Purpose and fit

Use this when fluent AI output, urgency or hierarchy makes it difficult to raise a concern. Rehearse a short, fictional discussion before the next consequential use. Focus on a visible behaviour: a colleague raises an evidence-based question, the recipient responds constructively and an authorised person resolves the next step. Use fictional cases; avoid investigating conduct or eliciting sensitive disclosures.

Before you start

Describe the team task, permitted decision roles and existing escalation routes; no new dataset or personal account is required. You may use an existing authorised, anonymised example or request a clearly fictional illustration. Avoid sensitive disclosures and retain unknown escalation details as placeholders for an owner to confirm.

Expect a fictional rehearsal guide and blank concern record, not an investigation, a resolved concern or proof of psychological safety. The guide should distinguish observed evidence from possible consequences and leave actual decisions to authorised people.

Required inputs

  • A fictional or sanitised example of AI-assisted work with one plausible unresolved issue.
  • Existing routes for specialist questions, urgent incidents and workplace concerns.
  • A facilitator and participants raising and receiving the concern; allow written participation.
  • An agreed note-taking approach and clear limits on who can see what is recorded.

Practical workflow

  1. Explain the exercise and its boundaries. Let participants use the fictional example rather than personal cases. State that no one must disclose an experience, and identify the appropriate existing route if a real urgent issue emerges.
  2. Give participants the same evidence. Ask them to identify one observable concern, such as an unsupported claim or unexplained mismatch, without speculating about a colleague's motives or an AI system's intentions.
  3. Practise raising it with a concrete request: “I can trace this claim to neither source. Before we use it, can we check the source or hold that part?” Distinguish an observation from its possible consequence.
  4. Practise the receiving response: acknowledge the specific issue, restate it accurately, ask what evidence is missing and identify who can decide. Avoid demanding that the person raising the question prove harm before the question is considered.
  5. Rehearse disagreement and time pressure. The authorised owner chooses a bounded next step, such as additional checking or holding the affected output. Record unresolved issues and the escalation route; silence or urgency is not resolution.
  6. Debrief the interaction. Ask what made the concern easier or harder to express and one change the leader will make. Revisit that change at a later routine meeting without publishing individual practice scores.

Expected output

A short conversation guide and concern record. Suggested opening: “I noticed [specific observation]. It may affect [work consequence]. The evidence I have is [source or gap]. Before [next action], could we [check or pause request]?”

Suggested response: “I understand the concern as [restatement]. We know [evidence] and still need [unknown]. [Authorised role] will decide [bounded next step] by [time]. Until then, the affected work will [holding arrangement].”

Concern record: Task/output: [ ]. Observation and source: [ ]. Possible consequence: [ ]. Evidence needed: [ ]. Current holding arrangement: [ ]. Resolution owner: [ ]. Next check/time: [ ]. Outcome and rationale: [ ]. Route if unresolved: [ ]. Access to record: [ ].

Human checks

The facilitator checks that the response addresses the concern rather than judging the speaker’s confidence or tone. The leader must follow through on the promised next step. Do not promise anonymity or protection that the available process cannot provide. Handle personal allegations through the appropriate established human process, with qualified support where needed.

Limitations

A rehearsal cannot establish psychological safety or guarantee how people will respond under pressure. Repeated concerns may reveal workload, authority or process problems that communication practice alone cannot solve. Do not use the exercise to label individuals as resistant or to automate employment decisions.

Starter prompt

Create a fictional challenge rehearsal for [team task], using these boundaries and escalation routes: [details]. Use an unsupported AI-generated claim as the issue; include no real employee information.

Provide a scenario and two practice rounds, the second with time pressure. Structure the concern as observation, possible consequence, evidence gap and requested check. Structure the response as accurate restatement, known evidence, unknowns, authorised decision owner, next step and holding arrangement.

Create a blank concern record with fields: Task/output; observation/source; possible consequence; evidence needed; holding arrangement; resolution owner; next check/time; outcome/rationale; unresolved-concern route; record access. Add debrief questions about the interaction and process. Leave decisions to the authorised human. Do not infer motives, promise confidentiality or treat confidence, silence or seniority as evidence.

Adapt the requested fields to the authorised inputs actually supplied. Use only supplied escalation routes; if none is provided, mark it for confirmation rather than inventing a policy or confidentiality promise. Keep fictional events visibly fictional and retain disagreements as unresolved instead of inventing consent or motives. State the expected output and acceptance criteria before drafting. Ask for essential missing information; where a useful partial draft is possible, label its limits and assumptions. Do not invent facts, evidence, results or approvals.

Data, model variability and your responsibility

Use relevant data that exists or that you provide, with accuracy and completeness sufficient for the task. Different AI models and versions may produce different results, and responses may vary between runs. You are responsible for checking and validating the output against your own requirements, source information and acceptance criteria before relying on it.