- CBT
- treatment planning
Graded exposure: how to build the hierarchy, and why "stay until the anxiety comes down" is no longer the rule
Wherever you trained, you were probably handed a stopping rule for exposure: stay in the situation until the anxiety comes down, by half or to about 2 out of 10. It is written into an English-language practice-software template, into that template’s Spanish translation, and into Spanish university and popular-psychology material; a 2025 review of the SUDS calls it “the common 50% reduction heuristic for habituation.” The rule has a real pedigree, Foa and Kozak’s emotional processing theory of 1986, but the data since do not support the drop as the condition for ending a trial. In Craske and colleagues’ 2014 review, “neither fear reduction nor ending fear levels predict long term outcome,” and “the end of an exposure trial is determined by conditions that violate expectancies and not by fear reduction.” You still build a hierarchy, and SUDS can still rank it. Each rung simply gains one more entry: what the client expects to happen, and how likely they think it is. Fear falling within the session is not required; testing that prediction is.
What circulates, in English and in Spanish
Carepatron’s Graded Exposure Worksheet, dated August 2025, puts it in step 4: “Have your client interact with this fear until their fear rating decreases by half or more. Once comfortable with this level, guide them to the next step.” The page cites no source. Its Spanish version carries the same instruction: «hasta que su índice de miedo disminuya a la mitad o más».
In Spanish, the same rule appears in material a clinician in Mexico is likely to meet, including a university handout. The handout, on the student-guidance site of Spain’s Universidad Carlos III de Madrid and signed by Alfonso Fernández-Martos Abascal, reads:
«Mantenerse en la situación tanto tiempo como sea necesario hasta que la ansiedad subjetiva sentida sea mínima (alrededor de un 2 en una escala de 0 a 10).»
(Stay in the situation as long as necessary until subjective anxiety is minimal, around 2 on a 0 to 10 scale.) It adds «No pase a la siguiente situación de la lista hasta que la ansiedad sea mínima» (do not move to the next item until anxiety is minimal).
The Spanish popular-psychology site Psicología y Mente says in its exposure and response prevention article (updated 12 June 2026) that the person «deberá permanecer en la situación hasta que la ansiedad se rebaje como mínimo a la mitad» (must remain in the situation until anxiety drops by at least half). To be fair, the same text adds that «el objetivo no es necesariamente que desaparezca la ansiedad» (the aim is not necessarily for the anxiety to disappear), but the trial’s length is still set by the drop.
Where the rule came from, and what the data showed
In Psychological Bulletin in 1986, Foa and Kozak proposed “three indicators of emotional processing”: activation of fear, “short-term (within-sessions) habituation, and long-term (across-sessions) habituation.” They added that “Long-term habituation must be preceded by short-term habituation.” On that premise, staying until it drops is the logical instruction.
Baker and colleagues (2010), who note the theory’s “wide subscription” among clinicians, tested it with in vivo exposure in 44 students scoring high on acrophobia measures. They found no support for initial fear activation or within-session habituation as predictors; between-session habituation predicted short-term change, but the effect was attenuated at follow-up. A small analogue sample, not patients; Craske and colleagues set it alongside animal and laboratory work and conclude that “the amount by which fear reduces by the end of an exposure trial or series of exposure trials is not predictive of the fear level expressed at follow-up assessment in fearful human samples.” They even name the instruction: exposure tasks are designed around “what do you need to learn” rather than “stay in the situation until fear declines.”
Not everything points one way. A 2017 meta-analysis (Rupp and colleagues, 21 studies) found within- and between-session habituation positively related to outcome, though the within-session estimate rests on only five studies and 116 participants, and concluded that its results “do not specifically strengthen EPT,” Foa and Kozak’s theory, “while matching other theoretical perspectives such as inhibitory learning and reality testing.” Nor does Craske say habituation is irrelevant: “Inhibitory learning is regarded as being central to extinction, although additional mechanisms, such as habituation, are likely to be involved.” The asymmetry fits in a line: falling fear is a product of the learning, not the condition for ending the trial.
Building the hierarchy
- Specific situations. The Madrid handout gets this right, with an example worth copying: «Ir sentado en el metro con poca gente (un 25% de las plazas ocupadas del vagón)» (sitting on the metro with few people, a quarter of the seats taken).
- The feared outcome for each. Not “I’ll get nervous” but what the client believes will happen. For social anxiety, Craske notes that predicting they will “get anxious” “would not be sufficient,” whereas predicting being ignored or rejected would.
- Two numbers per row. How much fear the item provokes (SUDS) and how likely the feared outcome seems, 0 to 100. In the review’s PTSD case, the hierarchy “contained distress and expectancy ratings.”
- A start the client will accept. In the OCD case they began with moderately difficult exposures, because “beginning with moderately difficult exposures increases the likelihood of success and facilitates patient buy in.”
- Not a straight line. The most pressing thing to learn sets the order, not the SUDS. The social anxiety client began with social events alongside his wife, rated as more frightening than dealing with grocery clerks, because he most needed to learn whether her friends would humiliate him.
- Each rung tied to the plan. “Ride the metro at rush hour without a companion” is a goal that fits a SMART treatment plan and can be verified.
SUDS: what it is for, and what it is not
Wolpe and Lazarus introduced the Subjective Units of Distress Scale in 1966, running from 0 (absolute calm) to 100 (the worst anxiety the person has felt or can imagine); Spanish handouts tend to use 0 to 10. It is useful for ranking situations and spotting gaps: if one rung is a 20 and the next an 80, a step is missing. That example comes from Mattera and Zaboski, whose 2025 paper is a theoretical construct-validity review, not a trial, and which concludes that “caution is warranted when interpreting SUDS scores.”
They suggest naming what is being rated (“Rate your fear of contamination from 0–100 right now”) rather than a generic “distress” and, for change across sessions, relying on validated multi-item questionnaires for the construct you are treating: the GAD-7 for generalized anxiety, or a disorder-specific questionnaire, administered regularly, as in measuring progress session by session. A trial’s SUDS is not your outcome measure. They also propose an expectancy rating from 0 (“will not occur”) to 100 (“definitely will occur”).
The written prediction: what ends a trial
Before each trial, in writing: the goal in behaviour and duration, what the client fears will happen, and how likely it seems, as in Craske’s case tables (“What are you most worried will happen?”, “On scale 0–100, how likely does this seem?”).
The prediction sets the duration. A client with a dog phobia predicted she would be bitten if she stood within 10 yards of a dog for 10 minutes on the sideline of a soccer match; she was challenged to stay five minutes longer, and at the end her fear “had not decreased substantially.” The trial had done its job. A client with panic disorder, whose medical tests had come back favourable, was 85% sure that 30 seconds of dizziness would bring on a stroke or heart attack; his spinning exercise lasted 60 seconds. Afterwards come the review’s questions: what did you fear would happen, what happened, “How was that surprising?” and “What did you learn?”
Craske and colleagues also move cognitive restructuring to after the trial. Lowering the expectancy beforehand (persuading the client that dogs rarely bite) narrows the gap between prediction and outcome, and may be “deleterious.” The Madrid handout says the opposite on the cognitive side: «Antes de afrontar la exposición sí se puede usar la relajación y la terapia cognitiva» (before facing the exposure, relaxation and cognitive therapy may be used). The review says nothing about relaxation, and its position on restructuring is a theoretical implication, not the result of a head-to-head trial.
When a client asks what happens if the anxiety does not come down, this explanation adapts the review’s “test it out”:
“We’ll use the 0 to 10 to put the list in order, not to decide when you stop. Before each practice you’ll write down what you’re afraid will happen and how sure you are, from 0 to 100. The practice lasts as long as we agreed it needs to test that, and afterwards we look together at whether it happened. Sometimes you may finish still feeling afraid, and that doesn’t mean it went badly: over time the fear usually does come down, as a result of what you find out. We don’t do any practice we haven’t agreed on first.”
That explanation belongs in informed consent: exposure involves temporary discomfort, and the client should know it before the first practice.
Safety signals: phase them out on a plan
The review lists the common ones: “the presence of another person, therapists, cell phones, medications, or food or drink.” It adds distraction and, for clients who fear fear itself, the drop in fear.
The evidence is less tidy here: the review concedes that “recent data have presented contradictory findings,” and Blakey and Abramowitz (2016) call the findings “limited, mixed, and controversial.” In England, NICE’s guideline on social anxiety builds into the Clark and Wells model exercises “to demonstrate the adverse effects of self-focused attention and safety-seeking behaviours” (CG159, 1.3.13). The consensus Craske reports is gradual withdrawal, with gradualness recommended “only in order to reduce treatment attrition”; “If willing, immediate removal of safety signals is preferred.”
In the review’s panic case, from session 4 the client’s wife stopped waiting outside and he stopped bringing his benzodiazepines to sessions. Not carrying them is not the same as stopping them: any change to how a prescribed medication is taken is the prescriber’s decision, not part of the exposure plan. An “accompanied” rung can be the first one, provided the next, unaccompanied rung is already written into the hierarchy. Blakey and Abramowitz define safety behaviours as actions “to prevent, escape, or minimize feared catastrophes and/or associated distress”; asking someone for reassurance fits, including when the someone is a chatbot at 3 a.m., which we take up in patients using chatbots between sessions.
Variability: stimuli, places and times
Traditional exposure repeats each item “until anxiety decreases” before moving on. The alternative varies stimuli, durations and intensities and takes items out of order, “without regard to fear levels or fear reduction.” In spider-fearful samples, progressively longer intervals between sessions gave better follow-up results than massed exposure.
The same goes for contexts: alone, in unfamiliar places, at different times and on different days, because extinction tends to stay tied to where it happened, such as the therapist’s presence. The review is candid that one rodent study and one human study found no benefit, and calls the effects “unstable.”
In vivo or imaginal
The review names three forms, “imaginal, interoceptive or in vivo,” and its cases show the criterion: expose the client to what they actually fear. A situation that can be approached safely (the metro, a meeting, a dog) goes in vivo. An image, a memory or a thought goes imaginal: in the OCD case the therapist built a detailed script of the intrusive images, and in the PTSD case imaginal exposure to the trauma memory was added to in vivo work from session 5.
In NICE guidance, the in vivo route appears in social anxiety treatment, with “graduated exposure to feared social situations, both within treatment sessions and as homework” (CG159, 1.3.14). The PTSD guideline, NG116, lists prolonged exposure therapy among trauma-focused CBT interventions (1.6.16) and says these should “be based on a validated manual” and “be delivered by trained practitioners with ongoing supervision” (1.6.17). Interoceptive exposure for panic calls for ruling out a medical cause of the symptoms first; the review does not state this as a rule. OCD, panic and trauma each deserve their own guide; here, only the principle fits.
What to record after each trial
The Madrid self-record tells you how far the anxiety fell, not what the client learned. A fictional example, built on the fields in Craske’s case tables, with rows added for safety signals, surprise and end anxiety:
| When | Field | Example |
|---|---|---|
| Before | Goal: behaviour and duration | Ride six metro stops alone at rush hour, standing, no earphones |
| Before | What I fear will happen | I’ll faint before I get there |
| Before | How likely, 0 to 100 | 80 |
| Before | Safety signals left behind | Companion, water bottle, music |
| After | Did it happen? How do I know? | No. I felt dizzy at the second stop and stood all the way to the sixth |
| After | Did I use any safety signal, before, during or after? (asking for reassurance, my phone, a chatbot) | No |
| After | What surprised me | The dizziness didn’t build |
| After | What I learned | I can feel dizzy on the metro and still arrive |
| After | Anxiety at the end, 0 to 10 (optional) | 7 |
In a BIRP note, the trial takes a few lines, split between R and P:
R — In vivo exposure between sessions: metro, six stops, unaccompanied. Predicted fainting (80%); did not occur; end anxiety 7/10. Learning, in her words: “I can feel dizzy and still arrive.”
P — Next trial: a different line, a different time.
That closing 7 is not a failure to explain away: the disconfirmed prediction is the result. If you use Gesell, attach your own log to the week’s task: the client checks it off and writes what happened; you record partial or not done, and the prediction-versus-outcome comparison goes in your note, as above. Whatever system you use, check how your jurisdiction expects homework and its review to appear in the record.
References
- Craske MG, Treanor M, Conway CC, Zbozinek T, Vervliet B. Maximizing exposure therapy: an inhibitory learning approach. Behaviour Research and Therapy, 2014; 58: 10–23 (author manuscript on PMC)
- Foa EB, Kozak MJ. Emotional processing of fear: exposure to corrective information. Psychological Bulletin, 1986; 99(1): 20–35
- Baker A, Mystkowski J, Culver N, Yi R, Mortazavi A, Craske MG. Does habituation matter? Emotional processing theory and exposure therapy for acrophobia. Behaviour Research and Therapy, 2010; 48(11): 1139–1143
- Rupp C, Doebler P, Ehring T, Vossbeck-Elsebusch AN. Emotional processing theory put to test: a meta-analysis on the association between process and outcome measures in exposure therapy. Clinical Psychology & Psychotherapy, 2017; 24(3): 697–711
- Blakey SM, Abramowitz JS. The effects of safety behaviors during exposure therapy for anxiety: critical analysis from an inhibitory learning perspective. Clinical Psychology Review, 2016; 49: 1–15
- Mattera E, Zaboski B. Rethinking the Subjective Units of Distress Scale: validity and clinical utility of the SUDS. Clinics and Practice, 2025; 15(7): 123
- NICE — Social anxiety disorder: recognition, assessment and treatment (CG159), 2013, recommendations
- NICE — Post-traumatic stress disorder (NG116), 2018, recommendations
Keep homework from getting lost between sessions
Assign the worksheet as homework in Gesell: your client sees it in their portal, marks it done, and leaves you a note before the next session.
Try gesell.aiHave a question about your practice?
How to document a case, which format suits you, or how Gesell fits your practice: write to us and we will answer.
About the author
Gesell Team
Clinical and product content written by the gesell.ai team together with certified clinical psychologists.