You can spell ‘OCD recovery’ without ‘A.I.’

If you’re in a rush:

• AI tools like ChatGPT are now widely used daily - even as emotional outlets - but their responses are just plausible patterns, not true understanding.

• For people with OCD (including Real Event OCD), using AI for reassurance can reinforce the obsessive-compulsive cycle and deepen reliance on external certainty.

• While AI can feel supportive in moments of stress, repeatedly turning to it for answers about fears or past events increases distress rather than helping long-term recovery.

ChatGPT is now used by close to 1 in 8 people around the globe each week. Collectively, we generate somewhere around 2.5 billion prompts every day.  For many of us, ChatGPT (amongst other AI services) has become a daily tool in our lives, whether it’s to generate a recipe with the ingredients in our fridge, research something we’ve heard in the news, or help with our studies.

Many people are also using AI as their confidante, as it provides an infinite space for emotional venting, advice-seeking, and supportive-seeming conversations.  To an extent, I can see how this might be useful – for example, for someone who feels they have no trustworthy relationships on which to rely, or who just wants a space to articulate their feelings and reach some greater sense of clarity.

However, Large Language Models are also a massive potential risk for obsessive-compulsive people who often find themselves on an endless search for certainty – including those of us who experience Real Event OCD.

I recall when I first experienced acute OCD-related anxiety, back in the early 2010s, I found myself regularly googling things like “why I am anxious?”.  I was desperate for answers.  Where did this panicky feeling come from?  Was it part of a mental health condition?  And most critically, how could I make it go away?  At that point, Google could ‘only’ provide a long list of articles, enabling me to disappear down rabbit-holes and obsess over whether I had depression, generalised anxiety, OCD, narcissism, nutritional issues, and all kinds of other conditions.

Fast forward 15 years and we now have, at our fingertips, the ability to ask any question we can think of and get a thoughtful-sounding response from an LLM.  And why is that problematic for those of us with OCD?

Well, there are many, objective reasons when it comes to accepting an LLM’s ‘opinion’.  Here are some of the key ones:

·         AI can’t have its own opinion – it merely pulls together a pattern of existing information and gives you a plausible synthesis of what people might tend to say.

 ·        LLMs are trained to produce answers that sound right – so it can state things too confidently and smooth over ambiguity.

 ·        It doesn’t have the full context of anything – only the prompt you’ve written in.  It doesn’t know you, and it can’t weigh up your whole experience or self.

 ·         Stepping back for a moment – why would I think that an information-aggregating piece of software could hold any kind of moral judgment over me, whether that judgment is ‘good’ or ‘bad’?  With OCD, it’s tricky enough that we can find ourselves so beholden to other people’s views of us, let alone a piece of software.

Things become even riskier when we compulsively use an LLM to seek reassurance about our (RE)OCD worries and doubts:

·         Reinforcing the vicious OCD cycle.  Intrusive worry —> feel anxious —> compulsion —> maybe get temporary relief —> doubt comes back even stronger later.  By compulsively seeking reassurance, we condition our minds that more compulsions are the only way to feel better again.

 ·         Relying on external certainty.  Just as compulsively seeking reassurance from other people is unhelpful, excessive use of LLMs shields us from the real, long-term antidote to OCD recovery – which is learning to tolerate uncertainty and to know that we’d survive even if our fearful worries came true.

 ·         ChatGPT tends to finish its responses with an invitation to continue examining issues from further angles.  In other words, it may take an obsessive fear, unpack it, and encourage you to analyse and ruminate on it even further.

 ·         One of the worst outcomes is that the LLM actually presents you with the possibility that you did do something unforgivably wrong or immoral, which can trigger a huge rush of panic and leave you feeling even worse.

As ever, I appreciate that the rational part of your mind will likely know and understand all the points I’ve highlighted in this piece.  And that’s only half the battle, because when we’re in an REOCD spike, that rationality can be clouded by the desperate wish for emotional safety.

Even so, there is benefit in holding an awareness of the pitfalls.  Next time you find yourself troubled by an anxious doubt, and feel the temptation to ask an LLM to unpack an obsession with you or give an opinion, consider how this might be a self-harming act which will only bring more obsessions and compulsions in future. 

Is there something more recovery-centred you could direct your energy to instead? In my book, I explore a range of healthier ways to tend to your mental, physical and emotional self.