Talking can help you see a pattern. It cannot show you whether the awkward response you expect will actually happen. That is why the Odyssey Coach sometimes turns a session into one specific Challenge in your real life.
The Challenge is not there to keep you busy or prove that you are committed. It should answer a question. If you expect an old friend to find your message strange, sending it gives both you and your coach information that another twenty minutes of discussion cannot.
Research on cognitive and behavioral therapy gives between-session action a stronger empirical basis than many adjacent practices. The studies do not establish an effect for the Odyssey Coach or life coaching, but they do explain why action outside the conversation deserves a defined place in the work.
Between-session action has the strongest support
A 2010 meta-analysis by Nikolaos Kazantzis and colleagues compared cognitive and behavioral therapies that included homework with otherwise similar therapies that did not. Across 46 studies, the analysis found a moderate benefit for including between-session assignments (Kazantzis, Whittington, and Dattilio, 2010).
A separate meta-analysis examined whether completing assigned work was associated with better outcomes. Across 23 studies, homework completion had a small-to-moderate relationship with treatment outcome (Mausbach et al., 2010). This second result is correlational. People who are already doing better may complete more work, and other differences can affect both. Read together, however, the two reviews support a restrained conclusion: what happens between conversations matters, and follow-through is part of the mechanism.
These studies concern psychotherapy, mainly cognitive behavioral therapy. They do not establish an effect size for life coaching, AI coaching, or work on building a social life. The transfer is in the design principle, not in a claim that the numbers carry over unchanged.
A useful Challenge tests something
Between-session work is easy to get wrong. A vague instruction such as “be more social this week” gives no clear action and little to review. A punishing assignment can turn a useful stretch into another experience of failure. Repetition without a question behind it may produce endurance but no new understanding.
Research on behavioral experiments and exposure offers a better frame. Before acting, the person identifies a concrete prediction. After acting, they compare that prediction with what occurred. The purpose is to learn from the mismatch.
Michelle Craske and colleagues describe this as expectancy violation: new learning becomes possible when experience differs from the feared outcome (Craske et al., 2014). Their paper concerns exposure therapy for anxiety disorders, so it should not be turned into a universal rule for coaching. One lesson is still especially useful for social fears. Anxiety does not have to disappear during the attempt for the attempt to count. Someone can feel tense, send the invitation anyway, and learn that the feared response did not happen.
This changes the review question. “Did you feel confident?” is often less informative than “What did you expect, what happened, and what do you make of the difference?”
Roleplay prepares; it does not replace the attempt
Practice inside a session serves a different purpose. Rehearsing how to invite a classmate for coffee can reduce uncertainty and reveal the exact moment where someone gets stuck. It may make the real attempt more likely.
The evidence for roleplay as a distinct client intervention is much less settled than the evidence for between-session assignments. A recent systematic review found broad variation in how adult psychotherapy studies defined and used roleplay, often as one element inside a larger treatment. Small samples, mixed methods, and inconsistent reporting made it difficult to isolate what roleplay itself contributed (Mazzucato, Savastano, and Iudici, 2026). A meta-analysis of chairwork found encouraging results in individual psychotherapy, but chairwork is a particular family of experiential methods, not evidence that any simulated conversation will work (Pascual-Leone and Baher, 2023).
Roleplay therefore belongs in the preparation category. It can be useful without being presented as a proven substitute for the real conversation.
Messages can support the work, but their effects are small
Messages and check-ins can help someone remember a plan, prepare for an upcoming situation, or reflect while the details are fresh. The available evidence suggests modest effects.
A 2025 meta-analysis of 23 just-in-time and ecological momentary intervention studies found a small overall effect on mental-health outcomes. The authors found benefits for symptom measures but not for psychological well-being, and they reported moderate-to-high risk of bias around adherence and missing data (von Lützow et al., 2025). These interventions varied widely and were not all conversational coaching.
That evidence does not support making notifications or chat the engine of change. At most, contact helps carry the work back into daily life. The action itself still happens away from the app.
What you should expect from your coach
A Challenge should be specific enough to do, connected to the life you want, and calibrated to what you are ready to attempt. Your coach should be able to explain what the step is for, what counts as doing it, and what the two of you will examine afterward. It is not required at the end of every session.
When a Challenge tests a fear, your coach should make the prediction clear before the attempt and review the result afterward. Completion means you took the real-world action. Feeling anxious during the attempt does not make it a failure.
Roleplay remains preparation. Messages can help you get ready or think through what happened, but neither replaces the real-world step. The conversation should lead back to your life, where the useful evidence is.
Sources
- Kazantzis, Whittington, and Dattilio (2010), Clinical Psychology: Science and Practice
- Mausbach et al. (2010), Cognitive Therapy and Research
- Craske et al. (2014), Behaviour Research and Therapy
- Mazzucato, Savastano, and Iudici (2026), Frontiers in Psychology
- Pascual-Leone and Baher (2023), Psychotherapy
- von Lützow et al. (2025), BMJ Mental Health