How to design a manager discussion guide with AI

Start with evidence, context, and limits

Define the assessment or work sample, task conditions, available materials, time limit, and evidence captured. Separate direct observations from interpretations. A submitted recommendation, a cited source, or a revision event can be observed. Motivation, general ability, and future workplace performance cannot be inferred from one bounded task.

Use a conversation evidence table:

| Element | Example | How to discuss it | |---|---|---| | Observation | Two external claims lacked traceable sources | Ask how sources were selected and checked | | Strength | Recommendation stated audience and decision | Explore how the participant framed the brief | | Missing opportunity | Research log was unavailable | Do not treat absence as poor evidence navigation | | Context | Final ten minutes used for format repair | Ask what changed and how time was prioritized | | Development hypothesis | Source mapping may need practice | Confirm with another task before prescribing support |

All examples are synthetic. The language stays close to the work rather than assigning a fixed label to the participant.

Structure the conversation around questions

Open by explaining the purpose: understand the work, recognize effective choices, explore gaps, and agree on a practical next step. Invite the participant’s account before presenting an interpretation.

The US Office of Personnel Management’s performance-management materials emphasize specific feedback connected to situations, behaviors, and impact, along with development discussion. Its performance-management toolkit also frames performance evidence and feedback as an ongoing process. These sources support specificity and dialogue in their employment context; they do not validate Two Prune scores or determine an employment outcome.

A useful sequence is:

The guide should contain optional probes, not a script that overrides listening.

• What was the participant trying to achieve?
• Which part of the task felt most uncertain?
• What evidence shaped the recommendation?
• Which AI output required the most checking?
• What would the participant change with another attempt?
• Which support or practice would make that change possible?

Translate evidence into development options

For each observed pattern, offer more than one explanation and more than one possible support. If citations were weak, possible causes include unclear standards, unavailable sources, time allocation, tool friction, or a verification skill gap. Ask questions and seek another sample before selecting the intervention.

A weak guide says, “The score shows poor judgment; assign remedial training.” That converts a bounded result into a broad verdict and assumes the cause. A stronger guide says, “In this task, two recommendations did not identify their evidence basis. What information did you rely on, and what made it difficult to preserve the links?” It creates room for context and diagnosis.

Possible development actions include a worked example, targeted practice, a checklist, manager modeling, access repair, peer review, or a second assessment opportunity. Name the behavior to practice and the evidence that would show progress.

Close with ownership and follow-up

Agree on one or two actions, the support owner, a realistic date, and the next evidence opportunity. Record the participant’s perspective and any disagreement. Do not rewrite uncertainty as consensus.

Before use, verify:

The artifact is the manager guide plus evidence table, question set, agreed action record, support owner, and follow-up plan. Protect participant information according to the organization’s access and retention rules.

• every statement maps to observed task evidence;
• missing records are not scored as weak performance;
• strengths and gaps are both specific;
• contextual factors are visible;
• questions precede causal conclusions;
• actions target observable behavior;
• no score is presented as a personal verdict; and
• follow-up specifies a new evidence opportunity.

Sources and claim map

The evidence, discussion prompts, and development options are synthetic. This resource does not support ranking, selection, diagnosis, or prediction about any person.

• US Office of Personnel Management, Measuring employee performance quick guide: supports specific feedback and development discussion in its federal workforce context.
• US Office of Personnel Management, Performance Management Toolkit: supports ongoing performance planning, monitoring, development, rating, and recognition practices in its scope.

Sources: [1] [2]