Scrolling for reassurance, finding more dread: Why turning to AI when you’re anxious might be making things worse

Jess Bigogno is a psychologist and psychotherapist based in Barcelona, where she has spent the past eighteen years working with clients on anxiety, low mood, relationship difficulties, stress and the disorientation that can come with a major life transition.

In this article for MumAbroad, she looks at a shift she has noticed in her own practice: clients who once Googled their worries at 2am now turning to AI chatbots instead, and why a more accurate, more articulate answer can end up feeding anxiety rather than easing it.

Drawing on real examples from her work, Jess explores the difference between using AI as a genuinely helpful tool and using it as a compulsion, and what actually helps when the question underneath the question is one no answer can ever fully settle.

A few years ago, the anxious clients in my practice were Googling looking for answers to their worries at 2 a.m. Now they’re doing something that looks similar but isn’t: they’re opening ChatGPT.

On the surface, it seems like an upgrade. Google gives you a chaotic list of forum posts and worst-case-scenario blogs. AI gives you something that sounds like a calm, informed, personalised answer. But as a therapist who works a lot with anxiety, I’ve started to see this shift as a problem, not despite AI being more accurate than a Google search, but because of it.

The trouble with almost-certainty

Anxiety doesn’t run on facts. It runs on the search for certainty. Am I healthy? Am I in the right relationship? Should I change my job? Does my child have a problem I should worry about? The anxious brain isn’t asking, “What’s probably going on here?” It’s asking, “Can you guarantee this isn’t dangerous?” And no honest source, human or AI, can ever answer that question with the airtight certainty the anxious mind is looking for.

This is where AI becomes uniquely dangerous for this particular struggle. A chaotic Google search is easy to dismiss; anyone can tell a random forum post isn’t a diagnosis. But a large language model gives thorough, coherent, individualised-sounding answers. It considers your specific combination of symptoms. It lists differentials. It hedges appropriately, the way a careful doctor would. Which means it’s genuinely more accurate and also, paradoxically, makes it more compelling to check “just one more time.”

I worked with a client who came to me for help with health anxiety that had become so severe she was missing work. She had a history of googling symptoms, but by the time we met, she’d mostly moved to ChatGPT. She’d describe a headache or a strange muscle twitch in exhaustive detail, then follow up with clarifying questions for an hour, sometimes two. The responses were thoughtful. They usually included some version of “this is most likely benign, but if X or Y occurs, seek medical attention.” For a moment, my client would feel calm. Then the “most likely” would snag on her, and she’d be back, rephrasing the same question to see if a different answer emerged.

This is the core mechanism of health anxiety: reassurance-seeking that works for minutes and fails structurally over time. Each search, each chat, is a compulsion in the same family as checking a lock repeatedly. It doesn’t resolve uncertainty; it reinforces the belief that uncertainty is intolerable and must be resolved before you can feel okay. AI doesn’t cause this pattern, but its fluency and apparent thoroughness make it an extraordinarily effective tool for feeding it. Worse, because the answers are better than a random web search, they feel more authoritative, which means the small, inevitable hedge at the end (“but please see a doctor if…”) lands with more, not less, weight.

The actual work my client and I did had almost nothing to do with her symptoms. It was about building her tolerance for not knowing, sitting with a sensation in her body without needing to resolve it immediately, and learning to notice the urge to check and let it pass. Every hour spent refining a prompt to extract more certainty from a chatbot was an hour not spent building that muscle. Treating the anxiety meant treating her relationship to uncertainty itself, not sourcing better information.

When AI is used well

I don’t think the answer is to swear off these tools, and one of my clients showed me what a healthier version of this can look like.

This client has generalised anxiety and has been in therapy on and off for years. She’d read several books on cognitive behavioural therapy and anxiety management that genuinely helped her: books we’d discussed together in session. At some point she uploaded those specific books into an AI tool and set a rule for herself: she’d only ask it to reference material from those texts, nothing else. When she felt stuck between sessions, she’d ask it to explain a concept again, or help her apply a technique from the book to a situation she was facing that week.

The difference is structural. This client wasn’t using AI to generate new information or chase certainty; she was using it to reinforce a framework she and I had already built together, one grounded in accepting discomfort rather than eliminating it. She’d created a boundary for the tool. It couldn’t offer her a diagnosis or a guarantee, because she’d never asked it to.

The real question isn’t “AI or not”

The tool itself isn’t the issue. The issue is what need you’re bringing to it. If you’re asking a question that has an actual, bounded answer, like “How does exposure therapy work?” or “What’s a grounding technique for panic?” AI, used with limits, can be a genuinely useful supplement to therapy. But if you’re asking a question that no answer can ever fully satisfy, am I safe? Is this really nothing? Can you promise…? No source will ever be enough, and a more articulate one will only prolong the search.

If you notice you’re returning to a chat window the way you might return to a compulsion, again and again, needing one more confirmation before you can relax, that’s not a sign you need a better answer. It’s a sign that the question itself needs a different kind of help.

If you would like help with anxiety, I’m here to help you. Book an initial session with me, and we can begin to help you to become free from the grip of worry and begin to live your life with more ease.

Jess offers sessions in English, Spanish or Italian, in person in Barcelona or online. The first step is just a conversation, get in touch to book your session by email jessica@internationaltherapybcn.com or via her website.

About Jess Bigogno

Jess Bigogno is a psychologist and psychotherapist from London who has spent the past eighteen years working in the Barcelona area, helping hundreds of people work through anxiety, low mood and depression, relationship difficulties, stress and burnout, ADHD, and the disorientation that can come with a major life transition, including building a life somewhere new. Her approach is warm and practical, combining real conversation with tools that help day to day, drawing on her training in CBT, Acceptance and Commitment Therapy and mindfulness-based therapy, shaped around what each client actually needs rather than a fixed programme.

FAQ:

Is it bad to use ChatGPT or AI when I’m feeling anxious?

Not necessarily. AI becomes a problem when it’s used to chase certainty on a question that has no certain answer, such as “am I safe” or “is this really nothing.” Used with clear limits, for a bounded question like how a particular technique works, it can be a genuinely useful supplement alongside therapy.

Why does asking AI about my symptoms feel worse than a normal Google search?

A chaotic search is easy to dismiss, since anyone can tell a random forum post isn’t a diagnosis. AI gives a calm, thorough, individualised sounding answer, which makes it feel more authoritative and more worth returning to for one more check, even though that repeated checking is what keeps the anxiety going.

How can I tell if my use of AI has become a compulsion?

A useful sign is whether you’re rephrasing the same worry to see if a different answer appears, or returning again and again for one more confirmation before you can relax. That pattern, rather than the accuracy of the answer itself, is usually what needs attention.

Is there a healthier way to use AI alongside therapy?

Yes. Some clients set clear boundaries, such as only asking AI to draw on material from books or frameworks they’ve already worked through in session, rather than using it to generate new information or new reassurance.

What actually helps with health anxiety, if not more information?

Often the work has little to do with the symptoms themselves. It’s about building tolerance for not knowing, learning to sit with a sensation without needing to resolve it immediately, and noticing the urge to check and letting it pass.

How do I know if it’s time to see a therapist about this?

If you notice yourself returning to searches, whether on Google or AI, the way you might return to a compulsion, that’s usually a sign the question itself needs a different kind of support, rather than a better answer.

Read more articles about health and wellbeing on the Mumabroad blog