Skip to content

Editorial Standard — experiment session (CBT behavioural experiment)

Purpose

A behavioural experiment session prompts the user to name a tinnitus-related activity they have been avoiding, write their prediction of what will happen if they do it, then go and try it. The session stays open -- the alert field reminds the user to return after the activity. This is a between-session assignment, not a contained in-app experience.

The therapeutic mechanism is prediction testing: cognitive restructuring identifies distorted automatic thoughts, but the experiment tests them against reality. A user who predicts "if I go to a restaurant without earplugs, I will panic" and then does not panic has updated their threat model with direct evidence. PMC12109689 (2025 Tonndorf lecture) rates behavioural experiments at 9/10 for patient-perceived effectiveness -- the highest-rated technique in that survey.

Behavioural experiments appear in Weeks 5-7 (Cognitive Work phase), after psychoeducation and basic relaxation are established and before ACT acceptance work begins.

Voice register

Default register is early habituation, leaning toward the authoritative end. The user is being asked to do something that feels risky. The intro framing must be honest about that ("this is designed to test a prediction, not to prove you are fine") while being directive enough that the user understands the exercise has a specific structure and purpose.

Do not soften the ask. "Pick something you have been avoiding" is direct and correct. "Consider perhaps exploring an activity you might have been a little hesitant about" is patronising and also fails to explain that avoidance is the thing being addressed. The knowledgeable guide names the mechanism (prediction, test, evidence), states the task, and trusts the user to do it.

Do not pre-reassure. The intro must not contain "you might find it's easier than you think" or similar framing -- that pre-empts the experiment's result and undermines the learning. The experiment's value is that the user discovers the result themselves.

Evidence / IP

Grounded in docs/session-content-evidence-base.md. Key sources:

  • Behavioural experiments as a CBT component: identified in the 25-component CBT scoping review (vault note #586, Behavioural category). Rated 9/10 by patients in the Beukes 2025 Tonndorf lecture (PMC12109689) -- joint highest-rated technique in that survey alongside thought diary.
  • iCBT4Tinnitus programme: includes behavioural experiments as a core module (7-module NHS-adapted programme). The Naluma experiment session operationalises the same structure: avoided activity + prediction + real-world test + return to record.
  • Graded exposure integration: the 25-component review identifies graded exposure (Behavioural category, section 2) as distinct from behavioural experiments but closely related. The experiment session can serve as the first step in a graded exposure sequence -- an avoidance identified here feeds into a graded exposure ladder in subsequent sessions.
  • Fear mediation: vault note #542 establishes that fear mediates CBT benefit for tinnitus. The behavioural experiment directly targets fear-avoidance patterns -- the most clinically important pathway.

IP: CBT techniques, including behavioural experiments, are academic and unprotectable. The schema structure (avoided activity, prediction, alert-to-return) is Naluma-authored -- no protected programme text is involved.

Length / reading level

  • Intro heading (intro_h4): 3-8 words. A directive frame, not a question. "Test your assumptions" or "What are you avoiding?" or "An experiment for this week."
  • Intro body (intro_body): 2-4 sentences. Explains the structure of the exercise: you name an avoidance, write a prediction, go try it, return to record. Must include the mechanism (prediction testing). No preamble before the task description.
  • Prompt labels: 5-10 words each. Clear, specific, non-judgmental. "Activity I have been avoiding because of tinnitus." "What I predict will happen." Do not use the word "negative" in prompt labels -- it pre-frames the prediction as wrong.
  • Hint text: 5-12 words. One concrete example. "e.g. going to a restaurant without earplugs." Not abstract: "e.g. something you've avoided" is unhelpful.
  • Alert text: 1-2 sentences. Reminds the user to return after attempting the activity. Should state what to do when they return ("come back and record what actually happened").
  • Reading level: Grade 8 or below throughout.

Editorial-required elements

  1. Mechanism in the intro, always. The user must understand that this is a prediction test, not an exposure exercise. "Write what you predict will happen -- you'll come back to see if your prediction was right" makes the structure explicit and sets up the learning. A session that omits the prediction-testing frame is just an avoidance list, which has no therapeutic mechanism.

  2. No pre-empting the outcome. The intro must not suggest what the result will be. "You might be surprised" and "often people find it isn't as bad as they feared" corrupt the experiment. The result is unknown -- that is what makes it an experiment.

  3. The alert text must be directive. The user needs to return after the activity to complete the experiment. The alert text is the only reminder mechanism in the session. It must clearly state the action ("come back after you've tried it and record what actually happened") not the feeling ("hope it went well!").

  4. Prompt labels must be non-judgmental. The avoided activity is not a "negative behaviour." The prediction is not an "irrational thought." Labels that imply the current state is wrong will cause users to under-disclose or idealise their answers.

  5. Hint text must be tinnitus-specific. Generic hints ("e.g. something challenging") do not help. Tinnitus-specific hints ("e.g. going to a busy restaurant", "e.g. spending an hour without background noise") give the user a concrete model of what belongs in the field.

  6. Obeys ai-patterns-en.md. All copy must obey ai-patterns-en.md. The experiment session is particularly vulnerable to toxic positivity in the alert field ("you've got this!") -- that is banned.

Examples

Good -- intro heading and body:

Test your prediction

Pick one thing you have been avoiding because of tinnitus. Write down what you predict will happen when you do it anyway. Then go and try it this week. Come back here to record what actually happened.

Why this works: states the structure precisely (avoid, predict, test, return), uses the clinical frame (prediction) correctly, makes no assumption about the outcome, and ends with a clear action. No wellness language.

Good alert text: "Come back after you have tried the activity. Record what actually happened -- not what you felt, but what occurred."

Bad -- intro body:

❌ Tinnitus can make us feel like we need to avoid certain situations, but often these fears are bigger than reality. Take this opportunity to step outside your comfort zone and find your peace by gently challenging yourself to try something you've been avoiding. You've got this!

Why it fails: "find your peace" and "you've got this" are banned under ai-patterns-en.md (Naluma-voice additions). "Comfort zone" is a banned construction (generic self-help). More critically, "often these fears are bigger than reality" and "bigger than reality" pre-empt the experiment's result -- that is exactly the opposite of how behavioural experiments work.