Study protocol

Participants assigned to an AI-personalized implementation intention will show a higher cue-contingent constructive voice rate than participants assigned to a generic implementation intention, when both arms receive matched cue opportunities, endorsement, interaction time, and measurement.

Primary contrast

C versus B

Secondary mechanism contrast

D versus C

Primary dependent variable

Cue-Contingent Constructive Voice Rate (CCVR): qualifying cue opportunities followed by an appropriate constructive contribution inside the protocol response window, divided by all qualifying cue opportunities.

Secondary dependent variables

• Cue recognition rate and cue-to-speech latency
• Plan execution rate and unique information contributed
• Clarification questions and unresolved questions resolved
• Repeated-argument cycles and dissent before decision closure
• False-positive cue rate
• Participant-rated appropriateness, usefulness, wording fit, and social cost
• Decision revision after new information, agenda items resolved, and decision reached or not reached

Qualifying opportunity

A time-bounded, recorded meeting event that matches an observable condition in the participant's endorsed IF clause, occurs in an allowed meeting phase, and passes cue governance. Silence alone is never an opportunity.

Appropriate constructive contribution

A participant-authored response relevant to the active decision that adds evidence, asks a clarification question, identifies an assumption, or respectfully states a decision-relevant difference. It need not change the decision.

False positive

A delivered cue for which the planned situation was not present, the cue was no longer timely, or speaking would not have added decision-relevant value, as coded independently and reported by the participant.

• Exclude a participant from transcription-derived analyses after transcription consent is withdrawn.
• Exclude research records when research-data consent is absent or withdrawn.
• Exclude technical test sessions, deterministic demonstrations, and duplicate reconnect events from study estimates.
• Exclude opportunities generated while a participant is speaking, within the post-speech grace period, after cues are disabled, or outside the endorsed meeting phase.
• Retain zero-opportunity sessions in reporting; do not treat them as failed participation.
• Pre-specify any further attrition, protocol-deviation, or missing-data rules before confirmatory analysis.

1. The engine records the transcript segment, structural signal, matched endorsed plan, cue time, and participant action.
2. Constructiveness is coded from a written rubric by a deterministic rule or a blinded human coder independent of the model that detected the opportunity.
3. A second blinded coder assesses a pre-specified sample; disagreements are resolved without access to experimental arm.
4. Participant reports of recognition, timing, usefulness, social cost, and false positives remain separate from behavioral coding.
5. The denominator and response window are fixed before confirmatory analysis; sensitivity analyses are labeled as such.

Transcription consent is explicit and required before capture. Translation and research-data consent are separate and optional. Raw audio is off by default and is not stored. Withdrawal stops the relevant capture or logging immediately. Private plans, barriers, cues, bridge outputs, and debriefs are not visible to other participants or the moderator.

Not preregistered. This is a protocol draft for prototype validation. No confirmatory data collection should begin until the hypothesis, sampling plan, response window, coding rubric, exclusions, power analysis, and analysis plan are time-stamped in an appropriate public registry. No outcome claim is made on this site.