DBT is the one modality honest enough to put this problem in writing. Most of us teach a skill and quietly hope it shows up when the client needs it. Marsha Linehan looked at that hope, decided it wasn't a plan, and built an entire function of treatment around the gap — the recognition that teaching a skill and getting a skill used in the wild are two different jobs, and the second one needs its own machinery.
I'm not a DBT purist and this isn't a fidelity lecture. But the way DBT names generalization is the clearest map I know for a problem every modality has and almost none of them admit to. Worth reading the model on its own terms before we talk about what the rest of us do in solo practice.
Generalization is a separate function of treatment — not a hope
Comprehensive DBT is built to serve five functions: enhancing the client's motivation, building their capabilities, generalizing those capabilities to the natural environment, structuring the environment to reinforce progress, and keeping the therapist motivated and skillful (the consultation team). Generalization sits in that list as a peer to skill-building, not a downstream consequence of it (Behavioral Tech Institute).
To deliver those functions, standard DBT runs four modes at once: individual therapy, group skills training, the consultation team, and phone coaching between sessions. The fourth mode — phone coaching — is the one most directly assigned to generalization. It exists because a skill rehearsed in a calm Tuesday-night group does not automatically appear in a Saturday-night crisis. The whole point of intersession coaching is to catch the client in the natural environment, at the moment the skill is supposed to fire, and walk them into it before the old pattern wins.
That's the structural insight worth stealing even if you never run a DBT program: the room where a skill is learned and the room where it's needed are different rooms, in different nervous-system states. Standard DBT staffs that gap on purpose. Most of the rest of us leave it unstaffed and call the result a motivation problem — which is really the general skills-generalization problem between sessions that DBT happens to name out loud.
Why the skill dies between group and the crisis moment
The reason this isn't a willpower issue is the same reason insight doesn't survive the trip home. What a client encodes in one physiological and contextual state is harder to retrieve in another — state- and context-dependent memory is well established in the basic literature (Godden & Baddeley, 1975; Tulving & Thomson, 1973). A client learns TIPP or paced breathing or a DEAR MAN script in a regulated, supported, low-stakes setting. Then the skill is supposed to be retrieved in a flooded, alone, high-stakes one. The retrieval cues that were present at encoding — the group, the worksheet, the co-regulation, you — are all gone exactly when the skill matters most.
So the failure isn't "they didn't practice." Often they practiced fine in group. The failure is that the skill is welded to a context the client can't reproduce at 11 PM. This is the nervous-system-regulation problem between sessions wearing a DBT uniform: when the sympathetic system is running the show, the prefrontal access the skill depends on is the first thing offline. You cannot reason your way to DEAR MAN from inside a fight-or-flight state you've already lost.
Phone coaching works on exactly this seam: a brief, in-the-moment contact — Linehan's teams describe these as short, often in the ten-to-twenty-minute range — that supplies the missing cue and re-regulates enough that the skill becomes reachable again. It's not therapy on the phone. It's a bridge from the encoding context to the retrieval context.
The in-vivo coaching gap solo practitioners actually face
Here's the part the DBT literature doesn't have to solve, because it assumes a team. You probably don't have one.
If you're a solo clinician delivering DBT-informed work — and most clinicians using DBT skills are — you're running, at most, two of the four modes. You teach skills in session; maybe you assign diary cards. But the generalization mode, the fourth one, built specifically because skills don't transfer on their own, is the one you can't realistically staff. Almost nobody can sustain genuine 24/7 phone coaching for a full caseload. The honest math of solo private practice doesn't include being on call.
So the modal reality looks like this:
| What full-fidelity DBT staffs | What solo / DBT-informed practice usually has |
|---|---|
| Individual therapy | Individual therapy ✓ |
| Group skills training | Skills taught in session (often no group) |
| Consultation team | A peer or two, informally |
| Phone coaching (the generalization mode) | Nothing — the skill is left to the 167 hours |
That bottom row is the whole problem. You've identified that a skill needs to generalize, you've taught it well, and then the structural mechanism for generalization — the thing the model calls a function of treatment, not a luxury — simply isn't there. The skill gets taught and then abandoned to the week. This is the same continuity gap behind why insight doesn't stick and the broader between-session engagement problem; DBT just makes the missing piece impossible to pretend isn't missing.
What does help short of full coaching: diary cards, skills rehearsed in-session rather than assigned cold, crisis plans written while the client is regulated, and explicit "when X happens, the move is Y" pairing. None of these reproduce a full coaching call; all of them narrow the gap. The question for a solo practice is how to get some of the generalization function without pretending you can offer all of it.
A skills cue is not a chatbot a client confides a crisis to
This is where I have to be careful, because the obvious-sounding fix is the dangerous one. "Give the client an app that talks them through the skill in the moment" slides, almost immediately, into "give the client an AI they confide a crisis to." Those aren't the same product, and the difference is a safety line, not a feature comparison.
A general-purpose AI companion that a dysregulated client turns to in a real crisis is a clinical hazard. The data isn't subtle: licensed clinicians respond appropriately to expressions of distress around 93% of the time, versus under 60% for general-purpose AI bots (Stanford HAI / ACM FAccT, 2025). A bot that takes the chair in the worst moment — absorbs the disclosure, attempts to counsel, replaces the human safety net — is solving the generalization problem by introducing a far worse one. The DBT case is the sharpest version of why AI belongs as a clinical tool and not a replacement: phone coaching's whole design is to keep the client connected to you and the skill, not to hand them off to a machine.
A between-session pattern mirror is a different object entirely. It doesn't counsel, advise, or hold a crisis. It reflects — surfaces the pattern the client is already in, in their own words, and points them back toward the skill you taught and the room you hold. The distinction is what the thing is for:
- A skills cue / mirror notices the rising pattern and reflects it back, so the client recognizes "this is the moment for the skill we worked on" and carries the material to the next session. It re-supplies the missing retrieval cue. It does not pretend to be the clinician.
- A crisis chatbot receives the disclosure and tries to manage it — stepping into the clinical role, with no license, no accountability, and a documented failure rate in exactly the situation where failure is most costly.
The first one is a partial, honest substitute for the generalization function. The second one is a liability wearing the costume of help.
How do I know which skill the client actually needs in the moment?
This is the quiet hard part of generalization. The skill a client practiced and the skill a client needs are often not the same, because in the moment the client can't read their own state accurately. Someone flooded reaches for problem-solving (DEAR MAN, fixing the situation) when the state actually calls for distress tolerance (TIPP, getting the body back online first). They try to negotiate a boundary while their nervous system is screaming, it doesn't work, and they conclude the skills don't work.
A mirror that reflects state — "your body is activated right now" — before the client picks a skill does the most useful thing intersession support can do: it gets the order right. Regulate, then choose. That's closer to what a coaching call does in its first thirty seconds — orient to state — than any skill-vending interface. It's a relative of the logic behind therapy homework that clients actually do: the support that works meets the client where they actually are, not where the worksheet assumed they'd be.
FAQ
What is generalization in DBT and why is it a separate mode of treatment?
Generalization is one of the five functions of comprehensive DBT — getting skills to transfer from where they're learned to the client's everyday life. It's treated as its own function (and staffed primarily through phone coaching, one of DBT's four modes) because Linehan's model takes seriously that skills don't transfer on their own. Teaching a skill and getting it used in a real crisis are two different jobs (Behavioral Tech Institute).
Why don't DBT skills transfer from group to the moment a client is dysregulated?
Because of state- and context-dependent retrieval. A skill encoded in a calm, supported group is welded to that context; when the client is flooded and alone, the cues present at learning are gone and prefrontal access is offline (Godden & Baddeley, 1975; Tulving & Thomson, 1973). It's not a practice failure — it's that the skill is hard to reach from the exact state it was meant for.
Do I have to offer phone coaching to deliver DBT skills effectively?
Full-fidelity DBT includes it; most solo and DBT-informed practice can't sustain genuine 24/7 coaching for a full caseload. The realistic move isn't all-or-nothing. You can serve part of the generalization function with in-session skill rehearsal, diary cards, crisis plans written while regulated, and between-session cues — while being honest that these don't replace a true coaching arm.
How can solo or DBT-informed practitioners support skills generalization without 24/7 coaching?
Rehearse skills in-session rather than assigning them cold; pair skills explicitly to triggers ("when X, the move is Y"); write crisis plans while the client is regulated; and give the client a way to re-access the right skill mid-week. The aim is to re-supply the retrieval cue that group can't travel with — not to be reachable around the clock.
What's the difference between a skills cue and an AI chatbot handling a crisis?
A skills cue or mirror reflects the client's pattern and state and points them back to the skill you taught and the room you hold — it doesn't counsel. A crisis chatbot tries to take the clinical role, which is where the risk lives: general-purpose AI bots respond appropriately to distress under 60% of the time versus around 93% for clinicians (Stanford HAI / ACM FAccT, 2025). One narrows the generalization gap; the other replaces the safety net with something measurably worse.
How do I know which skill a client needs in the moment versus what they practiced?
Get the order right: state first, skill second. A flooded client tends to reach for problem-solving when the moment actually calls for distress tolerance. Anything that helps the client read their state before they pick a skill — including a between-session mirror that reflects "you're activated right now" — does the same first move a coaching call does: regulate, then choose.