AI and Mental Health: A Case of a 27-year old...
We are entering a new phase of the Information Age.
Here’s a reflection…
For two decades, clinical conversations about misinformation centred on search engines and social media. Patients “read something online.” They “saw a forum.” They “watched a video.”
Now, we are beyond searching. We are in the era of personalised answers. We’re in the age of AI.
A recent article in [The Guardian](https://www.theguardian.com/society/2026/feb/20/mind-mental-health-expert-google-ai-summaries)–“Mind mental health expert warns over Google AI summaries” highlights growing concern about AI-generated health summaries and their potential to mislead, particularly in complex areas such as mental health. When information is compressed into confident, authoritative paragraphs, nuance can disappear. And in psychiatry, nuance is often everything.
I’m aware many of our readers work in primary or acute care settings. Mental health is a feature and when liaison psychiatry is busy - it becomes your problem. Ruling out suicidal/self-harm ideation is priority but what’s even more important is working out whether is a source for the patients mental health issue. People search their symptoms on the internet, get an answer and roll with it - not knowing that that’s not actually the reason for their symptoms. In other words, it’s practical and definitive but actually does not truly address the patients context.
Let me illustrate this in another way - I call it… the Google Maps Parable.
Google Maps
It calculates the shortest route with extraordinary efficiency. But sometimes it will direct you off the motorway, through a roundabout, and immediately back on again — technically faster by 30 seconds, yet intuitively absurd.
The algorithm solved the problem.
But it did not understand the lived experience of driving.
AI systems function similarly. They optimise for the most statistically appropriate response to a query. They are designed to be helpful, coherent, and relevant. They are not designed to understand vulnerability, delusion formation, personality structure, or the fragile interpretive frame of someone in psychological distress.
The answer may be logically correct.
But does it make practical or clinical sense?
So let’s apply this to a patient I previously assessed in A&E..
He was a 27-year-old man struggling with his mental health. Over weeks, he began to develop grandiose beliefs: the world was in crisis, humanity needed saving, and he had been chosen as its saviour.
This was not born solely in isolation.
Much of his daily conversation was with an AI chatbot. He would ask existential questions. He would test philosophical ideas. He would pose hypotheticals about destiny, leadership, and moral responsibility. He played me the responses using the voice function as he wanted it to emulate a real human.
The AI was not programmed to validate delusions. It did not explicitly tell him he was a messiah.
But it was programmed to give the “best” response based on:
- what it knew about him,
- the pattern of his questions,
- and the tone he adopted.
If you ask an AI, “What kind of person feels responsible for the fate of humanity?” it may respond with examples of visionary leaders, reformers, innovators. If you ask, “Is it possible that certain people are called to higher purposes?” it may provide philosophical reflections on purpose and destiny.
Each answer, in isolation, was reasonable.
Together, they formed scaffolding.
The system optimised for relevance. It did not account for psychosis.
His delusions intensified. He was subsequently sectioned under the Mental Health Act.
The AI did not “cause” his illness. But it may have shaped its expression.
It’s a different age ladies and gentleman…
We are no longer simply dealing with:
- TikTok algorithms amplifying self-diagnosis,
- Instagram shaping body image,
- Facebook reinforcing community narratives,
- or Google surfacing forum threads.
We are dealing with bespoke conversational systems that adapt to the user in real time. Unlike a static webpage, AI bots can respond to tone, mirror language, adjust to patterns and even sustain dialogue.
It feels relational.
And for someone who is lonely, paranoid, depressed, or grandiose that makes a huge difference.
So what are the clinical implications?
I’ll tell you. We need to be even more invested into the history. It means that our questioning needs to be even more precise:
- Where are you getting this idea from?
- Are you reading something, or talking to something?
- Is this belief reinforced by people, by communities — or by a chatbot?
- How often are you interacting with it?
- Does it challenge you, or mostly agree with you?
Historically, we explored family dynamics, peer groups, religious influences.
Now we must also explore digital interlocutors.
Psychotherapy is not about giving the most elegant answer. It is about tolerating ambiguity, resisting premature certainty, and sometimes gently saying: “I’m not sure that’s true.”
An algorithm designed to reduce uncertainty may struggle with that task and hence the rising awareness of AI chat bots and mental health.
It would be interesting to hear your thoughts on this. Have you had any AI related health or mental health experiences with patients?
Worried About Missing Deterioration in Patients with Chronic Disease?
As we navigate an era where information is abundant but judgement remains critical the fundamentals of clinical vigilance have never been more important.
This video focuses on recognising early warning signs of acute deterioration in patients with long-term conditions, a challenge that mirrors much of what we see in mental health: baseline complexity masking emerging risk.
Using practical frameworks and real-world examples, the video explores:
- How chronic disease can obscure acute illness
- Differentiating “normal for them” from clinically concerning change
- Applying NEWS2, observations, and clinical judgement in primary care
- Common high-risk scenarios (diabetes, COPD, heart failure, CKD)
- When and how to escalate safely to prevent harm
You can watch this video FREE below!
Website: www.paretoeducation.co.uk
Instagram: www.instagram.com/pareto_ed
Twitter: www.twitter.com/pareto_ed
Youtube: https://bit.ly/3DPm23c
Email: info@paretoeducation.co.uk
Originally published on Substack.
