rehearse difficult conversation
This identifier was used as an evidence-review target but was never published as a canonical Brali protocol. It is retained as provenance and deliberately does not link to a public route.
Evidence Ledger · Protocol candidate · reviewed 2026-08-31
Harnessing Artificial Intelligence to Facilitate Difficult Conversations
The reviewed source can support a conservative practical protocol.
Brali decision: Protocol candidate. This page summarizes a reviewed evidence boundary; it does not reproduce the source and it does not turn one paper into a universal prescription.
What the source supports
For adults already considering an ordinary difficult conversation, a brief chatbot preparation session modestly increased the probability that they reported actually initiating it over the following week: 65% in the preparation condition versus 59% in control. Brali can therefore test a bounded rehearsal protocol whose goal is to reduce the initiation barrier: clarify the goal, draft an opening, role-play a likely response and prepare one or two contingency moves, then return to the real human conversation.
What it does not establish
- Chatbot rehearsal makes difficult conversations go better once they occur.
- AI resolves interpersonal conflict or improves relationship quality.
- AI preparation is better than preparing with a friend, coach, therapist, manager or journal.
- The 6-12 minute duration is an established optimal dose.
- The result generalizes unchanged outside US and Canadian adults familiar enough with chatbots to complete the intervention.
- The protocol is appropriate for abuse, coercion, threats, harassment, legal disputes, HR investigations or other situations requiring safety planning, professional support or a durable written record.
- A specific chatbot model or prompt is necessary to obtain the effect.
Important limitations
- The primary real-world outcome was participant self-report rather than independently observed conversation behavior.
- The absolute difference in conversation initiation was modest: six percentage points.
- The experimental and control chatbot interactions differed in duration, engagement and prompting, so the study does not isolate which preparation component caused the effect.
- The sample was limited to US and Canadian Prolific participants, restricting cultural generalizability.
- Among participants who did have the difficult conversation, the study found no significant differences in satisfaction, perceived quality, difficulty or thoroughness.
- The study followed behavior for roughly one week and does not establish repeated-use or long-term effects.
- Preparing the conversation increased negative affect immediately after the exercise, and repeated reliance on a chatbot was not studied.
- Safety-sensitive, coercive, legal and high-power-imbalance conversations were not established as appropriate use cases.
Source context
Harnessing Artificial Intelligence to Facilitate Difficult Conversations
- Source type: preregistered-randomized-behavioral-experiment
- Design: Preregistered two-part randomized behavioral study. Participants first identified a real difficult conversation they were considering, then were randomized to spend 6-12 minutes preparing it with a chatbot or to ask the chatbot three neutral factual questions. Of 1,465 participants completing Part 1, 1,371 returned for a follow-up survey beginning five days later. The primary behavioral outcome was whether the participant reported having the difficult conversation in the intervening period.
- Population: Adult Prolific workers in the United States and Canada who could identify a difficult conversation they were considering having. Most target conversations involved romantic partners, family or friends; about one-fifth were professional conversations.
- Exposure / intervention: A short GPT-4o-mini conversation used to prepare a real interpersonal conversation. Participants commonly planned what to say, role-played the other person, sought phrasing or contingency advice, and anticipated possible reactions. Control participants interacted with the chatbot on unrelated factual questions after identifying and describing their difficult conversation.
- Outcomes: Self-reported initiation of the difficult conversation during follow-up; Anticipated conversation quality; Perceived importance, difficulty and anxiety; Post-intervention affect; Among participants who conversed: satisfaction and perceived quality, difficulty and thoroughness
Citation: Maheshka C, Folk D, Dunn E. Social Psychological and Personality Science. First published online May 5, 2026.
DOI: 10.1177/19485506261444068
What this changes in Brali
Editorial note: Protocol direction: use AI as a private rehearsal surface, not as the relationship partner or final authority. Ask the user to name the conversation goal, draft one opening sentence, role-play a plausible response, identify the main failure mode and prepare one recovery phrase. The protocol should stop or redirect when safety, coercion, harassment, legal/HR sensitivity or professional mental-health support is relevant. Success should initially be measured as whether the intended safe conversation was initiated, not whether AI supposedly improved the relationship.
Use the boundary, not just the headline
When an AI agent or a person retrieves this decision, preserve both the supported claim and the unsupported or overstated claims. Dropping the boundary would turn a reviewed source into a stronger claim than Brali actually maintains.
All Evidence Decisions → · Research & trends → · Machine-readable review →