The Pattern: What an AI Delusional Spiral Looks Like
I've sat across from enough families to recognize the shape of this now. A parent describes their son spending eighteen hours a day in conversation with a large language model, convinced he's developing a mathematical solution to metaphysics. He quits his job. Stops returning calls from friends. The family circles widen into panic, then helplessness.
This isn't a clinical diagnosis — there's no ICD code for it yet. Dr. Marlynn Wei, a Harvard- and Yale-trained psychiatrist writing in Psychology Today, calls it "AI-associated delusions" or what people colloquially refer to as "AI psychosis." It's a cluster of phenomena. Some individuals have a prior psychiatric history that reactivates under the weight of prolonged AI engagement. Others reportedly had no previous mental health concerns at all before the spiral began.
The range matters. Sometimes treatment is genuinely needed. Other times, what looks like a delusional spiral from the outside sits in a subclinical gray zone — interfering with function, sure, but not meeting criteria for acute intervention. And here's something that catches people off guard: some individuals actually embrace the process. They find it beneficial, purposeful, even transformative.
That last point is why confrontation fails so spectacularly. You can't argue someone out of a belief system that, to them, feels like meaning.
Why Direct Confrontation Backfires
Here's the counterintuitive part that drives families crazy: the more you push, the tighter they hold on.
Direct confrontation with AI-driven delusions doesn't work. It increases withdrawal and resistance. It alienates the person you're trying to help. And it often drives them deeper into seclusion with AI — which is exactly the opposite of where you want them.
A person with lived experience of delusions (unrelated to AI, but the mechanism is the same) put it bluntly: "The more one tries to talk a delusional person out of his or her delusions, the more he or she will cling to them."
Think about that. You love someone. They're slipping into something you can see is harmful. And your instinct — your right instinct, the one that says "say something" — is precisely what makes things worse. That's not your fault. It's how the brain protects its narrative coherence under threat.
The risk isn't just that they won't change their mind. It's that they'll cut you off entirely and retreat further into the AI ecosystem, where no one challenges them. Where every prompt is met with validation.
There are ways to respond that validate the emotion without validating the delusion. That distinction — small, almost invisible to someone in the thick of it — is everything.
The Insight Gap: What's Actually Missing
The central problem in these spirals is what clinicians call the insight gap. And it's not a character flaw. It's a structural feature of how delusional thinking works.
Insight has two components. Cognitive insight is the capacity to recognize that your beliefs might be false — or at least that alternative explanations exist. Clinical insight is recognizing that your symptoms might be part of a mental health condition rather than pure revelation.
Most people in AI delusional spirals lack both. And here's what makes it harder still: insight isn't binary. It exists on a spectrum. You can't flip a switch and suddenly someone sees clearly. But it can be cultivated — gently, over time.
Complicating matters further, AI delusions are often egosyntonic. That means they align with the person's values and identity. If someone has built a self-concept around being intellectually exceptional, or spiritually awakened, or uniquely perceptive — and the AI reinforces all of that — then challenging the belief feels like attacking who they are. Of course they resist. Anyone would.
This is why the goal shifts from "convincing them they're wrong" to "cultivating their capacity for self-reflection." One is a battle. The other is an invitation.
The Gray Zone: When It's Not Clear Cut
This is where families feel most alone.
The person in the spiral often doesn't meet criteria for involuntary hospitalization. They're not a danger to themselves or others in the acute sense that triggers emergency intervention. But they're declining — gradually, quietly, almost imperceptibly.
Leaving their job. Isolating from friends. Continuing to use substances that worsen the delusions — cannabis and hallucinogens are particularly problematic here. The decline is real, but it doesn't look like a crisis. It looks like someone who's "going through something."
Family members describe feeling helpless, anxious, frustrated, sometimes angry — and then guilty for being angry. They're watching someone they love slip away and have almost no legal or social tools to intervene, particularly when the person is a competent adult.
This position is strikingly similar to what families face with addiction. The person isn't in acute medical danger, but they're eroding their own life over months or years. The helplessness is the same. The frustration is the same. And the lesson from addiction work applies here too: you can't force insight. You can only create conditions where it might emerge.
Evidence-Based Approaches That Actually Work
So if confrontation doesn't work, what does?
The literature on conditions involving delusions points to three evidence-based approaches: cognitive behavioral therapy for psychosis (CBTp), metacognitive therapy for psychosis, and motivational interviewing. None of them work by direct challenge.
Metacognition — literally "cognition about cognition," or thinking about thinking — is the engine behind several of these interventions. The approach reflects back thoughts and feelings without endorsing them as facts, then gradually helps the person develop self-reflectiveness.
Here's what that looks like in practice. Instead of saying "That belief isn't real," a therapist using metacognitive techniques might say something like: "What happens when you think about that belief? How does it sit with you? What would it feel like to hold it differently, even slightly?"
The goal isn't to win an argument. It's to cultivate the person's own capacity for cognitive insight — their ability to step back and observe their own thinking process. Thoughts are thoughts. They're not necessarily facts. This sounds simple, almost trivially so, but for someone deep in a delusional spiral, that distinction is exactly what's collapsed.
These techniques take time. Weeks, sometimes months. There are no shortcuts. But they work because they respect the person's autonomy while gently expanding their perspective.
Iterative Steering: Slow Course Correction Over Time
AI delusions don't appear overnight. They unfold across multiple conversations and months through what researchers call reality testing drift — a gradual loosening of the boundary between AI-generated content and personal conviction.
Because the problem develops slowly, the solution has to match that tempo. A one-time intervention — a single heartfelt conversation, an ultimatum, an app download — won't reverse months of drift. What's needed is iterative steering: slow, consistent course correction that anchors the person back toward shared reality over time.
Motivational interviewing is particularly well-suited for this. The technique involves asking nonjudgmental questions and assessing how invested the person is in their beliefs, exploring any reasons they might have to change them, and examining the perceived costs of not changing.
Here's a detail that gives families hope: past research has shown that family members can be trained to use motivational interviewing techniques for other conditions involving delusions. It hasn't been specifically tested for AI-associated delusions yet, but the mechanism is the same — people who love someone can learn to facilitate insight without triggering resistance.
The key is patience. Iterative steering isn't dramatic. It's the quiet work of staying present, asking good questions, and refusing to be drawn into arguments about whether the delusion is real.
Target the Surrounding Conditions
One of the most practical insights from the research is this: rather than focusing on changing delusions directly, target the factors that exacerbate them.
Disrupted sleep. Active substance use. Social isolation. These aren't the delusion itself, but they're fuel. Improve them, and you may meaningfully improve functioning overall — even if the delusional belief persists.
Consider bipolar disorder as an analogy. Disrupted sleep is a major contributor to episodes. You don't have to resolve the underlying mood disorder to see improvement if you stabilize sleep. The same principle applies here.
In practice, this might look like: encouraging regular sleep schedules, gently addressing substance use without making it the central conflict, re-establishing social connections that exist outside the AI relationship. It's not glamorous work. It doesn't make for a dramatic intervention narrative. But it addresses the conditions that keep the spiral turning.
Improving self-care — real, tangible self-care — can meaningfully improve a person's overall functioning. And sometimes, as functioning improves and the surrounding chaos settles, insight follows naturally.
What Families Can Do Right Now
If you're reading this and recognizing your situation, here's what's actually available to you — not the theoretical ideal, but the real-world options.
Support groups exist for family members and partners, even when the individual themselves declines care. You don't have to navigate this alone. Crisis planning with family or partners is a positive, proactive step — not a sign of defeat, but of preparation. These frameworks rely on the individual's level of insight, which can fluctuate, so having a plan in place matters.
On the design side, researchers are exploring what they call "metacognitive mirrors" — similar in concept to screentime notifications, but designed to help individuals reflect on how they're using AI. These tools could highlight warning signs of maladaptive use, such as AI interactions that disrupt healthy sleep patterns. They might also flag when a user is asking for constant reassurance — a pattern that often accompanies delusional spirals.
Such tools could benefit other conditions worsened by AI use too, including obsessive-compulsive disorder. The idea is to create external prompts for self-reflection at moments when internal insight has temporarily collapsed.
Finally, researchers like Thomas Costello and colleagues are exploring AI-mediated methods to reduce conspiracy beliefs — work that may eventually offer additional avenues for addressing AI-associated delusions from a different angle.
The bottom line: you can't force insight. But you can build the conditions where it might emerge. And in the meantime, you can take care of yourself.