Skip to content
PromptsBuddy
All prompts
ManagementadvancedClaudeChatGPT

Performance feedback that lands instead of stinging

Turns a vague concern about someone's work into specific, dated, behavioural feedback — and checks whether what you're about to say is actually about them or about your own preferences.

The prompt· 0/3 variables filled
~300 tokens
Help me give this feedback well.

MY CONCERN: {{CONCERN}}
SPECIFIC INSTANCES: {{EVIDENCE}}
ABOUT THEM: {{CONTEXT}}

**1. Pressure-test me first.** Before drafting anything, answer:
- Is this a pattern or an incident? The evidence I gave — does it support a pattern?
- Is this about impact on the work, or about a preference of mine? Be blunt.
- What might be causing it that I have not considered, including something I
  am doing?
If the evidence does not support the concern, say so and stop. Do not draft.

**2. The behavioural version.** Restate the concern as observable behaviour and
its effect — no adjectives about the person, no "you always".

**3. The conversation.** What I actually say, in order: the specific opening,
the evidence, the space for their side, the concrete change I am asking for,
and how we will both know it changed. Keep my words plain.

**4. Their likely response**, and how I respond to that without retreating or
escalating.

Rules:
- Do not use a feedback sandwich.
- Do not characterise the person. Behaviour and effect only.
- Do not soften the ask to the point where they could leave unsure what to change.
- Do not invent evidence. Work only from what I gave you.

Fill in the blanks

Run it in:ClaudeChatGPT

What goes in each blank

CONCERN
What's bothering you, in your own unfiltered words.e.g. They over-engineer everything and it's slowing the team down.
EVIDENCE
Specific instances with dates. If you can't list three, that's a finding.e.g. The caching layer in March; the config refactor in May; the abstraction on the export job last week.
CONTEXT
Their level, how long they've been here, what they're working toward.e.g. Mid-level, 18 months in, wants to be promoted to senior this cycle.

Notes from people who use it

  • Step 1 stops perhaps a third of these conversations from happening, which is the right outcome.
  • 'Is this impact or preference?' is uncomfortable and is the question worth paying for.
  • If you can't list three dated instances, you have an impression, not feedback. Wait.

Use this as a skill

A prompt is a skill with the serial numbers filed off. Install it once and your agent reaches for it on its own.

Paste this into ChatGPT, Claude Code, Codex, Cursor, or whatever you use:

Install the performance-feedback-that-lands skill globally from https://promptsbuddy.com/prompts/performance-feedback-that-lands/skill.md
Add to Cursor

.claude/skills/performance-feedback-that-lands/SKILL.md

---
name: performance-feedback-that-lands
description: "Turns a vague concern about someone's work into specific, dated, behavioural feedback — and checks whether what you're about to say is actually about them or about your own preferences. Use when the user asks for help with management tasks like management, feedback, people."
license: CC-BY-4.0
metadata:
  source: https://promptsbuddy.com/prompts/performance-feedback-that-lands
  author: "Dinesh"
  version: "1.0"
---

# Performance feedback that lands instead of stinging

Turns a vague concern about someone's work into specific, dated, behavioural feedback — and checks whether what you're about to say is actually about them or about your own preferences.

## Inputs to collect first

Ask the user for anything below that they have not already given you. Do not
invent values for these.

- `CONCERN` — What's bothering you, in your own unfiltered words. (e.g. They over-engineer everything and it's slowing the team down.)
- `EVIDENCE` — Specific instances with dates. If you can't list three, that's a finding. (e.g. The caching layer in March; the config refactor in May; the abstraction on the export job last week.)
- `CONTEXT` — Their level, how long they've been here, what they're working toward. (e.g. Mid-level, 18 months in, wants to be promoted to senior this cycle.)

## Instructions

Help me give this feedback well.

MY CONCERN: {{CONCERN}}
SPECIFIC INSTANCES: {{EVIDENCE}}
ABOUT THEM: {{CONTEXT}}

**1. Pressure-test me first.** Before drafting anything, answer:
- Is this a pattern or an incident? The evidence I gave — does it support a pattern?
- Is this about impact on the work, or about a preference of mine? Be blunt.
- What might be causing it that I have not considered, including something I
  am doing?
If the evidence does not support the concern, say so and stop. Do not draft.

**2. The behavioural version.** Restate the concern as observable behaviour and
its effect — no adjectives about the person, no "you always".

**3. The conversation.** What I actually say, in order: the specific opening,
the evidence, the space for their side, the concrete change I am asking for,
and how we will both know it changed. Keep my words plain.

**4. Their likely response**, and how I respond to that without retreating or
escalating.

Rules:
- Do not use a feedback sandwich.
- Do not characterise the person. Behaviour and effect only.
- Do not soften the ask to the point where they could leave unsure what to change.
- Do not invent evidence. Work only from what I gave you.

## Notes from people who use this

- Step 1 stops perhaps a third of these conversations from happening, which is the right outcome.
- 'Is this impact or preference?' is uncomfortable and is the question worth paying for.
- If you can't list three dated instances, you have an impression, not feedback. Wait.

---

Performance feedback that lands instead of stinging · by Dinesh · v1.0
From PromptsBuddy — https://promptsbuddy.com/prompts/performance-feedback-that-lands
Licensed CC BY 4.0.

Every prompt is also available at /prompts/performance-feedback-that-lands/skill.md — see all install options.

The pressure test earns its place

Most feedback that lands badly is not badly worded — it is unfounded, or it is taste dressed as standards. Asking the model to check whether the evidence supports a pattern, before it writes anything, catches both.

It will sometimes tell you not to have the conversation. That is the feature.

Behaviour, not character

"You over-engineer" is a trait and invites defence. "The caching layer took three weeks against a one-week estimate, and two of us couldn't review it" is an observation, and it can be discussed. The rewrite in step 2 is mechanical but it changes how the whole conversation goes.

Try these next

SupportintermediateStaff pick

The support reply that de-escalates without grovelling

Drafts a response to an angry customer that acknowledges the real complaint, commits to something specific, and never apologises for things that aren't your fault.

ClaudeChatGPT3 vars
DineshDinesh
Designintermediate

A design critique partner that argues back

Critiques a screen or flow against stated goals, then argues the opposite position — so you get a real second opinion instead of polite agreement.

ClaudeChatGPTGemini3 vars
Lena FischerLena Fischer