American Psychiatric Association’s 2026 annual poll. That figure should prompt a more useful conversation than whether AI will replace therapists. People are already turning to chatbots when they want to talk about their mental health. The more immediate question is what happens when the advice they receive follows them into professional care.
The poll, conducted by Morning Consult among 2,201 adults from 13th–15th April, found that 17% had discussed mental health with a chatbot. More strikingly, 35% said they would consider talking to a chatbot instead of a mental health professional if they were struggling. That does not mean 35% of adults intend to abandon professional care. It measures willingness, not actual substitution. The poll’s reported margin of error was two percentage points.
Still, the finding is difficult to dismiss. A chatbot is available immediately, does not require an appointment and does not carry the social discomfort that can come with telling another person about anxiety, depression, loneliness or thoughts that are difficult to articulate. For someone who is awake at 2am and wants an answer now, those characteristics matter.
There is also evidence that AI conversations are beginning to enter clinical practice indirectly. A separate American Psychological Association survey of more than 1,200 practising US psychologists found that 39% reported having patients who used AI to diagnose themselves. That statistic needs to be read carefully. These are psychologists reporting about their patients. It does not mean that 39% of all patients, or 39% of the wider public, have used AI for self-diagnosis.
What it does suggest is that clinicians cannot assume that the consultation begins with a blank page. A patient may already have asked a chatbot what their symptoms mean, whether they have a particular disorder, what medication they should take, or whether something they are experiencing is normal.
The problem is that a convincing answer is not necessarily a sound clinical assessment.
A chatbot can produce a response that feels attentive, coherent and reassuring without having the information required to make a proper assessment. It may also fail to recognise when reassurance is inappropriate, when symptoms require further investigation, or when a person needs urgent help. In its consumer guidance, the Psychological Association cautions that AI can produce inaccurate advice and should not replace a licensed mental health professional.
This is where the debate around AI and mental health can become too simplistic. The question is not whether every conversation with a chatbot is dangerous, any more than every Google search about a headache is a medical emergency. People have always looked for information before speaking to professionals. AI simply makes that process more conversational, personalised and persuasive.
That last characteristic deserves particular attention. Search results present information. A chatbot can appear to engage with you. It can remember the details provided within a conversation, respond to what you have just said and generate language that sounds remarkably tailored to your circumstances. That can make its output feel more authoritative than it actually is.
Privacy introduces another problem.
The US Department of Health and Human Services explains that information entered into personal health apps is often outside HIPAA protections unless the app is provided by a covered organisation or its business associate. Entering health information into an app does not, by itself, give that conversation the same protections as a medical record.
This is easy to overlook because the interface may look private. A person can sit alone with a phone and disclose deeply personal information without considering where that information goes afterwards. Before sharing sensitive information, users should check whether conversations are retained, reviewed, used for training, or shared with other organisations. A reassuring interface is not a privacy policy.
Removing a person’s name does not necessarily make a conversation anonymous either. A detailed account of someone’s relationships, employment, family circumstances, location and experiences can contain enough information to make that person identifiable.
There is nevertheless a more constructive way to approach the issue than simply telling people not to use AI.
The Psychological Association suggests that AI may help people organise their thoughts or prepare questions before an appointment. That is a considerably different role from asking a chatbot to determine a diagnosis or provide treatment. Someone who cannot quite explain what they have been experiencing might use a chatbot to organise their thoughts before speaking to a psychologist, psychiatrist or other qualified professional.
Clinicians should also know when AI has been involved. Telling a professional that a chatbot suggested a particular diagnosis or gave specific advice allows that information to be considered alongside the person’s history, symptoms and circumstances. The chatbot’s output can then become something to examine rather than an invisible influence on the consultation.
For those considering these tools, SLS Psychiatry’s educational guide to AI therapy discusses possible uses, limitations and privacy questions. The Texas practice provides adult psychiatric care. Its resource is patient education, rather than evidence that any particular chatbot is effective, and it should be read alongside independent research and professional guidance.
There is an important limitation to what the current survey findings can tell us. They show that people are using these systems and that many are open to doing so. They do not establish whether chatbots improve symptoms, worsen them, encourage people to seek help or cause them to delay it. Nor do they show that one chatbot performs like another.
That uncertainty should not prevent clinicians from asking about AI use. In fact, the opposite is probably true.
If patients are already bringing advice from chatbots into their lives, asking about that advice gives clinicians an opportunity to see what information people are actually receiving outside the consultation. It may reveal misconceptions, inappropriate reassurance, useful questions that the patient had struggled to articulate, or advice that needs correcting.
The technology does not need to be treated as either a miracle or a menace. The more practical approach is to recognise what is already happening. People are using AI to talk about their mental health, and some are considering it an alternative to professional care.
The task for mental health professionals is therefore not simply to tell people that chatbots are not therapists. It is to understand how these tools are being used, what people believe about them, and whether they are helping people move towards appropriate care or giving them reasons to stay away from it.
That conversation is likely to become increasingly important as AI becomes a routine part of how people seek information, advice and support.
Dr Rana Kaleemullah, MD, is a board-certified psychiatrist and founder of SLS Psychiatry in Southlake, Texas. She provides adult evaluations and medication management in person and by telehealth across Texas and is known as Dr K.
