Avoidance is what keeps anxiety going, and the way out is duller than you think

Every anxious person has worked out a way to make the feeling stop. Leave early, sit near the door, check one more time. It works, which is the problem. Here is what the trials say about unwinding it, and why the crutch you never actually reach for is still costing you.

Anxiety is not really maintained by the thing you are afraid of. It is maintained by what you do to make the fear stop.

You leave the party at nine. You sit near the door. You check the lock once more, send the message to a friend first to see if it reads oddly, drive the long way round. Each of these works. The fear drops within minutes, every time, reliably. That reliability is exactly the trouble: you have just taught yourself that the fear was right and that the escape is what saved you.

The loop

The anxious prediction is never tested. You believed something bad would happen, you left, nothing bad happened, and the obvious conclusion is that leaving worked. The belief survives intact and slightly stronger. Next time the pull to leave arrives sooner.

This is the same shape as the depression spiral, where doing less leaves less to feel good about, which makes doing anything harder. We wrote about that one separately. Anxiety runs the same engine with the direction reversed: depression contracts your life through absence of reward, anxiety contracts it through relief.

One honest caveat before going further. Avoidance is the central mechanism in every major model of anxiety, and it is the thing treatments are built to target. But I could not find a meta-analysis that puts a number on how much of anxiety's persistence avoidance accounts for. It is a well-supported framework rather than a measured quantity, and anyone telling you the precise percentage is making it up.

What the trials actually show

The treatment that follows from the model is exposure: going toward the thing, deliberately, repeatedly, without the escape hatch, and staying long enough for the prediction to be tested rather than dodged.

The number to start from comes from a meta-analysis of randomized placebo-controlled trials of cognitive behavioral therapy for anxiety and related disorders: a Hedges g of 0.56 on the symptoms of the target disorder. Placebo-controlled is the part that matters, because waiting-list comparisons flatter whatever is measured against them. Protocols built mainly on exposure did better than ones built mainly on cognitive work, though that difference was not significant, which comes back in the next section.

Most of the newer trial evidence is on exposure delivered through virtual reality, because it is far easier to run. A 2026 meta-analysis of 26 randomized trials in people with diagnosed anxiety disorders found a moderate effect on symptoms, Hedges g of -0.61, confidence interval -0.90 to -0.33, pooled from 13 of those trials. An earlier meta-analysis of 13 trials and 397 people found a Cohen d of 1.11 against control conditions. Effects that large almost always come down as the trials get bigger, and that is what has happened. The moderate figure is the one to trust.

Virtual reality results carry over to the real-world version better than you might assume: a 2019 meta-analysis comparing the two head to head found no significant difference, g of -0.07. The behavioral finding is the one to be careful with. The 2026 analysis found approach behavior improving at g 0.62, meaning people went toward things they had been going around rather than just reporting they felt better, and that figure rests on four trials with a confidence interval of 0.11 to 1.12. It barely clears zero. It is the most interesting number in the paper and the least secure, so take it as a hint rather than a finding.

The part that surprises people

You might assume that the talking half of therapy is where the work happens, and the uncomfortable half is the price you pay for it. A systematic review of 20 trials and 1,308 patients compared pure exposure against pure cognitive therapy and found no significant difference for obsessive compulsive disorder, PTSD, or panic disorder, at the end of treatment or at long-term follow up.

With one exception, and it runs the other way. For social phobia, cognitive therapy was significantly better than exposure at both time points. The review's authors call that strong evidence while also noting it came from only two research groups and needs independent replication. If what you have is social anxiety, that is the result on this page pointed most directly at you, and it is not an argument for going straight at exposure.

The no-difference findings need reading carefully too. The review reports no equivalence test and no power calculation, and its authors flag the small number of trials and their small samples themselves, so this is a difference that was not detected rather than one shown to be absent. What it supports is narrow: there is no good evidence that you have to think your way into feeling ready before approaching anything. Readiness is not the entry requirement. It does not support the larger claim that the two routes are interchangeable.

The crutches

The usual instruction is that safety behaviors must all go. No water bottle, no phone in your hand, no friend on standby, no sitting near the exit. The reasoning is clean: a crutch stops the prediction being properly tested, so the learning never lands.

A meta-analytic review of twenty studies pooled the trials on this and could not produce compelling evidence either for removing safety behaviors during exposure or for allowing them. On its own that reads like the rule is unsupported.

It is not, and this is where a single pooled null is misleading. A placebo-controlled trial of 72 people with marked claustrophobic fear compared exposure alone against exposure with safety aids available and exposure with them used. High end-state functioning afterwards was 94% for exposure alone, against 45% and 44% for the two safety-behavior conditions, with 25% for a credible placebo. The authors' reading is the uncomfortable part: it was the perceived availability of the aid, not whether anyone actually used it, that did the damage.

So the water bottle in your bag is not neutral just because you never reach for it, and that is the best single trial on the question.

What the evidence does support is a distinction rather than a slogan. A systematic review splitting safety behaviors by function found that ones still allowing full confrontation with the thing you fear did not spoil the exposure, while ones that stopped you engaging with it weakened the outcome. The crutch that keeps you out of the situation is the problem. The one that gets you into it may not be.

Which leaves a practical answer rather than a pure one. Plan to drop them, from the start, and treat a crutch as a way in rather than a way to stay. If the real choice is between an exposure with your phone in your pocket and no exposure at all, do the one that will actually happen, then do the next one without it.

What this looks like on a Tuesday

The useful unit is small, repeated, and boring.

Pick one avoidance you can name. Not your largest fear. The specific thing you did last week to make the feeling stop.

Do the smaller version on purpose, and stay past the point where you would normally leave. The staying is the active ingredient. Leaving at peak fear teaches the old lesson again.

Repeat it before moving up. Once is an anecdote. The learning comes from the same prediction failing several times.

Expect it to feel worse first. Anxiety usually rises before it falls, and the rise is not evidence that it is going wrong.

None of this is a substitute for working with a professional, particularly for panic disorder, obsessive compulsive disorder, or trauma, where doing exposure badly can make things worse and the structure of a trained person matters. If what you are carrying is trauma, this is not the page to start from.

The honest summary

Exposure has as much evidence behind it as anything in this area, and its effect is moderate rather than miraculous. It works by letting a prediction fail in front of you, which is why it cannot be done comfortably and cannot be done in your head. The discomfort is not a side effect of the treatment. It is the treatment.

And if you are currently taking medication for anxiety, nothing here is a reason to change it. That is a conversation with the person who prescribed it, and behavioral work sits alongside it rather than in place of it.

This article is for general information and is not a substitute for professional care. If you are in crisis or thinking about harming yourself, please contact your local emergency services or a crisis line. See crisis lines.