AI Tools for Work

Give employee feedback without labels: a Claude prompt

Updated: 2026-09-26

Feedback lands when the person can recognize the situation you describe and cannot argue with the facts. Claude is good at stripping notes down to that form: situation, behavior, impact, expectation, with no labels and no "you always". A draft takes about 15 minutes. The most useful part of the prompt is the last line, which asks Claude to point out where your own notes judge the person instead of describing what happened.

Notes first, and specific

Write down two or three situations you saw yourself, each with a date. What did the person do, and what did it affect: a missed deadline for a client, a colleague redoing work, a meeting that ran without the numbers it needed. Then write what you expect from now on and what you will do to help. If you cannot name a concrete situation, the feedback is not ready yet; "I feel you are disengaged" gives the other person nothing to change.

In the notes, the person is [Employee 1]. Notes about an employee with their name are personal data, and the feedback works just as well without it.

The prompt

Role: you are an experienced manager and coach who gives feedback without labels. Context: employee [Employee 1] in the role of {{position}}. Situations: {{situation 1: what they did, when, what impact}}, {{situation 2}}. What I expect going forward: {{expectation}}. How I am ready to help: {{help}}. Task: frame the feedback as situation, behavior, impact, expectation for each situation. No labels, no assumptions about motives, no "always" or "never" generalizations. Add five open questions for the conversation and a plan to review progress in a month. Format: text for an in-person conversation up to 250 words, no long dashes, then the questions as a list. At the end, separately list the places in my notes where a judgment of personality replaces a fact, if there are any.

An example of what changes

A manager's note might read: "Irresponsible with the weekly report, clearly does not care about the team." Claude will typically flag both halves. "Irresponsible" is a label. "Does not care about the team" is a guess about motive. What remains is the fact: the weekly report arrived after the team meeting it was meant for, twice this month, and the team made decisions without the numbers. That version is harder to dispute and easier to act on. It also opens a real question: what is getting in the way of sending it on time?

Reread before the conversation

Go through the draft sentence by sentence and ask: did I see this myself? The model occasionally adds a plausible cause ("perhaps due to workload") that sounds kind and is still an assumption. Remove it and let the person explain in the conversation. The five open questions are there for exactly that.

What to do

  1. Write two or three situations with dates: what the person did and what it affected.
  2. State the behavior you expect and how you will help.
  3. Paste the prompt into a new Claude chat with the employee as [Employee 1].
  4. Read the list of places where your notes judged personality, and fix your own wording.
  5. Remove any statement about motives that you did not observe.
  6. Hold the conversation in person, using the text as your plan, and schedule the review in a month.

How to tell it went well

Every statement was tied to a specific situation, no labels like "irresponsible" appeared, there was an expectation and an offer of help, and the conversation happened in person. Sending the text as a message defeats the purpose, since the person cannot ask questions or explain.

Plans and privacy

Feedback drafts fit easily within the free plan. Pro at 20 USD per month (17 with annual billing) adds Projects, useful for managers who keep a running record of one-to-ones in a single place; Team at 25 USD per seat with annual billing keeps team data out of training by default. See Claude pricing. Check Settings, Privacy on personal plans before pasting notes. HR managers will find the saved template on the employee feedback page, and Claude vs ChatGPT covers the difference in writing style.

FAQ

How do I give constructive feedback to an employee with AI?

Write two or three specific situations with dates, what the person did and what it affected, plus your expectation and the help you offer. Claude frames them as situation, behavior, impact, expectation, without labels or guesses about motives.

Why does AI-written feedback sound judgmental?

The model adds assumptions about why the person acted as they did. Remove every statement about motives you cannot know and keep only facts and their impact.

Can I use the text from Claude as a written review?

Use it as the plan for an in-person conversation. The text is a structure for what you say, and the conversation is where the person can respond.

Sources

Useful pages

AI tools weekly for your role

One email a week: new tools, price changes and one tested prompt for your job. Free, unsubscribe in one click.

Need an AI agent for your task?

An agent built around your workflow: we pick the models and tools and connect them to your systems.

Describe your task