How do I prepare a performance conversation?

The trigger

A performance conversation with [EMPLOYEE_NAME] is booked for tomorrow morning, and you have records but no plan and no rehearsed opening. This recipe builds both, because AI can carry the writing but not the facts or the fairness the conversation rests on.

The inputs

• Contemporaneous notes: incidents, dates, what was said
• Logged feedback and KPI or performance data from the record
• The exact wording of the standard not being met, current version dated [POLICY_VERSION_DATE]
• Any prior warning or improvement plan on file
• Meeting logistics: attendees, and whether anything will be recorded

The steps

1. Assemble the facts from the records before you open a model: incidents, dates, the standard, prior feedback.
2. De-identify: replace names with [EMPLOYEE_NAME], [MANAGER_NAME], [EMPLOYEE_ROLE] and [TEAM] before anything is pasted.
3. Run the planning prompt so the model structures a plan from only the facts you supplied, flagging gaps.
4. Open the actual policy and confirm the quoted wording, the same discipline as a policy-grounded HR query.
5. Verify the plan line by line against your records; delete anything that does not trace to a record.
6. Rehearse with the second prompt: brief an archetype, tell it not to go easy, and run your real opening line three ways.
7. Decide the outcome yourself, after [EMPLOYEE_NAME] has responded. The plan never carries the decision.

The prompts

Run the plan first, then the rehearsal in the same chat.

Help me prepare a performance conversation plan. Use only the facts
I supply. Do not add incidents, dates, quotations, ratings or
policy references not in the source material. Flag any gap.

Context (de-identified):
- [EMPLOYEE_NAME], [EMPLOYEE_ROLE], in [TEAM]; manager [MANAGER_NAME]
- Standard not met, quoted exactly from the policy, version dated
  [POLICY_VERSION_DATE]: [paste exact wording]
- Verified incidents from my records, with dates: [paste notes]
- Feedback already given, with dates: [paste logged entries]

Produce a one-page plan: a neutral opening; the examples I will
raise, each tied to one record above; where I invite
[EMPLOYEE_NAME] to respond; and likely questions. Do not propose
an outcome or a rating.

Stop drafting. Be my rehearsal partner. Play the counterpart as an
archetype only: a [EMPLOYEE_ROLE] who believes the concerns are
overstated. Never use or ask for a real person's name, records or
medical information. Do not be agreeable, do not concede quickly,
stay in character until I type STOP. I deliver my real opening
line; react in two to four sentences. Replay the same opening three
ways: defensive, withdrawn, agrees too fast. On STOP, switch to
coach: where I put them on the defensive, where I buried the
message, and the one sentence to open with instead.

The checks

• Every incident, date and quotation traces to a record; delete anything the model added
• Policy wording matches the current instrument, not the model's paraphrase
• No rating or outcome appears in the plan; remove any the model proposed
• The plan includes a genuine point for [EMPLOYEE_NAME] to respond
• Placeholders only in the chat; no identifying detail went in

Governance controls

• The manager owns the conversation and its outcomes. AI structures and rehearses; it never sets a rating or an outcome, the boundary LM-S02 draws.
• Procedural fairness is built in: [EMPLOYEE_NAME] hears the substance of the concern and gets a real opportunity to respond before anything is decided.
• Prompts about a real person use placeholders, in your approved enterprise tool, never a public chatbot.
• If anything will be recorded or an AI note-taker will run, ask for consent on the record first; take manual notes if declined.
• The prep plan is a record about an identifiable person: store it under the same access and retention rules as the review documents, and clear the chat copy.

The output

A one-page, record-grounded conversation plan plus a rehearsed opening. It files into the performance record under your normal rules.

Time it takes

About 30 minutes once practised: 15 for the plan, 15 to rehearse. Allow 45 the first time.