Inspect the client turn immediately following eligible therapist moves

See how Intuita links a therapist turn to the client's next turn, with transcript evidence, raw counts and no claim of causality or effectiveness.

One exchange at a time

“Which interventions land?” is the question a therapist naturally asks. Intuita does not answer it with an effectiveness score or a recommendation. It observes something narrower:

What did the therapist do in one turn, how did the client respond in the very next turn, and what subject was the exchange about?

Only an adjacent therapist → client pair qualifies. A response that arrives later in the session, a change outside the room or an eventual treatment outcome cannot be attributed to that one move. “Followed by” is the right causal language. “Produced”, “worked” or “failed” is not.

The result accumulates through this client’s analyzed individual sessions up to the session being viewed. It is personal in the sense that the observations come only from this client’s sessions. It is not a separately trained personal model and it does not require a minimum sample before showing an observation.

Twelve therapist moves

Each therapist turn receives one dominant conversational technique:

1. Open question — invites an extended answer.
2. Closed question — seeks a bounded or factual answer.
3. Simple reflection — restates the client’s expressed meaning.
4. Complex reflection — extends or deepens meaning already present.
5. Interpretation — introduces a hypothesis not directly stated by the client.
6. Validation — acknowledges the legitimacy or understandability of experience.
7. Challenge — directly points to discrepancy or invites confrontation with a pattern.
8. Directive — gives an instruction, exercise or homework.
9. Psychoeducation — explains a concept or mechanism.
10. Self-disclosure — brings in the therapist’s own experience or reaction.
11. Process management — structures time, agenda or the practical frame.
12. Other — a last resort when no specific family fits.

A turn can contain more than one move. The classifier chooses the dominant therapeutic intent rather than pretending each sentence is pure. The taxonomy is deliberately general: this chain currently does not adapt its categories to a selected therapeutic modality.

Several boundaries are explicit. A complex reflection adds to meaning already present; an interpretation adds a new hypothesis. Teaching a concept is psychoeducation; organising the agenda is process management. A task or instruction is directive.

Twelve immediate response types

The following client turn is placed into one of twelve response categories and then into one of four readable bands.

Opened up

• Affective deepening: movement into present feeling.
• Cognitive exploration: tentative, curious discovery.
• Insight: a newly articulated connection.
• Change language: self-directed movement toward change.

Stayed with it

• Elaboration: develops the material without a clear shift in depth.
• Confirming: agrees or affirms.
• Information giving: supplies relevant facts.

Held back

• Narrative avoidance: tells a story without moving into the experience at issue.
• Intellectualisation: analyses from a distance rather than discovering in the moment.
• Deflection or minimisation: reduces, jokes away or redirects the significance.
• Topic shift: moves to another subject.

Pulled away

• Active resistance: explicitly rejects the therapeutic task, frame or move.

These bands describe the immediate turn. “Held back” is not a trait, and “opened up” is not proof that an intervention was clinically right.

Psychological distinctions designed into the rubric

The response taxonomy is more careful than counting words or emotion.

Volume is not depth. A long, cooperative story can still remain at the level of factual narrative. A short sentence can carry present emotional contact.

Discovery is not analysis. Tentative meaning-making can count as exploration. Confident, outside-looking-in psychological explanation can remain intellectualisation when it creates distance from experience.

Feeling is not automatically progress. Emotional intensity alone does not establish reflective contact or therapeutic benefit.

Relational protest is not automatically resistance. Separation distress, a demand for reassurance or anger directed toward the therapist can represent movement toward the relationship. Active resistance is reserved for rejection of the therapeutic task or frame. The distinction is bounded, but it prevents a common conceptual error: treating every uncomfortable response as withdrawal.

Bound to the topic

The action-and-response analysis receives the session’s topic-tree leaves as numbered context. Code, not the model, binds each stored pair to the corresponding topic label. This helps separate a genuine topic change from a colder response within the same subject.

On the cumulative Radar view, each response is rebound through the persisted topic analysis for its patient turn. If an exchange cannot be matched to a topic, it stays visible in a “Without topic” bucket. It is not silently discarded to make the per-topic chart look complete.

This creates two complementary readings:

• an overall distribution of techniques and response bands;
• a per-topic view of how the exchanges were distributed.

Raw counts preserve contradiction

The live surface uses counts, not one average.

Suppose a complex reflection is followed three times by deeper exploration, twice by ordinary elaboration and once by active disagreement. A single midpoint would smooth away the interesting part. Intuita keeps all six observations in their original bands.

In the overall view, a technique’s count appears inside every response band in which it occurred. In the per-topic view, the topic and band totals remain visible and the technique names are ordered by frequency. Because there is no minimum-sample gate, one observation remains one observation — not a stable personal tendency.

The distribution supports a question. It does not answer whether a difficult intervention was inappropriate. Productive therapy can contain discomfort, disagreement and delayed effects that an immediate-turn classifier cannot see.

How the pair is grounded

The engine walks the transcript mechanically to find non-empty therapist turns immediately followed by non-empty client turns. The AI makes two bounded judgments: the dominant therapist move and the immediate client response. Code assigns:

• the therapist segment;
• the next client segment;
• the four-band response tier;
• the topic label;
• timing where available;
• engine provenance.

The transcript surface can show a technique badge on the therapist turn and a response badge on the following client turn. Those exact turns can be opened for inspection. The cumulative Radar summary itself is a count view; it does not place a quote beneath every aggregate.

Coverage and degraded analysis

Long sessions are processed in bounded slices. The system checks whether the model returned the expected number of pair judgments and whether too many reactions are missing. Suspicious slices are retried in smaller pieces. Over-generated or persistently mismatched slices are excluded rather than forced into the record.

If too much of the eligible exchange stream remains uncovered, the result is marked degraded. Trusted pairs can remain available while incomplete coverage is acknowledged. A full chain failure makes the broader analysis partial.

The transcript view can warn that current-session annotations are degraded or unavailable. The cumulative Radar view may still contain valid observations from earlier sessions; it should therefore be read as “through this point in the record”, not proof that every exchange in the selected session was classified.

Sessions with fewer than four transcript segments do not enter extraction. Recap and couple sessions do not use this individual-session feature.

What “lands” cannot mean here

This capability does not establish:

• that a technique caused the following response;
• that an immediate opening produced long-term progress;
• that pushback means the intervention was wrong;
• that the therapist should repeat or stop a technique;
• that every modality-specific intervention was recognised;
• that the response classification is clinically validated ground truth;
• that a one-off count is a reliable patient tendency;
• that a quiet or intellectual response reveals a personality or diagnosis.

Its value is accountability to the actual exchange. It lets a therapist revisit the mix of moves and immediate responses without replacing the therapist’s understanding of timing, formulation, modality and relationship.

Synthetic evidence trace

Therapist turn: “What changed when you paused before answering?”
Immediate client turn: “I noticed I was trying to give the right answer.”
Generated labels: open question → elaboration. The turns are source material; both labels are AI classifications. Adjacency does not establish cause or effectiveness.

Evidence, implementation and review status

• Primary research context: sequential consequences of therapist actions.
• Conversation-analysis context: systematic review of emotion talk.
• Product implementation reviewed against the repository on 25 July 2026; the classifier has not been independently clinically validated.
• Current rendered limitation: persistent AI provenance and a correction path are incomplete, and directional styling can overstate a descriptive response band.

Frequently asked questions

Does the next turn prove an intervention worked? No. It records immediate sequence only.
Are the labels therapist-authored? No. They are AI classifications that require review.
Can it recommend a technique? No.

Review the other science methods or ask about the current workflow.