Among the techniques used in Cognitive Behavioural Therapy for psychosis, behavioural experiments are arguably the most powerful and the most under-used. The principle is simple: a belief makes a prediction about the world; a small, safe experiment lets the patient see whether the prediction holds. When the prediction does not hold — and the patient was the one who saw it — the belief tends to weaken in a way that no amount of conversation could achieve.
A behavioural experiment is a small, planned, real-world test of a prediction made by a distressing belief — designed collaboratively, performed safely, and reviewed in detail afterwards.
Why experiments work when arguments do not
Talking changes beliefs slowly. Lived experience changes beliefs quickly. This is true for everyone — most of us could be told for years that public speaking is not really dangerous and still feel terrified, until the first time we do it and notice we are still alive. The same dynamic applies in psychosis. A patient who believes that leaving the house will be punished by a voice can sit through twenty sessions of cognitive work and not shift much. Walking around the block once, with the therapist or a trusted person, can shift more.
The cognitive theorist Christine Padesky has described behavioural experiments as the most effective single intervention in the cognitive toolkit. James Bennett-Levy and colleagues' textbook Oxford Guide to Behavioural Experiments in Cognitive Therapy remains the standard reference.
The structure of a behavioural experiment
A well-designed experiment has several parts:
- Identify the belief — what specific belief drives distress? ("If I leave the house the voices will get worse and they will follow through on their threats.")
- Identify the prediction — what specifically does the belief predict? ("If I walk to the corner shop, within an hour something bad will happen — I will hear the voices much louder, or someone will threaten me, or I will collapse.")
- Rate the prediction — how sure are you, 0 to 100? (Often 80 or 90.)
- Design the test — what is the smallest, safest, most informative experiment? (Walk to the corner shop, with a friend, at a quiet time, having timed the trip in advance.)
- Identify safety behaviours — what would you normally do to "make sure" the bad thing does not happen? (Walk with headphones, avoid eye contact, hold a particular object.) Plan whether to drop them.
- Run the experiment — do the planned activity
- Record what actually happened — voice levels before and after, what people did or did not do, what felt different from prediction
- Re-rate the belief — how sure are you now? What does this say about the original prediction?
An example
A composite case. A young man — call him Mehdi — believed that the people in his apartment block had organised against him and would harm him if he was seen alone in the corridor. He had been ordering food in for six months and only leaving with his sister. The cost was substantial: he had stopped working, was gaining weight, and felt his life shrinking.
An experiment was designed: Mehdi would walk from his apartment door to the lift and back, alone, on a weekday morning. He would time it. He would record what people in the corridor did or did not do. His prediction was 80% certain that someone would shout at him, block his way, or threaten him.
He did the walk. He passed two neighbours. One nodded. The other did not look up. Nothing happened. The experiment was repeated several times across the week with different routes. After six experiments he rated his certainty at 35%. The belief did not vanish — but it had been shaken in a way no amount of conversation had managed.
Safety behaviours: the hidden saboteurs
Most patients with persistent paranoid beliefs have a set of safety behaviours — small actions they perform to prevent the feared outcome. Carrying a particular object, walking a particular route, avoiding eye contact, checking specific things. The trouble with safety behaviours is that they prevent disconfirmation: when nothing bad happens, the patient credits the safety behaviour rather than the falseness of the belief.
Skilled experiments often involve dropping safety behaviours one at a time, so the patient sees that the bad outcome does not follow even when the protection is removed. This is done carefully and collaboratively, never imposed.
Special considerations in psychosis
Take it slow
Patients with psychosis often have higher baseline distress. Experiments are smaller, safer, and more incremental than in standard CBT.
Build in support
The therapist or a trusted person often accompanies the first attempts. Phone backup is common.
Plan for any outcome
What will the patient and therapist do if the predicted bad thing does occur? Having an explicit plan reduces fear of the experiment itself.
Respect the limits of evidence
One experiment never proves a belief wrong. The aim is gradual erosion of conviction, not knockout. Many small experiments accumulate.
Beyond paranoia: experiments for other targets
Voices
"If I do not obey the voice for one hour, it will hurt me." Experiment: do not obey for one hour, with a planned safe activity.
Negative symptoms
"If I try to call my friend, it will be awkward and I will feel worse." Experiment: make the call, rate mood before and after.
Negative beliefs about self
"If I show up at the support group, people will stare at me." Experiment: attend, observe what people actually do.
What the evidence shows
Behavioural experiments are core to most CBTp protocols and are embedded in the manualised therapies tested in trials reviewed by NICE (CG178). Specific component-level dismantling studies are limited, but the overall therapy effect — small to moderate on positive symptoms, larger on distress — depends substantially on this technique. Studies of Daniel Freeman's work on persecutory delusions, including the SlowMo and Feeling Safe trials, have shown that targeted behavioural experiments combined with cognitive work produce meaningful reductions in conviction and distress.
For people considering this work
If your therapist is good, they will not push experiments before you are ready. They will ask you to rate predictions, design tests collaboratively, and respect your sense of safety. If you are doing CBTp and have not yet done any behavioural experiments, that is worth a conversation — it may simply be early in the work, or it may indicate that the therapist is not using the full toolkit.
This article is for educational purposes only and is not medical advice, diagnosis, or treatment. Always consult a qualified mental health professional. If you or someone you know is in crisis, call or text 988 in the US, or your local emergency number.