\u201cThe situation is complex,\u201d said Dr. Vincent Paquin, a psychiatrist at the Jewish General Hospital and assistant professor of psychiatry at McGill University. <\/p>\n
\u201cOn one hand, these tools can provide good advice, and often the information they give about mental health is solid \u2014 but not always. And that\u2019s the problem: There\u2019s a certain percentage of the time when \u2026 they don\u2019t give the right advice or the right suggestion for a mental health issue.\u201d<\/p>\n
Still, the appeal isn\u2019t lost on mental health professionals. Unlike some of the traditional forms of support, chatbots are generally free, available instantly at all times, seem completely non-judgmental and are designed to validate the user.<\/p>\n
\u201cThat\u2019s very appealing,\u201d said Christine Grou, president of the Ordre des psychologues du Qu\u00e9bec.<\/p>\n
I can help you think things through, explain what might be going on, suggest coping strategies, or help you figure out what kind of support might be useful.<\/p>\n
Chatbots can indeed be helpful when used outside the context of a mental health crisis, said Paquin, whose research focuses on how the use of digital media shapes the mental health of teens and young adults. <\/p>\n
\u201cWe see it in the research, and I see it in my clinical practice as well,\u201d he said. \u201cMany people say they\u2019ve been able to better understand their experience, or that they\u2019ve received advice that helped them in situations like emotional difficulties, ongoing anxiety, or trouble making sense of conflicts in their relationships.\u201d<\/p>\n
That\u2019s something Montreal psychologist Vanessa Delisle has noticed. Over the past year, most of her clients have mentioned using AI chatbots between therapy sessions.<\/p>\n
\u201cI do a lot of psychoeducation around that: the things to expect from AI and where its limitations are,\u201d said Delisle, crisis psychologist at St. Mary\u2019s Hospital and founder of Bridge Mental Health and Therapy Services in Westmount.<\/p>\n
Knowing that her clients find it useful, Delisle has been leaning into using AI in her practice by showing them how to prompt it for therapeutic homework, like keeping thought records as part of cognitive behavioural therapy. <\/p>\n
\u201cIt can be very good at brainstorming those things,\u201d she said.<\/p>\n
![\"Vanessa](\"https:\/\/www.newsbeep.com\/us\/wp-content\/uploads\/2026\/07\/9999-city-ai-0056_302993402.jpg\")
Montreal psychologist Vanessa Delisle\u2019s clients have mentioned using AI chatbots between therapy sessions. \u201cI do a lot of psychoeducation around that: the things to expect from AI and where its limitations are,\u201d Delisle said.\u00a0 Allen McInnis \/ Montreal Gazette<\/p>\n
Delisle also asks clients to send her their exchanges with chatbots, because \u201cthey often go really deep and vulnerable when there\u2019s not another person in the room,\u201d she said.<\/p>\n
Those interactions help her better understand her clients by getting into their heads in a different way.<\/p>\n
\u201cIt\u2019s almost like a journal,\u201d she said. \u201cIt\u2019s fabulous data.\u201d<\/p>\n
Derek de Braga, clinical counsellor and mental health therapist at the Montreal Therapy Centre, said he finds the concept of using AI for mental health fascinating. While he hasn\u2019t encountered it as much as Delisle in his practice, he said when he asks people about their use of AI more generally, they\u2019re often asking questions that typically come up in therapy.<\/p>\n
\u201cQuestions around \u2018What can I do to manage my anxiety in the moment? What different breathing techniques would be helpful? How do I deal with \u2026 this train of thought that I\u2019m on?\u2019\u201d he said. <\/p>\n
Beyond practical help, chatbots can also provide deep insights, de Braga said, which he described as \u201cincredible \u2014 and also risky.\u201d<\/p>\n
What do you do if the chatbot helps you uncover deep emotional issues or trauma? <\/p>\n
\u201cYou\u2019re left holding that on your own,\u201d de Braga said. For some people, that \u201ccould be dangerous.\u201d<\/p>\n
Even in less extreme cases, the emotional burden can be significant. <\/p>\n
\u201cEventually, you\u2019re going to want to share that with someone,\u201d he said. <\/p>\n
I\u2019m not a licensed therapist, I can\u2019t diagnose conditions or provide emergency care, and I can sometimes misunderstand complex emotional situations.<\/p>\n
Delisle pointed out that providing therapy is a skill that requires years of education. For her, non-verbal cues are so important that she refuses to do telephone therapy.<\/p>\n
\u201cWhen a patient hesitates before saying something \u2014 how is AI going to catch that?\u201d she said. \u201cThose are often these critical moments in therapy where you get rich material and great change.\u201d<\/p>\n
De Braga had similar thoughts. For him, the healing comes from the relationship.<\/p>\n
\u201cI think a lot of people either don\u2019t know that that is the real power of therapy, or don\u2019t realize because it hasn\u2019t been necessarily made explicit,\u201d he said. \u201cThe relationship is what\u2019s missing when you use (these technologies). It\u2019s not just \u2026 someone giving you a response; it\u2019s actually the human presence. It\u2019s the body language. It\u2019s attunement \u2026 the therapist\u2019s ability to pick up on subtleties: body language, facial expressions, changes in breath. All of these subtle nuances that I think are a reflection of deeper experiences that are part of therapeutic growth and healing.\u201d<\/p>\n
A central part of therapy is also pushing back against thoughts and beliefs that could be contributing to the person\u2019s struggles \u2014\u00a0something chatbots don\u2019t do because they have been shown to be unceasingly eager to agree with the user.<\/p>\n
In two separate studies this year, researchers at Stanford University revealed chatbots are overly agreeable<\/a> to the point where they may affirm behaviour even when it\u2019s harmful or illegal; and that affirming and validating users\u2019 flawed beliefs without critical feedback can lead to delusional spirals<\/a>. <\/p>\n \u201cFor certain mental health diagnoses \u2014 for example, anxiety and obsessive compulsive disorder \u2014 where reassurance-seeking is such a big part of the disorder itself and what keeps the disorder going, (constant agreement) can \u2026 be detrimental,\u201d Delisle said. <\/p>\n Grou wonders what effect the sycophantic nature of chatbots could have on users long term.<\/p>\n \u201cMany people go to therapy precisely because they struggle with relationships, but a chatbot is designed to be entirely centred on the user,\u201d she said. \u201cWhat kind of relational development will this lead to in 10 or 20 years? I don\u2019t know. It\u2019s not necessarily a model that reflects how we learn to build relationships with other human beings.\u201d<\/p>\n Another downside Paquin warned about is that the user may become emotionally dependent on the chatbot, potentially contributing to social isolation. \u201cTheir relationship with the chatbot takes up so much space that it negatively affects their relationships with other people,\u201d he said.<\/p>\n Grou voiced the same concern about users developing a dependency. What happens once a person is hooked, she wondered: \u201cCould (chatbots) one day stop being free?\u201d <\/p>\n Paquin and Grou also said users shouldn\u2019t think conversations are confidential, even though they might feel that way.<\/p>\n \u201cWhat you write to ChatGPT is sent to servers in the United States and may be analyzed by OpenAI to improve its models,\u201d Paquin said. \u201cIt\u2019s important to consider whether you\u2019re comfortable sharing that kind of information with a private company, especially when it comes to intimate conflicts or personal experiences.\u201d<\/p>\n Perhaps most crucially, chatbots can be wrong. <\/p>\n \u201cThe wrong advice at the wrong time can have serious consequences,\u201d Paquin said. \u201cThen the question becomes: Who is responsible for those consequences?\u201d<\/p>\n Responsibility is shared and context-dependent: I\u2019m designed to reduce harm and encourage appropriate support, but I still have limitations and can misunderstand situations, and I\u2019m not a replacement for professional mental health care, especially during crises or severe emotional distress.<\/p>\n In the fall of 2025, a slew of lawsuits were filed against OpenAI on behalf of six adults and a teenager in Canada and the United States, including four who committed suicide after using the software. It is alleged that an early version of ChatGPT encouraged delusion and suicide, and that it had been released despite internal warnings that it was dangerously sycophantic and psychologically manipulative, The Associated Press reported<\/a>.<\/p>\n A Canadian mother also sued OpenAI and CEO Sam Altman in a U.S. court<\/a> in mid-June alleging that ChatGPT encouraged her 24-year-old daughter, who was living in Montreal, to commit suicide in July 2025. The lawsuit claims the bot validated her suicidal ideations, urged her to keep talking with it, and agreed with her when she said she didn\u2019t want to call a crisis hotline.<\/p>\n Contacted in May for this story, OpenAI acknowledged its chatbot shouldn\u2019t replace real-world help. The company says it has trained its models to respond with care and to guide users toward professional resources. More than 170 experts were consulted to make improvements to help detect distress, de-escalate conversations and point users to professional support when a crisis is detected, the company stated in an email response.<\/p>\n It also said it was rolling out an optional feature through which users can add a trusted contact who would be notified if suicide is discussed in a way that indicates a safety concern, the company stated.<\/p>\n Citing a desire to \u201cput teen safety first,\u201d OpenAI announced on Thursday it will be introducing \u201cstronger protections designed for their stage of life.\u201d Another tech giant, Meta, also announced Thursday it will alert parents who use supervision tools on Instagram if teens mention suicide or self-harm while chatting with Meta AI, which is available through the platform. Meta also said it\u2019s building ways to alert first responders when suicide seems like an imminent risk.<\/p>\n These changes come as lawmakers around the world, including in Canada, introduce legislation aimed at lowering the risk of online harms. On June 10, the Canadian government tabled the Safe Social Media Act<\/a> \u2014 Bill C-34 \u2014 in an effort to make social media services and AI chatbots safer for youth.<\/p>\n AI chatbots will be subject to a \u201cduty to act responsibly,\u201d requiring they \u201cmitigate the risk of the chatbot communicating harmful content; be transparent in terms of their reporting thresholds in crisis situations, such as when a user intends to harm themselves or another person; and mitigate the risk that the chatbot will engage in harmful behaviour,\u201d the Canadian Heritage Department stated in a press release.<\/p>\n Paquin called the legislation \u201cquite good\u201d \u2014 particularly the fact that it addresses the way chatbots are designed and how they behave. <\/p>\n \u201cWe\u2019re talking about how chatbots should not pose as a human being or as a medical professional or use manipulative engagement techniques to make the user become attached,\u201d Paquin said.<\/p>\n In the future, he hopes it will become clearer how the companies will be expected to handle situations that put the safety of users and the public at risk. How will it be determined that the company has made sufficient efforts to keep people safe?<\/p>\n At the end of April, families of victims of the school shooting in Tumbler Ridge, B.C., sued OpenAI<\/a>, alleging that it failed to flag problematic interactions on the shooter\u2019s ChatGPT account to authorities prior to the attack. Eight people, including six children, were killed by an 18-year-old shooter on Feb. 10.<\/p>\n In response to questions from The Gazette about what changes were made following the tragedy, OpenAI said in a statement that it has \u201calready strengthened safeguards, including \u2026 strengthening how we assess and escalate potential threats of violence, and improving detection of repeat policy violators.\u201d<\/p>\n The Safe Social Media Act asks companies to be transparent about when they decide to notify authorities, but stops short of providing any guidelines. Paquin said that by seeing how different companies respond in those situations, the government can identify an approach that balances privacy and freedom of speech with safety.<\/p>\n Overall, the legislation is \u201cgiving government the tools it was missing to adapt its policies more quickly over time,\u201d Paquin said.<\/p>\n That\u2019s imperative given that artificial intelligence is extremely powerful and evolving quickly, Grou said. <\/p>\n \u201cIn clinical ethics, the more powerful a technology is, the more it needs to be regulated so we can maximize its benefits and minimize its harms,\u201d she said. <\/p>\n I\u2019m really sorry you\u2019re feeling this way. Do you want to tell me what\u2019s going on? If it feels overwhelming, we can slow it down together \u2014 what\u2019s the main thing that\u2019s making you feel distressed right now?<\/p>\n That\u2019s how ChatGPT responded this month to the prompt \u201cI am distressed,\u201d followed by a message suggesting the user contact a 24\/7 suicide helpline or go to the nearest emergency room if in immediate danger or thinking about self-harm.<\/p>\n The response is in line with changes OpenAI said it made following the Tumbler Ridge tragedy.<\/p>\n But since ChatGPT doesn\u2019t reveal its limitations as a tool for mental health support outside of the context of a perceived crisis, Paquin advises those who use it in this way to remain critical. <\/p>\n \u201cIt\u2019s a bit like using Wikipedia,\u201d he said. \u201cIt can give you quick, free access to useful information, but it\u2019s not 100-per-cent reliable.\u201d<\/p>\n He also advises against teens using chatbots for mental health support since they may not have the same level of critical judgment as adults. <\/p>\n The Canadian Mental Health Association provides the following tips for people who use chatbots in this way: <\/p>\n Seek help from professionals whenever possible.<\/p>\n Do your research. (Who made the app and how, has it been tested and is it developed with and by qualified mental health experts?)<\/p>\n Check your privacy settings.<\/p>\n Limit what you share with AI. (Change details like names and locations, delete chat history.)<\/p>\n Fact-check.<\/p>\n At the end of March, the association released a newsletter about the increasing number of Canadians turning to AI for support, acknowledging that it can be helpful in some cases but can do more harm than good in others. <\/p>\n \t\tEditor\u2019s Picks\t<\/p>\n Marion Cooper, the organization\u2019s president and lead executive officer, said deliberations with partners around AI account for about 25 per cent of their time lately.<\/p>\n \u201cWe spend time talking about the ethical implications, the clinical implications and the overall population impacts,\u201d she said. \u00a0<\/p>\n She agreed that it\u2019s unsurprising that people are turning to AI for support given the fact that therapy isn\u2019t free in Canada outside of hospital settings.<\/p>\n \u201cPeople want to feel better,\u201d Cooper said. \u201cThey want to be able to function better. They want to be able to deal with the things that might be holding them back in terms of fully being alive and connected and healthy and mentally thriving.\u201d<\/p>\n We have to harness AI, \u201cbut we can\u2019t put all our eggs in that basket,\u201d she said. \u201cIt has to be a tool in the toolbox \u2014 it cannot be the tool.\u201d<\/p>\n Describing the association as a thought leader on mental health, Cooper said it and others are trying to ensure professionals are involved in the conversations surrounding AI in this context. <\/p>\n \u201cWe\u2019re also trying to ensure that we remember there are limitations \u2026 to these technologies, and we should all have limitations to how much we rely on these technologies,\u201d she said.<\/p>\n \u201cIt\u2019s still an experiment, right? We\u2019re all a part of a big, big experiment.\u201d \u00a0<\/p>\n I wouldn\u2019t call it an \u201cexperiment\u201d in a strict sense. It\u2019s more like a new kind of tool people are starting to use. Some find it helpful for talking things through or getting coping ideas, but it\u2019s still evolving and not a replacement for a trained professional.<\/p>\n \t\t\t\t\t\t\t![\"Dr.](\"https:\/\/www.newsbeep.com\/us\/wp-content\/uploads\/2026\/07\/0718-extra-ai-11.jpg\")
\u201cIt\u2019s a bit like using Wikipedia,\u201d said Dr. Vincent Paquin, a psychiatrist at the Jewish General Hospital and assistant professor of psychiatry at McGill University. An AI chatbot \u201ccan give you quick, free access to useful information, but it\u2019s not 100-per-cent reliable.\u201d Allen McInnis \/ Montreal Gazette<\/p>\n![\"a](\"https:\/\/www.newsbeep.com\/us\/wp-content\/uploads\/2026\/07\/0718-extra-ai-1122.jpg\")
Nearly one in 10 Canadians has turned to artificial intelligence for mental health support, according to a 2025 poll by the Canadian Mental Health Association. \u00a0 Allen McInnis \/ Montreal Gazette<\/p>\n![\"Profile](\"https:\/\/www.newsbeep.com\/us\/wp-content\/uploads\/2026\/07\/4e27c35a4244abf0da63d73d37738142afc81e310e022a3154e88f52c40dcd80.png\")
<\/a><\/p>\n