Draft difficult feedback using situation-behaviour-impact
Turn a hard observation into feedback structured as situation, behaviour, impact and a request — so it reads as specific and fair, not as a verdict.
Last updated
Fill in the blanks
Assembled in your browser — nothing you type is stored or sent anywhere.
Your prompt
Model-agnostic — works in Claude, ChatGPT, Gemini or any AI chat.
Why this prompt works
- →Naming the situation-behaviour-impact structure explicitly, rather than just saying "give feedback," is what keeps the output anchored to one incident instead of drifting into a general character assessment.
- →Banning trait labels like "unprofessional" and requiring observable behaviour instead is the single change that most reliably keeps difficult feedback fair and defensible rather than accusatory.
- →Ending on a specific request rather than a verdict reframes the message as forward-looking, and the instruction to flag thin input rather than generalize stops the model from inventing a pattern from a single data point.
Part by part
- situation-behaviour-impact structure
- A named framework rather than a request to be constructive. Each of the three parts is separately checkable, which is what makes the feedback arguable rather than assertive.
- what was actually said or done, not a character judgment
- The distinction the whole structure rests on. Behaviour can be disputed on facts; a trait can only be denied, which turns a conversation into a defence.
- stated plainly, not exaggerated
- Overstating impact is the fastest way to lose the point, because the recipient argues with the exaggeration instead of the behaviour.
- a specific request for what to do differently
- Feedback without a request is a complaint. The request is what makes the next conversation possible — either it happened or it did not.
- tell me what's missing instead of writing something generic
- Guards against the worst outcome: a plausible-sounding script describing a situation that did not happen the way it is written.
Common mistakes
Help me tell someone they're being unprofessional.Unprofessional is a label, not a behaviour, so the draft argues about the label. Naming what was said or done in a specific meeting gives the person something they can actually change.
Write me a script for a difficult conversation.A script produces delivery rather than a conversation, and it falls apart at the first unexpected response. The structure is a way to organise what you already know, not a performance to read out.
Soften this feedback so it doesn't upset them.Softening past a certain point removes the message. The most common failure in difficult feedback is that the person leaves not knowing anything needs to change.
Questions
Why avoid words like lazy or unprofessional?
Because they describe a person rather than an action, so there is nothing specific to respond to. The conversation becomes about whether the label is fair, which is unwinnable for both people and changes nothing.
Should I really write this out beforehand?
Writing it out is how you find out whether you have a specific example. If you cannot fill the situation and behaviour sections, the feedback is not ready — and delivering it anyway is how it lands as a general impression.
What if the impact is hard to state without exaggerating?
Then state it small. "I had to redo the section before the deadline" is more persuasive than "it put the whole project at risk", because it is checkable and nobody can dispute it.
Is a model the right help for this?
For structuring an example you already have, yes. For deciding whether feedback is warranted, no — that judgement depends on context and history the model has no access to, and it will support whatever framing you supply.