Inhibitory-learning exposure

1. Introduction

Version followed: Craske, Treanor, Conway, Zbozinek & Vervliet (2014) inhibitory learning approach (eight optimisation strategies plus affect labelling), with the strategy tiers of the 2022 "inhibitory retrieval" update (Craske, Treanor, Zbozinek & Vervliet 2022; abstract and trial protocol only [uncertain]); OCD-specific translation from Jacoby & Abramowitz (2016); per-exercise record structure from the PROTECT-AD protocol (Pittig et al. 2023).

Inhibitory-learning exposure is not a stand-alone therapy or manual. It is a set of principles and strategies, drawn from extinction research (Bouton's model, among others), for running the exposure component of any CBT for anxiety, OCD or trauma-related problems. Its core idea of change: exposure does not erase the original threat association (feared cue predicts harm). It builds a second, competing inhibitory association (feared cue does not predict harm) [source: Craske et al. 2014; Craske et al. 2008]. Treatment works when that new learning is strong and when it can be retrieved later, in other places and at other times. Fear may come back through spontaneous recovery, a change of context or a stressful event; the approach plans for that from the start. The learning is driven chiefly by expectancy violation: the bigger the mismatch between what the client predicted and what happened, the more there is to learn. Fear reduction within a session is not the index of success. Fear tolerance and updated expectancies are [source: Craske et al. 2014; Craske et al. 2008].

The 2022 update renames it the "inhibitory retrieval approach". It groups strategies as fundamental (expectancy violation, attention to the feared stimulus, removal of safety signals, mental rehearsal after exposure), advanced (deepened extinction, occasional reinforced extinction) and generalisation-promoting (retrieval cues, multiple contexts, stimulus variability, positive affect). It also introduces a clinician mapping tool, the OptEx Nexus [source: Craske et al. 2022, Caltech record abstract].

2. Framework

No phase model, session count or stopping rule is published as a manual. The OptEx trial manual is "available to clinical researchers upon request" [source: NCT04048824 protocol]. The phases below are therefore my synthesis [synthesis]. Each is anchored to the published trial structures named.

Reference formats from published protocols

• OptEx trial (UCLA; social anxiety or panic disorder): 9 weekly individual sessions, session 1 of 90 minutes, then 60 minutes. Treatment starts with an OptEx Nexus. Exposures "test out" whether the feared outcome occurs, and the trial adds deepened extinction, increased attention to the feared stimulus, context and stimulus variability, occasional reinforcement and mental reinstatement. Therapists had an 8-hour workshop and weekly supervision [source: NCT04048824 protocol].
• PROTECT-AD (Germany; panic/agoraphobia, social anxiety, specific phobias): 14 sessions of 100 minutes in four phases. Sessions 1–4 are cognitive preparation; sessions 5–10 are exposure (5 therapist-guided exposures plus self-guided ones), delivered weekly or three times per week with equal results; sessions 11–12 are self-management and relapse prevention; sessions 13–14 are boosters at 2 and 4 months. Exercises were usually not repeated but varied in stimulus and context [source: Pittig et al. 2023].
• PTSD application (Netherlands): 12 sessions of 90 minutes over 4 weeks of imaginal and in vivo exposure, with variability across memories, therapists, rooms and online delivery [source: Kooistra et al. 2025, via Leiden thesis].

Phases [synthesis, built on the above]

Phases C1 to D overlap in practice. The 2014 paper uses variability and multiple contexts from early on, and treats retrieval cues as mainly a late, relapse-prevention tool [source: Craske et al. 2014]. Expanding spacing (sessions progressively further apart) was associated with less return of fear in the authors' own studies, but direct comparisons are scarce [source: Craske et al. 2008]. Jacoby & Abramowitz suggest twice weekly, then weekly, then every other week; they favour massed, supervised sessions when insight, response prevention or family accommodation are problems [source: Jacoby & Abramowitz 2016].

3. Guiding principles

[source: Craske et al. 2014 and 2008 unless tagged otherwise]

• Learning, not fear reduction, is the target. Do not judge a session by whether SUDS fell; judge it by what the client learned and whether the expectancy changed.
• Fear toleration is a primary goal. Present fear during exposure as expected and survivable, not as a sign that something is going wrong [source: Craske et al. 2008].
• The old fear is not erased. Tell clients that fear can return and plan generalisation and relapse prevention from the start.
• Expectancy violation drives change. Design every exposure around a specific prediction that the exposure can contradict.
• The test sets the length. End an exposure when the prediction has been tested or the agreed behavioural goal is met, not when fear drops.
• Performance in session is not learning. "Desirable difficulties" (variability, spacing) can make sessions feel harder while improving long-term retention [source: Craske et al. 2008; Jacoby & Abramowitz 2016].
• The hierarchy is a menu, not a ladder. Pick items for the learning they offer and their life relevance, not by fear rank; often start with a moderately hard item.
• No expectancy-lowering before exposure. Keep cognitive restructuring out of the pre-exposure period; use cognitive discussion afterwards to consolidate.
• Attention stays on the feared cue and on what does not happen. Distraction counts as a safety behaviour.
• Safety signals and safety behaviours block learning. Identify and remove them, abruptly if the client will, gradually only to prevent dropout.
• Vary more than you repeat. Vary stimuli, intensity, duration, order, context and spacing.
• The client is a personal scientist. Every exercise is an experiment to "test it out".
• Consolidate every time. Each exposure ends with a structured debrief of what happened and what was learned.
• Target what can be tested. For distal or unverifiable fears, test the near-term expectancies about tolerating doubt, anxiety and intrusive thoughts [source: Jacoby & Abramowitz 2016].
• Track expectancies, not only SUDS. Rate the expectancy before, the outcome after, and the revised expectancy [source: Pittig et al. 2023].
• Collaborate and agree in advance. Unpredictable lengths and random item order are used only after the client agrees to the general principle [source: Craske et al. 2014; Jacoby & Abramowitz 2016].
• Do no real harm. Never expose to genuine danger. Use reinforced extinction only where the aversive outcome is safe and agreed. Use imaginal exposure for outcomes that cannot be tested ethically [source: Craske et al. 2014; Jacoby & Abramowitz 2016].

4. Intervention inventory

All example therapist language below is my own illustrative wording unless it is a short, cited quote. The 2014 paper's worksheet questions are quoted only in fragments. The 2014 paper is a translational review with case illustrations, not a session-by-session manual [source: Craske et al. 2014].

4.0 Safety, contraindications and consent (applies to every exposure)

• Purpose: to make sure exposure is safe and voluntary before any of the strategies below are used.
• Steps:
  1. Screen before exposure. Check for current suicidal intent or recent self-harm, active psychosis, mania, substance intoxication or withdrawal, and ongoing real danger (for example a current abusive relationship in PTSD). Address these first or treat around them [synthesis, consistent with standard CBT exposure practice].
  2. For interoceptive exposure (spinning, hyperventilation, running, caffeine), get medical clearance where there is cardiac, respiratory, seizure, vestibular or pregnancy-related risk. Do not use caffeine or pharmacological panic induction without medical input. The 2014 paper mentions yohimbine-induced panic only as a research example, not a routine clinic procedure [source: Craske et al. 2014; synthesis].
  3. Never arrange genuine harm. The feared outcome may be tested in imagination when real testing would be unsafe or unethical [source: Jacoby & Abramowitz 2016]. Reinforced extinction is "ethically prohibitive" where it could retraumatise or the setting is dangerous [source: Craske et al. 2014].
  4. Obtain informed consent to the general approach: that fear may stay high, that exposures will not always go in order, and that lengths may be unpredictable. Consent again to each specific exposure. The client can always stop; stopping is discussed, not punished.
• Watch for: dissociation, medical symptoms that differ from the client's usual anxiety, and consent given only to please the therapist.

4.1 Case mapping (functional analysis and OptEx Nexus)

• Purpose: identify what to expose to and what to test before any exposure is planned. The 2022 update provides the OptEx Nexus as a clinician tool for this. In the Dutch PTSD study, therapists used it to find the principal feared cue, and drafted a new Nexus when new feared outcomes emerged. The OptEx trial lists its elements as conditional stimuli, feared outcomes, safety signals and occasion setters [source: NCT04048824 protocol; Kooistra et al. 2025]. The full Nexus template is not drawn on here [uncertain].
• Steps:
  1. Identify the principal feared cue, meaning the cue most central to the fear (for example, bodily sensations of a racing heart in panic), and the secondary cues (places, people, objects, images, thoughts, sensations).
  2. For each cue, write the specific feared outcome(s). Separate near-term, testable outcomes ("I will faint") from distal ones ("I will get cancer in 20 years").
  3. List safety signals (people, phone, medication, water, exits) and safety behaviours, including subtle ones (holding a rail, scripting speech, distraction, reassurance seeking, mental checking).
  4. List avoidance, both outright and partial.
  5. List contexts and occasion setters: when and where the fear is worse or better, such as alone or accompanied, night or day, tired or rested [the role of occasion setters is inferred from extinction theory; uncertain].
  6. Build an exposure menu. For each item, record the target feared outcome, an expectancy rating and an estimated distress rating. Do not commit to a fixed order [source: Craske et al. 2014].
• Example language (illustrative): "Let's map what you're afraid of in detail: the thing that sets the fear off, what you think will happen, and everything you do to stay safe. That map tells us what to test."
• Watch for: feared outcomes stated as feelings ("I'll get anxious"), which cannot be tested; missed subtle safety behaviours; a single feared outcome hiding several.

4.2 Psychoeducation and rationale

• Purpose: secure buy-in for a method that deliberately does not aim to reduce fear in session.
• Steps:
  1. Explain that fear is learned and that avoidance and safety behaviours stop new learning.
  2. Explain that exposure builds a new memory that competes with the old one. The old one is not deleted, so fear can come back, especially in new places or under stress. That is why practice is varied.
  3. Explain fear tolerance. Anxiety may rise and stay up; the aim is to learn what happens, not to calm down. Relief usually comes later, across weeks [source: Craske et al. 2014; Craske et al. 2008].
  4. Introduce the personal-scientist frame: each exposure is an experiment to "test it out" [source: Craske et al. 2014].
  5. Explain desirable difficulties: practice will be mixed and sometimes unpredictable because real life is [source: Jacoby & Abramowitz 2016].
  6. Agree in advance to random order and unpredictable lengths, if they will be used.
  7. Ask the client to explain the rationale back.
• Example language (illustrative): "We're not trying to make the anxiety go away in the room today. We're running experiments to find out whether what you fear actually happens, and to show you that you can handle the anxiety while we find out."
• Watch for: the client equating success with feeling calm; agreement without understanding. The client's paraphrase is the check.

4.3 Specifying a testable feared outcome

• Purpose: an exposure can only violate an expectancy that is concrete and observable.
• Steps:
  1. Ask what the client is most afraid will happen. The 2014 worksheet wording is "What are you most worried will happen?" [source: Craske et al. 2014].
  2. Turn it into something observable, with a time frame: who would do what, and how we would know.
  3. Reject predictions that are only about anxiety ("I'll panic") unless the prediction is about tolerability ("I'll lose control and have to run out", "the anxiety will never stop"). PROTECT-AD therapists were trained to turn vague beliefs into testable, observable outcomes, not fear responses [source: Pittig et al. 2023].
  4. If the true fear is distal or cannot be checked, use 4.18.
• Example language (illustrative): "If your fear came true in the next ten minutes, what would we see or hear? What would someone filming it capture?"
• Watch for: hedged predictions ("something bad"); the prediction shifting after the event ("well, they were just being polite"). Record the exact wording before the exposure.

4.4 Pre-exposure expectancy rating

• Purpose: to make the prediction explicit and measurable so its violation is salient.
• Steps:
  1. Write down the behavioural goal, including what, where, how long and how often (for example, a set number of minutes in a specified situation without the safety signal) [source: Craske et al. 2014].
  2. Write down the feared outcome in the client's words.
  3. Rate how likely it is, 0–100 [source: Craske et al. 2014]. PROTECT-AD also rated anticipatory anxiety 0–100 [source: Pittig et al. 2023].
  4. If the expectancy is low, redesign the exposure to raise it: lengthen it, add a cue or remove a safety signal (see 4.5).
  5. Do not challenge the prediction now.
• Example language (illustrative): "Before we start: right now, how likely, from 0 to 100, does it feel that this will happen?"
• Watch for: therapist reassurance creeping in ("I'm sure it'll be fine"), which lowers the expectancy and weakens the test.

4.5 Designing exposures to maximise expectancy violation

• Purpose: the larger the mismatch between expected and actual outcome, the more is learned.
• Steps:
  1. Choose the item whose feared outcome is most strongly held and most central (often the principal cue), not just the next rung up.
  2. Set the exposure length by what would convincingly test the prediction. If the client fears fainting after 5 minutes of a racing heart, run it well past that point.
  3. If the expectancy rating is low, increase the test: longer, more intense, combined cues, or fewer safety signals. In the 2014 panic case, combining spinning with jogging raised the expectancy to 99% [source: Craske et al. 2014].
  4. Continue until the expectancy is violated or the goal is met, regardless of SUDS [source: Craske et al. 2014; Jacoby & Abramowitz 2016]. In one interoceptive study cited, exposure continued until expectancy fell below 5% [source: Craske et al. 2014].
  5. Use the 2014 case tables as models. The OCD case placed a hand on a sleeping infant's neck for 10 minutes (expectancy 80%). The social anxiety case gave a professional opinion to a coworker (95%). The dog-phobia case stood near a dog for 15 minutes (99%) [source: Craske et al. 2014].
• Example language (illustrative): "Your fear is that if your heart races for more than a few minutes, you'll pass out. So the experiment needs to go longer than that. Shall we do 10 minutes?"
• Watch for: exposures chosen because they are comfortable; exposures too short to test the prediction; clients who "survive" by covert safety behaviours.

4.6 Attention to the feared stimulus

• Purpose: learning requires noticing the cue and noticing that the feared outcome did not happen. The 2022 update lists attention as a fundamental strategy [source: Craske et al. 2022 abstract]. Distraction is described in 2014 as a "pernicious" safety behaviour [source: Craske et al. 2014].
• Steps:
  1. Before the exposure, ask the client to keep their attention on the feared cue: its features, sensations and details.
  2. During it, prompt brief, non-reassuring descriptions ("What do you notice about it now?").
  3. Ask the client to watch specifically for whether the feared outcome occurs.
  4. Block distraction: phone, chatting about other topics, counting, mental escape.
• Example language (illustrative): "Keep looking at the dog. Tell me what it's doing. And keep checking: is what you predicted happening?"
• Watch for: glazed or averted gaze, a sudden change of topic, "zoning out" (in PTSD, check for dissociation).

4.7 Removing safety signals and safety behaviours (includes response prevention)

• Purpose: safety signals lower the expectancy of harm and let the client attribute non-occurrence to the safety aid, which blocks learning [source: Craske et al. 2014; Craske et al. 2008].
• Steps:
  1. Using the map (4.1), list every safety signal and behaviour relevant to this exposure, including subtle ones (fidgeting, earphones, a companion waiting outside, pepper spray, medication carried, rehearsing) [source: Craske et al. 2014].
  2. Remove them all at once if the client is willing. Phase them out only if needed to prevent dropout [source: Craske et al. 2014].
  3. In OCD, apply full response prevention: no rituals, reassurance seeking or escape during or after exposure, including ordering and arranging for "not just right" experiences [source: Jacoby & Abramowitz 2016].
  4. Have the client do exposures alone where a companion functions as a safety signal. In the 2014 OCD case, the wife's presence lowered the expectancy of harm [source: Craske et al. 2014].
  5. Check for safety behaviours after every exposure ("Did you do anything to keep yourself safe?").
• Example language (illustrative): "If you hold the rail, part of you will think: 'I didn't faint because I held on.' Let's find out what happens without it."
• Watch for: substituted safety behaviours, such as swapping a phone for counting. The evidence on safety behaviours is mixed: one study found hand wipes did not increase relapse, and one claustrophobia replication failed [source: Craske et al. 2014]. Still treat removal as the default.

4.8 Post-exposure consolidation and adjusted expectancy

• Purpose: to make the violation explicit and carry it into the updated expectancy. In PROTECT-AD, expectancy change and the "learning rate" (how much of the violation turned into change) predicted outcome, while violation alone did not [source: Pittig et al. 2023].
• Steps:
  1. Straight after the exposure, ask whether the feared outcome occurred (yes/no), "How do you know?" and what the client learned [source: Craske et al. 2014 worksheet].
  2. Ask Socratic follow-ups: "What happened?", "How was that surprising?" [source: Craske et al. 2014].
  3. Rate how far the feared outcome actually occurred (0–100%). Then re-rate the expectancy for repeating the same exercise (0–100%) [source: Pittig et al. 2023].
  4. Now, and only now, use cognitive discussion to consolidate: compare prediction and outcome, and spell out what this means beyond today [source: Craske et al. 2014; Jacoby & Abramowitz 2016].
  5. Ask the client to summarise and synthesise the learning, not just recall it [source: Jacoby & Abramowitz 2016].
  6. Refine the next hypothesis to be more specific or more demanding [source: Craske et al. 2014].
  7. Over sessions, hand the debrief to the client so they ask themselves the questions.
• Example language (illustrative): "You gave it 80% that you'd faint. You didn't. How likely does it feel if we did it again tomorrow?"
• Watch for: discounting ("I was lucky", "it was only because you were here"). Note what the discount is, then design the next exposure to remove it, for example by doing it alone. If the revised expectancy barely moves despite clear violation, see section 5.

4.9 Mental rehearsal after exposure

• Purpose: to strengthen the memory of what was learned. The 2022 update lists post-exposure mental rehearsal as a fundamental strategy [source: Craske et al. 2022 abstract]. The OptEx trial uses "mental reinstatement of extinction learning" [source: NCT04048824 protocol]. A UCLA experimental protocol has participants recount exposures out loud and then in writing after every two exposures [source: NCT06132425 protocol]. The exact clinical procedure in the 2022 paper is [uncertain].
• Steps [partly synthesis]:
  1. At the end of the debrief, or at home that evening, the client mentally replays the exposure, focusing on what they predicted, what happened and what they learned.
  2. Have them tell it aloud or write a short paragraph.
  3. Keep the focus on the non-occurrence of the feared outcome and on coping. Do not let the replay become post-event rumination about what went badly.
• Example language (illustrative): "Tonight, take two minutes to replay it in your mind, from walking in to walking out, and write down the one thing you learned."
• Watch for: the replay drifting into self-criticism. If it does, redirect to the factual evidence about the feared outcome.

4.10 Affect labelling

• Purpose: naming emotions in words during exposure may dampen the threat response without trying to suppress it. The 2014 paper cites spider-phobia work showing reduced skin conductance and more approach at one-week follow-up [source: Craske et al. 2014].
• Steps:
  1. During exposure, ask the client to describe their emotional state out loud in simple words.
  2. Do not ask them to change, reappraise or reduce the feeling.
  3. Keep it short and repeat it periodically. Combine it with attention to the cue. The 2014 PTSD case used affect labelling during imaginal exposure [source: Craske et al. 2014].
• Example language (illustrative): "Just name what you're feeling right now, as it is. No need to change it."
• Watch for: labelling turning into reassurance or analysis ("I feel scared but I know it's irrational"); in that case, go back to plain naming.

4.11 Deepened extinction (compound exposure)

• Purpose: combining two separately extinguished cues that predict the same feared outcome increases expectancy violation and reduced relapse in animal and some human work [source: Craske et al. 2014; Craske et al. 2008].
• Steps:
  1. Run exposure to cue A alone and cue B alone first, until each expectancy is lower.
  2. Then combine them, making sure both predict the same feared outcome.
  3. Examples from the sources: interoceptive arousal during in vivo exposure in panic (for example caffeine while shopping, or spinning then jogging); two spiders, then both together; an obsession plus a knife near the loved one in OCD [source: Craske et al. 2014]. For OCD, combine situational, imaginal and interoceptive cues, such as leaving the house with the oven preheated while imagining not knowing whether it is off and hyperventilating [source: Jacoby & Abramowitz 2016].
• Example language (illustrative): "You've done the spinning and you've done the crowded shop. Now we put them together: get your heart racing, then go straight into the shop."
• Watch for: a spike in the expectancy rating, which is expected and is the point. Compounding without first extinguishing each cue alone ("super extinction") produced strong relapse in one study, so do the sequential steps [source: Jacoby & Abramowitz 2016].

4.12 Occasional reinforced extinction

• Purpose: including occasional real (safe) aversive outcomes makes the new learning less vulnerable to a single bad experience later and sustains arousal [source: Craske et al. 2014]. The authors say evidence is still preliminary [source: Craske et al. 2014].
• Steps:
  1. Choose only aversive outcomes that are safe, temporary and agreed, such as mild social rejection, embarrassment or a full panic attack.
  2. Design exposures that make the outcome likely. Social examples are "shame attacks": the 2014 social anxiety case rode an elevator for 30 minutes calling out floors [source: Craske et al. 2014].
  3. Encourage clients to seek occasional negative outcomes in daily life (ask for something likely to be refused) [source: Craske et al. 2014].
  4. Debrief on coping: the outcome happened, and how bad was it, really, and for how long?
  5. Never use it where the outcome is dangerous or could retraumatise [source: Craske et al. 2014].
• Example language (illustrative): "This week the goal is to get turned down at least once. If it happens, the experiment worked."
• Watch for: the client experiencing the outcome as a catastrophe. Process it, and do not retreat to safety behaviours.

4.13 Variability (stimuli, intensity, duration, order)

• Purpose: variability improves retention and provides more retrieval routes, offsetting fear return in new situations [source: Craske et al. 2014; Craske et al. 2008].
• Steps:
  1. Vary stimuli (different dogs, different audiences, different contaminants).
  2. Vary intensity and duration within and across sessions; with agreement, use unpredictable lengths [source: Craske et al. 2014].
  3. Select menu items out of rank order, or at random by prior agreement, to teach that any level of anxiety is tolerable [source: Jacoby & Abramowitz 2016]. Often start with a moderately difficult item to support buy-in [source: Craske et al. 2014].
  4. Consider choosing items by life interference and values [source: Jacoby & Abramowitz 2016].
  5. Rarely repeat the identical exercise. PROTECT-AD exercises were usually not repeated [source: Pittig et al. 2023].
• Example language (illustrative): "Today we'll pull an item from the list at random. Whatever comes up, we'll treat it as an experiment."
• Watch for: resistance to non-hierarchical work. Explain the reason carefully and allow some choice [source: Jacoby & Abramowitz 2016]. Experimental evidence on randomised order is mixed [source: Jacoby & Abramowitz 2016].

4.14 Multiple contexts

• Purpose: to counter renewal, where fear returns in a context different from the one it was extinguished in [source: Craske et al. 2014; Craske et al. 2008].
• Steps:
  1. Run exposures in different places, at different times of day and on different days, alone and with others, including unfamiliar places [source: Craske et al. 2014].
  2. Prioritise contexts where the client will meet the feared cue in real life [source: Craske et al. 2008].
  3. Vary internal states where safe (tired, after caffeine for non-panic clients, after exercise) [synthesis].
  4. In therapy, vary rooms, therapists or online sessions, as in the Dutch PTSD study [source: Kooistra et al. 2025].
• Example language (illustrative): "You've done this in the clinic. Now let's do it at night, in a part of town you don't know, on your own."
• Watch for: benefits have been inconsistent across studies [source: Craske et al. 2014]. Treat it as low-cost insurance, not a guarantee.

4.15 Spacing (expanding schedule)

• Purpose: gradually widening gaps between sessions or practices may reduce fear return [source: Craske et al. 2008].
• Steps:
  1. After the early, closer-spaced exposure sessions, widen the intervals (for example twice weekly, weekly, then every other week) [source: Jacoby & Abramowitz 2016].
  2. Keep sessions close together or massed when response prevention is weak, family accommodation is high or the rationale is poorly understood [source: Jacoby & Abramowitz 2016].
  3. PROTECT-AD found the same outcomes for 3 times per week and weekly delivery, so intensified scheduling is a legitimate option [source: Pittig et al. 2023].
• Example language (illustrative): "From now on we'll meet every two weeks; the gaps are part of the training."
• Watch for: relapse in long gaps. The optimal schedule is unknown [source: Jacoby & Abramowitz 2016].

4.16 Retrieval cues and mental reinstatement

• Purpose: to help the client retrieve the new learning in later situations where the old fear may resurface [source: Craske et al. 2014; Craske et al. 2008].
• Steps:
  1. Use mainly late in treatment, as a relapse-prevention skill. Do not use early, because cues can lower the expectancy and blunt violation [source: Craske et al. 2014].
  2. Prefer mental reinstatement: before entering a feared situation, the client briefly recalls the treatment context and a specific successful exposure, including what was predicted and what happened. This was more robust than physical cues [source: Craske et al. 2014; Craske et al. 2008].
  3. If a physical cue is used (for example a wristband worn during exposures), use it sparingly. Make sure it reminds the client of the learning, not of protection [source: Craske et al. 2014].
• Example language (illustrative): "Before you walk into the meeting, take ten seconds to remember the elevator exercise: what you predicted, and what actually happened."
• Watch for: the cue turning into a safety signal ("I can only do it with the wristband"). If so, fade it and test without it.

4.17 Positive affect

• Purpose: the 2022 update lists positive affect as a generalisation-promoting strategy [source: Craske et al. 2022 abstract]. A UCLA experimental protocol compares rehearsal of the positive emotional aspects of exposures with neutral factual rehearsal [source: NCT06132425 protocol]. How the 2022 paper recommends doing this clinically is [uncertain].
• Steps [synthesis]:
  1. In the debrief or mental rehearsal, invite the client to notice positive feelings: relief, pride, satisfaction at having done it.
  2. Where it fits, build exposures into rewarding, valued activities (for example a dog-phobic client walking a friend's dog in a park they enjoy).
• Example language (illustrative): "Alongside the fear, was there anything good in that? Pride, relief, a moment you enjoyed?"
• Watch for: positive feelings used as a reason to stop at a "good" moment, or relief that comes from escape rather than from completing the exposure.

4.18 Targeting tolerability expectancies for distal or unverifiable fears

• Purpose: when the core fear cannot be disconfirmed in session (going to hell, future illness, harming someone years from now), test near-term expectancies instead [source: Jacoby & Abramowitz 2016].
• Steps:
  1. Say openly that the distal outcome cannot be proven or disproven, and do not try to.
  2. Identify near-term predictions: "I won't be able to stand the uncertainty", "the thought will drive me mad", "the anxiety will last all day", "the disgust will never fade" [source: Jacoby & Abramowitz 2016].
  3. Rate and test these. For example, the client stays in a pesticide aisle while deliberately holding uncertainty about future cancer [source: Jacoby & Abramowitz 2016].
  4. For disgust and "not just right" experiences, which habituate less readily, test the belief that the feeling is unbearable or permanent, not whether it goes away [source: Jacoby & Abramowitz 2016].
  5. Use imaginal exposure to the distal feared scenario where useful (see 4.19).
• Example language (illustrative): "We can't know about twenty years from now. What we can test today is your prediction that you couldn't stand not knowing for an hour."
• Watch for: reassurance-seeking about the distal outcome slipping back in; mental neutralising.

4.19 Imaginal exposure in an inhibitory-learning frame

• Purpose: to expose to memories, images or feared scenarios that cannot or should not be produced in real life [source: Craske et al. 2014; Jacoby & Abramowitz 2016].
• Steps:
  1. Specify what the client predicts will happen during or after the imagery, such as losing control, going mad or being unable to function afterwards. Rate it 0–100 [source: Craske et al. 2014].
  2. Run the imaginal exposure for a predetermined time set by the test. The 2014 PTSD case used 20 minutes, then minor tasks to test the prediction of not being able to function [source: Craske et al. 2014].
  3. Add affect labelling (4.10) and keep attention on the image.
  4. For PTSD, vary memories and move to hotspots early. The Dutch PTSD study alternated the worst memory with randomly drawn others [source: Kooistra et al. 2025].
  5. For OCD, imaginal scripts of the feared scenario can be followed immediately by in vivo exposure to the related cue (compound). In the 2014 OCD case, imagined harm was followed by a hand on the infant's neck [source: Craske et al. 2014].
  6. Debrief as in 4.8.
• Example language (illustrative): "You expect that if you go through the memory fully, you won't be able to do anything for the rest of the day. Let's find out: 20 minutes of the memory, then we'll see what you can do."
• Watch for: dissociation and overmodulation (flat, detached narration), which mean the client is not attending to the material. Prolonged exposure for PTSD should follow a recognised protocol; inhibitory-learning strategies are adapted onto it, not a replacement [synthesis].

4.20 Interoceptive exposure

• Purpose: to violate expectancies about bodily sensations (panic, health anxiety, some PTSD and OCD presentations) and to serve as one element of compound exposure [source: Craske et al. 2014; Jacoby & Abramowitz 2016].
• Steps:
  1. Complete medical screening (see 4.0).
  2. Identify the feared consequence of the sensation (fainting, heart attack, losing control).
  3. Pick an induction that produces the feared sensations, such as spinning, hyperventilation, running on the spot or breathing through a straw [the standard panic-treatment exercises; the 2014 paper uses spinning and jogging in its panic case].
  4. Rate the expectancy and run it long enough to test the prediction, not until sensations fade.
  5. Once single inductions are done, combine them with in vivo exposures (4.11).
• Example language (illustrative): "You predict that if you spin for a minute you'll collapse. Let's spin for a full minute and see."
• Watch for: the client stopping as soon as sensations peak, or sitting down "just in case" (a safety behaviour).

4.21 Self-guided exposure between sessions

• Purpose: to carry learning into daily contexts and build independence. In PROTECT-AD, at least one self-guided exposure followed each exposure-phase session, and self-management sessions focused on it. Therapist-guided exposures produced more expectancy violation and change than self-guided ones [source: Pittig et al. 2023].
• Steps:
  1. Assign exposures in contexts different from the session, using the same record: goal, feared outcome, expectancy before; occurrence, learning and adjusted expectancy after [source: Craske et al. 2014; Pittig et al. 2023].
  2. Remove companions or safety aids that served as safety signals [source: Craske et al. 2014].
  3. Vary timing and setting.
  4. Review the records at the start of every session; repeat a self-guided exposure in session when violation was weak.
• Example language (illustrative): "Same experiment as today, but on your own, at a different time and place. Fill in the prediction first and the result afterwards."
• Watch for: ratings filled in after the fact; exposures done with hidden safety behaviours; homework chosen only from easy items.

4.22 Relapse prevention and generalisation plan [synthesis built on 4.13–4.16]

• Purpose: to make the inevitable return of some fear expected and manageable [source: Craske et al. 2014; Jacoby & Abramowitz 2016].
• Steps:
  1. Teach that a return of fear is a retrieval problem, not proof that treatment failed.
  2. Write a plan: likely triggers for renewal (new places, stress, life events); a mental reinstatement routine (4.16); a rule to respond to any fear return by doing a varied exposure within days, not by avoiding.
  3. Schedule booster sessions (PROTECT-AD used boosters at 2 and 4 months) [source: Pittig et al. 2023].
• Example language (illustrative): "If the fear comes back after a move or a stressful month, that's the old memory resurfacing. Your job is to run an experiment that week, not to wait it out."
• Watch for: the client treating a lapse as collapse.

4.23 Retrieval-extinction (reconsolidation) timing: experimental, not routine

• Purpose: the 2014 paper reviews a laboratory procedure: a brief presentation of the feared stimulus about 30 minutes before sustained exposure, intended to update the original memory [source: Craske et al. 2014].
• Status: findings have not reliably replicated, and the effect may reflect retrieval rather than memory updating. The 2022 update discusses reconsolidation as an alternative model [source: Craske et al. 2014; Craske et al. 2022 abstract]. No clinical protocol recommending it is known [uncertain]. Do not build routine practice around it.
• Example language and watch-for: not applicable for routine use.

5. Handling problems

The source papers give limited troubleshooting, mostly through case examples [source: Craske et al. 2014; Jacoby & Abramowitz 2016]. Items tagged [synthesis] are my own guidance, consistent with the model.

Vague or untestable prediction. The client says "something bad" or "I'll get anxious". Ask for the observable version: who does what, by when. If the feared outcome is genuinely distal, switch to tolerability expectancies (4.18) [source: Craske et al. 2014; Pittig et al. 2023; Jacoby & Abramowitz 2016].

Low expectancy before exposure ("I know it won't happen here"). Make the test harder: lengthen it, combine cues, remove the companion or safety aid, or move to a context where the fear is stronger. In the 2014 cases, shorter exposures did not raise expectancy, so exposures were lengthened [source: Craske et al. 2014]. Check whether the therapist's presence is acting as a safety signal; if so, move toward self-guided or therapist-absent exposures [synthesis].

Fear stays high or rises, and the client wants to stop. This is expected and is not a reason to stop by itself [source: Craske et al. 2014; Craske et al. 2008]. Respond [synthesis]:

• Re-anchor on the test ("Has what you predicted happened yet?").
• Use affect labelling (4.10).
• Remind the client of the agreed goal and how much is left.
• If the client must stop, stop without reassurance. Record the expectancy at that point and debrief: what did they predict would happen if they stayed? That becomes the next test.
• Re-attempt soon, varied if needed. Avoid a pattern where escape ends sessions.

Overwhelm beyond fear (dissociation, rage, uncontrollable crying, panic that blocks attention). [synthesis] Learning needs attention to the cue. Ground the client enough to restore attention (eyes open, describe the room, feel feet on the floor). Then return to the cue at a level that allows attention, without adding safety behaviours. In PTSD, monitor for dissociation throughout imaginal work; persistent overwhelm suggests a need for stabilisation or a different protocol pace. Do not use grounding as an escape that ends the exposure every time.

Expectancy violated but barely updated ("it didn't happen, but next time it might"). PROTECT-AD found learning rates varied widely, and only expectancy change, not violation alone, predicted outcome [source: Pittig et al. 2023]. Respond [synthesis]:

• Identify the discount ("luck", "the therapist was there", "I did X"), then design an exposure that removes it.
• Repeat across varied stimuli and contexts.
• Use the adjusted-expectancy question and mental rehearsal explicitly.
• Check for hidden safety behaviours, rituals or mental neutralising.

Covert safety behaviours and reassurance-seeking. Ask after every exposure what the client did to stay safe. Name the subtle ones (distraction, scripted speech, holding on, checking a phone, mental reviewing). In OCD, maintain full response prevention [source: Craske et al. 2014; Jacoby & Abramowitz 2016]. When reassurance is sought from the therapist, decline it warmly and return to the experiment [synthesis].

Family members or partners acting as safety signals or accommodating. Plan exposures without them. In the 2014 OCD case, the spouse's presence was removed [source: Craske et al. 2014]. Where accommodation is heavy, use massed, supervised sessions and involve the family in reducing it [source: Jacoby & Abramowitz 2016; synthesis on family involvement].

Resistance to random order or unpredictable lengths. Explain desirable difficulties and why real life is unpredictable. Offer choice within limits (the client chooses from three items). Start with a moderately difficult item. Agree on the principle in advance [source: Jacoby & Abramowitz 2016; Craske et al. 2014]. If refusal persists, use a more graded but still varied order rather than losing the client [synthesis].

Feared outcome actually happens (rejection, a panic attack, a mistake noticed). Treat it as occasional reinforced extinction. Debrief how bad it was, how long it lasted and how the client coped. This protects against later relapse after real bad experiences [source: Craske et al. 2014]. If the outcome caused real harm, stop, review the plan and the safety screen, and repair [synthesis].

Post-event rumination (especially social anxiety). Redirect to concrete evidence about whether the predicted outcome occurred [source: Craske et al. 2014]. Use structured mental rehearsal (4.9) rather than open-ended replay [synthesis].

Non-adherence to self-guided exposures. [synthesis] Find out which part failed: planning, starting, completing or recording. Make the assignment smaller in scope but not safer (keep safety-behaviour removal). Pre-commit to a time and place. Do the first instance in session or on a call. Consider closer spacing until the client completes exposures reliably [source for massed sessions: Jacoby & Abramowitz 2016].

Client asks for (or the clinician habitually uses) cognitive restructuring before exposure. Explain that lowering the prediction beforehand leaves less to learn. Move the cognitive work to after the exposure [source: Craske et al. 2014; Jacoby & Abramowitz 2016]. Note that this caution is theory-based and not directly tested [source: Jacoby & Abramowitz 2016].

Alliance ruptures ("you're pushing me", "you don't care how I feel"). [synthesis] Acknowledge the distress and validate that the method is demanding. Restate the rationale and the client's goals. Reaffirm the client's right to stop. Renegotiate the next exposure collaboratively, keeping a real test in it. Avoid two failure modes: the therapist becoming a reassurer (a safety signal), and the therapist escalating without consent.

Therapist's own discomfort. [synthesis] Clinicians often end exposures early, reassure or pick easy items. Use supervision, plan the exposure length before starting, and use the record (prediction and outcome) as the shared anchor rather than the client's moment-to-moment distress.

Return of fear after treatment. Normalise it as renewal, spontaneous recovery or reinstatement. Use mental reinstatement, run varied exposures promptly and offer booster sessions [source: Craske et al. 2014; Pittig et al. 2023].

Disgust or "not just right" feelings that do not fade. Do not wait for habituation. Test the belief that the feeling is intolerable or permanent [source: Jacoby & Abramowitz 2016].

Telehealth delivery. [synthesis] This is feasible (the OptEx trial moved to Zoom [source: NCT04048824 protocol]). Plan exposures in the client's real contexts, agree in advance what to do if the connection drops, and watch more closely for off-camera safety behaviours.

Children and adolescents. [synthesis; uncertain] The principles carry over, but the primary sources reviewed are mostly adult. Simplify the prediction language, involve parents in reducing accommodation, and keep the consent process developmentally appropriate.

Sources

Sources:

• Maximizing exposure therapy: An inhibitory learning approach (Craske et al. 2014, PMC full text)
• Craske et al. 2014 record, Association for Contextual Behavioral Science
• Optimizing inhibitory learning during exposure therapy (Craske et al. 2008, PDF)
• Optimizing exposure therapy with an inhibitory retrieval approach and the OptEx Nexus (Craske et al. 2022, Caltech record; abstract only)
• OptEx trial protocol NCT04048824: Inhibitory Learning vs. Habituation (UCLA)
• Reward Processes and Rehearsal in Exposure Therapy, protocol NCT06132425 (UCLA)
• Jacoby & Abramowitz 2016, Inhibitory learning approaches to exposure therapy: translation to OCD (UNC repository)
• Pittig et al. 2023, Change of threat expectancy as mechanism of exposure-based psychotherapy (HU Berlin repository)
• Kooistra 2026 doctoral thesis chapter: Inhibitory retrieval-based exposure therapy for PTSD (Leiden)
• Heinig et al. 2017 PROTECT-AD design paper record (TU Dresden)

Cited by author, year, title:

• Craske, M. G., Kircanski, K., Zelikowsky, M., Mystkowski, J., Chowdhury, N., & Baker, A. (2008). Optimizing inhibitory learning during exposure therapy. Behaviour Research and Therapy, 46, 5–27.
• Craske, M. G., Treanor, M., Conway, C. C., Zbozinek, T., & Vervliet, B. (2014). Maximizing exposure therapy: An inhibitory learning approach. Behaviour Research and Therapy, 58, 10–23.
• Craske, M. G., Treanor, M., Zbozinek, T. D., & Vervliet, B. (2022). Optimizing exposure therapy with an inhibitory retrieval approach and the OptEx Nexus. Behaviour Research and Therapy, 152, 104069. [uncertain]
• Jacoby, R. J., & Abramowitz, J. S. (2016). Inhibitory learning approaches to exposure therapy: A critical review and translation to obsessive-compulsive disorder. Clinical Psychology Review, 49, 28–40.
• Pittig, A., Heinig, I., Goerigk, S., et al. (2023). Change of threat expectancy as mechanism of exposure-based psychotherapy for anxiety disorders. Clinical Psychological Science, 11(2), 199–217.
• Heinig, I., Pittig, A., Richter, J., et al. (2017). Optimizing exposure-based CBT for anxiety disorders via enhanced extinction: Design and methods of a multicentre randomized clinical trial. International Journal of Methods in Psychiatric Research, 26, e1560. [record only]
• Kooistra, M. J., Oprel, D. A. C., Schoorl, M., van der Does, W., & de Kleine, R. A. (2025). Inhibitory retrieval-based exposure therapy for patients with posttraumatic stress disorder. Cognitive and Behavioral Practice. [read via thesis chapter]