You have probably been on both sides of this. Someone says "can I give you some feedback?" and your shoulders rise before they have said a word. Or you are the one with the difficult thing to say, and you spend three days rehearsing a sentence that still lands wrong.
Defensiveness is not a character flaw. It is what happens when a person hears a verdict about who they are rather than a description of something that occurred. The job of feedback is to keep the second thing and remove the first.
Say what happened, not what it means
"You do not care about quality" is a conclusion. "The report went out with the old numbers in section two" is an event. Only one of those can be discussed, and it is not the conclusion.
This sounds obvious and is remarkably hard to do in the moment, because the conclusion is usually what annoyed you. The discipline is to describe the incident precisely enough that the other person recognizes it, without adding the interpretation that makes them defend themselves.
If you cannot name a specific incident, you may not be giving feedback. You may be sharing a feeling about a pattern you have not yet pinned down. That is worth saying honestly: "I have a vague sense something is off and I want to check it with you" is a fair opening. It is not a verdict, and it invites the person in rather than pushing them back.
Keep the effect on you, not on their character
The second sentence should say what the event cost. Not "this was careless", but "I had to redo the figures the next morning, which pushed the client call." That sentence is hard to argue with, because it is about consequences rather than intentions.
This is where people with a strong logic preference often go wrong. They build an airtight case for why the behaviour was inefficient, and win the argument while losing the relationship. The other person concedes the point and quietly decides not to work with them again.
People with a strong values preference go wrong in the opposite direction. They lead with how it made them feel, and the other person hears an accusation they cannot verify. The effect is real, but it needs an anchor in something that happened.
Ask before you prescribe
Before offering the fix, ask how they saw it. "How did that look from your side?" is not a rhetorical question, and it is not politeness. It sometimes produces information that changes your read: a deadline moved, a message never arrived, a decision was made above them.
If their account changes the picture, say so. Changing your mind in public is the single strongest signal that the conversation is real, and it buys you credibility for the parts you are not willing to move on.
Then make one request. Not five. One request that is small, specific, and observable, plus a time to check whether it worked. Five requests read as a list of faults; one request reads as a change.
Timing and setting do most of the work
Never in front of an audience. Never in the last ten minutes of a meeting. Never in writing when the relationship matters, unless writing is the only channel you have.
Private, early, and short. A twenty-minute conversation in the same week as the event is worth more than a thorough review three months later, because memory has already smoothed the details and the other person has already built their own version.
If you are the manager, say why you are raising it. "I am telling you this because I want you on the project in six months" changes how the same sentence is received. People can hear hard things when they understand the intent behind them.
When they get defensive anyway
Some people will hear the first sentence and start explaining, minimising, or counter-attacking. Do not fight the reaction. Name it once, calmly: "I am not trying to catch you out. I want to fix one thing so it does not happen again."
Then stop talking. Defensiveness needs an opponent to push against; if you stop supplying one, it usually runs out of energy in a few seconds. If it does not, end the conversation and pick it up later. There is no version of this where winning the exchange produces the change you wanted.
And if the same conversation has happened three times with no change, stop giving feedback and start managing consequences. Feedback is for people who can hear it; repeated feedback with no consequence just teaches people that the standard is optional.
Conclusion
Good feedback is short, specific, and about something that happened. It says what it cost, asks how it looked from the other side, and makes one request. The rest is timing and setting. If you want to know how you naturally handle disagreement and directness, the communication test covers that in about four minutes. The full test adds how you decide and how much structure you need, and this guide covers the pattern that makes the same clash repeat every week.