How anxiety patterns actually change: the neuroscience of updating a threat prediction

There is a particular kind of frustration that comes with catching yourself doing it again. The same anxious loop. The same retreat from the same kind of conversation. The same overthinking at three in the morning. Whatever the pattern is, you have understood it for some time now. You can name it. You know where it came from. And it has barely shifted.

If that has been your experience, the question worth asking is not whether you have understood the pattern well enough. You probably have. The question is whether the kind of attention you have been giving it is the kind that anxiety patterns actually respond to. The honest answer, supported by the science of how fear-related learning is stored, is that it usually is not.

Knowing has not changed the pattern because the pattern is not stored where knowing lives. Anxiety patterns are not just habits. They are predictions held in long-term threat memory. They do not weaken from insight alone, and they do not weaken from forced repetition of the right thought. They update when something specific happens at a specific time, in a specific kind of state. What follows is a brief tour of why, and a way to work with the pattern that takes the biology of it seriously.

A habit is a sequence the brain has automated because it pays off. Brushing your teeth in the same order each morning, taking the same route home, reaching for your phone when you sit down. Habits change by being interrupted and replaced. The system does not need much convincing because it does not, fundamentally, believe anything is at stake.

Anxiety patterns are different. They sit on top of a much older system, one whose job is to keep you alive. When the brain learns that a particular cue, situation, or internal sensation goes with danger, it stores that as a prediction. The next time the cue appears, the system runs the prediction automatically and without consulting you. Your heart rate climbs before you have decided to be worried. Your shoulders rise before you have noticed the room. The pattern is not waiting for permission. It is doing the job it was built for.

Bouton, Maren and McNally set this out in a 2021 review in Physiological Reviews. One of the central findings is that the original learning is not erased. Even after a pattern has stopped showing up, the research suggests the original prediction is still on file. Something new is laid down beside it. That new learning can become the dominant response, but the old one remains, which is why the pattern can come back under stress.

Most of us, when faced with a pattern we want to change, default to reasoning with it. We tell ourselves it is irrational. We list the reasons it does not make sense. We rehearse what we should think instead. And, sometimes, we feel a bit calmer for ten minutes. Then the cue shows up and the prediction fires anyway.

There is a reason for this, and it is not a personal failure. Kredlow and colleagues (2022) reviewed this in Neuropsychopharmacology. The parts of the brain that hold the threat prediction and the parts that hold considered, language-based knowledge are doing different jobs. The amygdala and its circuits build and run the prediction. The prefrontal cortex, slower and more deliberate, can sometimes regulate it. The two are connected, but they are not the same system. One cannot rewrite the other simply by talking.

Put plainly, the part of you that knows is not the part of you that predicts. The research suggests the pattern is not waiting to be argued with. It is waiting for evidence. Specifically, it is waiting for an experience that is different enough from what it predicted to count as new information.

If the pattern updates by mismatch rather than by argument, the next question is what kind of mismatch counts. The clinical translation is set out by Craske, Treanor, Zbozinek and Vervliet (2022) in Behaviour Research and Therapy. Their model centres on a single idea: expectancy violation. The feared thing is contacted. The feared outcome does not arrive in the way the system predicted. The gap between expected and actual is what gives the brain something new to encode.

Three conditions need to be present for the gap to do its work. First, the pattern has to be active. A prediction can only be updated if it has been retrieved. Thinking about it from a distance is not the same as having it online. Second, the present-moment experience has to be different enough from the prediction to count as a mismatch. The bigger the gap between what was expected and what actually happens, the stronger the new learning. Third, the system has to be regulated enough to register the mismatch as information rather than as another threat.

That third condition is where many honest attempts at change come unstuck. If the system is overwhelmed, the new experience does not land. It gets filed alongside the threat rather than as evidence against it. This is not a failure of effort. It is a feature of how the encoding works. The window in which an old pattern can update is real, and it is reachable, but it is narrower than people usually expect.

What follows is one structure for translating those three conditions into something you can actually do. It is not a script. The exact shape will depend on the pattern. The aim is to give you four moves to make, in roughly this order, when you next find yourself facing the cue that fires the old prediction.

Take a worked example. The pattern is a familiar spike of dread before sending an email that asks for something. The old prediction, learned somewhere along the way, is that the request will land badly and the relationship will tilt. The pattern fires before you have read the email back. The new pattern looks like this.

Two practical notes. First, the size of the mismatch matters more than the size of the cue. Doing something small that genuinely lands as different is more useful than doing something large while the system is too flooded to encode it. Second, the fourth step is not optional. The new memory is laid down beside the old, not on top of it. The new pattern needs to be rehearsed across enough situations to become the response the system reaches for first.

There is a question that comes up at this point, and it deserves a direct answer. If patterns can update, why does it so often feel slow? Cuijpers and colleagues (2025) looked at this on a large scale in JAMA Psychiatry. They pooled three hundred and seventy-five randomised trials of cognitive behavioural therapy across anxiety, depression and related disorders, with almost thirty-three thousand patients. The average effect for anxiety disorders was moderate to large. The takeaway is not that change does not happen. It reliably does. The takeaway is that even the most effective psychological treatments work in steps, not switches.

The other thing slowing the work is often the response to the work itself. A pattern fires, the person notices it, and the response is self-criticism. Anthes and Dreisoerner (2026), in Mindfulness, noted that harsh self-criticism switches on the body's threat physiology. It puts the system in exactly the state in which a new mismatch experience cannot be encoded. Kindness toward yourself is not, in this picture, a niceness. It is the regulation condition that lets the third step of the work happen at all.

It helps to remember what the pattern is doing. It is running an old prediction about danger. It is not stupid, and it is not your enemy. It is doing exactly what it learned. The work is not to get rid of it. The work is to give the system, on enough occasions and under regulated enough conditions, an experience that is different enough from the prediction to count as new information. The pattern does not need to vanish. The new one needs to become more accessible.

Getting rid of the pattern was never the assignment. Building a second response, layered alongside it, more available with each repetition, is.