Prepare for the next therapy session with source-linked open threads

See how Intuita turns eligible transcript-grounded loose ends into a short therapist-reviewed preparation queue—not a generated treatment plan.

Start with the thread already in hand

After an analyzed individual session, Intuita can prepare one short headline and a prioritized set of threads that may be worth carrying into the next conversation. A thread is not an instruction. It is an observable loose end: something that was explicitly promised, left open, connected to a goal, raised as friction, or stated in safety-related language.

This distinction matters. Guidance does not decide what the next session should be about, plan treatment, assess risk, or tell a therapist how to intervene. It reduces the work of finding the relevant moment again. The therapist reads the thread, checks its evidence and decides whether it belongs in the room.

Guidance is generated only for an individual session that completed the relevant analysis. Couple sessions follow a separate analysis path. A session without a usable transcript, an interrupted analysis, or a result in which no thread passes the gates does not produce a normal guidance queue.

Six kinds of observable follow-up thread

Intuita keeps different kinds of loose ends separate:

1. Explicit safety language. A quoted statement about suicidality, self-harm, abuse or violence. This is not a risk score and the engine may not infer risk from mood, topic or tone alone.
2. Commitment. Something the therapist or patient explicitly said they would do before the next session.
3. Left open or deferred. A subject that was raised and explicitly postponed before it was resolved. A topic merely fading from the conversation is not enough.
4. Possible friction. An observed moment of disagreement, distance or disconnect in the therapeutic relationship, surfaced without explaining its meaning or prescribing a repair.
5. Between-session work. A thread about a task or agreement intended to continue outside the session.
6. Connection to a goal. Current-session material that refers back to a treatment goal set by the therapist.

They are follow-up categories, not six recommendations. A queue can contain fewer than six, more than one of a category, or nothing at all.

What the engine reads

The main source is the complete, speaker-labelled transcript. Each turn receives a stable segment number so verified items can link back to the exchange. Intuita also receives a deliberately bounded continuity view:

• up to five recent guidance cues and their status;
• the therapist’s most recent written plan for the following session, capped to keep it secondary;
• the current goals and their descriptions;
• a limited set of current topic summaries and detected gaps;
• any existing guidance for this same session, so therapist edits can be preserved when the same thread returns.

It does not simply pour every analysis result into one prompt. The current guidance stage does not directly consume the CCRT, collaboration, engagement or full trajectory outputs. Topics and gaps are supporting context and are deliberately lower in authority than what was actually said in the session.

From session to a short queue

The pipeline has several separate responsibilities:

1. Project the transcript. Patient and therapist turns are labelled, sanitised and indexed.
2. Draft structured candidates. The AI may propose a headline, category, cue, short context and evidence reference. It cannot assign the final transcript pointer, speaker role, status or stable identity.
3. Reject directive wording. A deterministic English-and-Dutch screen removes cues that start like instructions. “Explore…”, “ask…” or “consider…” belongs too close to treatment direction.
4. Check the category and evidence. Transcript-backed categories need a match in the session. Goal and between-session threads need a relevant reference.
5. Put explicit safety first. Qualifying safety threads sort ahead of other items. Non-safety output is kept short; safety items are not discarded merely to meet that normal cap.
6. Merge therapist state. When the same evidence-backed thread returns, its stable identity allows eligible edits and review state to survive regeneration.
7. Write an explicit result. A verified empty result is different from a failed or unavailable analysis.

The result is intentionally a queue, not a generated agenda.

Safety without pretending to assess risk

Safety handling is deliberately narrow. The guidance stage may surface explicit language; it does not estimate intent, severity or probability and does not perform a clinical risk assessment.

If an explicit safety candidate cannot be matched to the transcript, it is withheld instead of presenting an unverifiable quote. A warning is raised inside the system. When a safety thread does qualify:

• it is placed first;
• it cannot be dismissed like an ordinary loose end;
• reviewing it requires an explicit confirmation;
• a previously dismissed non-safety thread is reactivated if the same evidence is freshly classified as safety.

Those controls keep the item visible. They do not replace a therapist’s safeguarding procedure, professional judgment or direct conversation with the patient.

The therapist owns the final reading

The clinician can rewrite the headline or cue, mark an item reviewed and set aside a non-safety item. A reviewed item remains visible; a dismissed ordinary item is hidden. Save failures remain visible as failures.

Same-session continuity is deterministic when the verified thread returns with the same stable identity. Cross-session continuity is softer: prior cues and statuses inform the next generation, but new evidence can legitimately reopen a subject. An explicit safety statement is deliberately allowed to resurface.

Guidance is shown as AI-generated material that must be reviewed before it is relied on. The editable state is not cosmetic. It is the boundary between a machine-found loose end and the therapist’s agenda.

What it does not claim

Next-session guidance does not claim that:

• every session will produce a queue;
• every important subject was detected;
• a surfaced thread should be discussed next;
• an unsurfaced subject is unimportant or absent;
• homework and goal references have the same quote verification as transcript-backed items;
• a moment of friction is an alliance rupture;
• explicit safety language constitutes a risk assessment;
• dismissed subjects can never return;
• generated text is clinically correct because it passed a structural gate.

The system is deliberately better at saying “this did not qualify” than inventing material to fill the page.

Synthetic evidence trace

Source: “I will bring the letter next week.”
Generated thread: “Client stated an intention to bring the letter.”
Provenance: the source excerpt is transcript-linked; the category and summary are AI-generated. The therapist decides whether the thread is relevant.

Research context

Read the research on how therapists prepare for sessions.

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