How to get honest feedback

You've asked how you're doing and been told you're doing great. You've come out of a performance review with nothing you could actually act on. Then a promotion goes sideways, or someone quietly moves a project off your plate. You find out that several people had the same opinion about you, and not one of them told you.

This article covers why honest feedback rarely reaches you, and how to get more of it out of an AI. It also goes into where AI gets it wrong and what you still need people for. Read the first half if you want to understand why it's happening, and skip to the prompting section if you have something to rehearse tonight.

The sources are people who have sat on both sides of this: Kim Scott, who led teams at Google and Apple and wrote Radical Candor; Jules Walter, a product leader at YouTube and Slack; Deb Liu, CEO of Ancestry and previously a VP at Facebook; Alisa Cohn, an executive coach of more than 20 years; Mellody Hobson, co-CEO of Ariel Investments; Carla Harris, vice chairman at Morgan Stanley; plus Hilary Gridley of Whoop, investor Tomasz Tunguz, and Zapier's Reid Robinson on what AI does and doesn't do well here.

Nobody told you because it would have cost them

If something were really holding you back, wouldn't someone have told you by now? Usually several people can already see it. They talk about it when you're not there, and saying it to your face is awkward enough that nobody quite gets around to it.

Jules Walter spent years at YouTube and Slack watching careers stall for reasons the person never heard.

The gap between what people say about you and what they say to you is the whole problem. Deb Liu watched it happen to someone she managed and rated highly.

Kim Scott tells a version of this that tends to stick with people. She had an employee, Bob, whose work wasn't good enough, and she said nothing for almost a year because she liked him and didn't want to upset him. When she finally had to let him go, he said something she's been repeating ever since.

Nobody in that story was being malicious. Scott stayed quiet precisely because she cared about Bob, which is what makes this so hard to fix.

What it costs the person who tells you

Giving you real feedback is genuinely hard work for the other person. They have to plan it, pick their words, watch your face while you hear it, and then deal with whatever happens next. Jules Walter describes that trade-off plainly.

Mellody Hobson understood this from the receiving end early on. Senator Bill Bradley once told her she could be a ball hog, and while she sat there she was working out exactly what her own reaction would cost her.

She's right, and it compounds. React badly once and you can pick up a reputation as someone who can't take feedback, and that reputation travels ahead of you. People start deciding in advance which subjects are safe to raise with you, so you don't only lose feedback from the person who was there that day. You lose whole categories of it, and nobody tells you which ones got moved off the list.

You can also tell how hard this is by how much coaching exists to help people do it. Alisa Cohn coaches executives through it, and she has them dig up specific evidence before they open their mouths, because if they skip that preparation they usually avoid the conversation entirely.

Cohn also keeps a rehearsed script for the moment someone gets defensive. Matt Mochary developed his curriculum with startup CEOs and publishes it free. It goes further and builds a whole ritual around this: a feedback protocol, a five-step method for receiving it, and a prompt you're supposed to say out loud to get anything real: please share the thought you think will be most painful for me to hear. You only write that down as a procedure because people default to silence. If you're on the other side of this and have to deliver the hard message yourself, the giving feedback playbook covers the delivery.

Why people soften what they tell you

Even when someone does tell you something, they're telling you a version shaped by their relationship to you. Each of these people softens what they tell you for a different reason, and each one needs a different workaround.

Your manager matters most here, because that's the relationship most people are asking about. Alisa Cohn describes what actually happens inside a manager who has decided that keeping you happy is part of the job.

"They don't really know there's a problem" is the part to notice. From your side, a manager doing this looks exactly like a manager who thinks you're fine.

Your manager is only one source. Everyone else who could tell you has their own reason to soften it, and knowing which reason you're dealing with tells you how much to trust what you hear from them.

None of these people are behaving badly. Each one is responding sensibly to their own position. The result is that you hear plenty of opinions and almost none of them tell you where you actually stand.

The interviewer

The purest version of this is the person who just interviewed you. They have the most rigorous read on you that anyone has taken in years: several people, a defined bar, notes written down within the hour. Then you get a template email. Once they've decided to pass on you, there's nothing in it for them. They're under no obligation to tell you anything, writing it up takes time they don't have, and saying anything specific only creates risk they don't need.

You can sometimes get a little of it by asking before the interview ends rather than after the rejection, which our playbook on asking for feedback after an interview covers. Mostly, though, this is the gap you have to fill yourself, and it's the clearest case for rehearsing with something that will tell you what it sees.

What an AI can do here that a person can't

AI is genuinely useful here, and not because it's smarter than the people around you.

It's available at 11pm the night before

The moment you most need a read on your promotion doc or your talk track is usually the night before, and that's precisely when you can't ask anyone. You can ask a mentor once a quarter, and your manager in the 1:1 slot. An AI has no schedule, so you get the feedback while you can still change the work.

It can take a perspective nobody in your network has

People underrate this one. Your question is usually very specific: how a staff engineer promo packet reads at a company of this size, what a Series B head of sales actually gets judged on, how a skeptical VP of finance will react to this framing. The person who has done exactly that, at your level, in your industry, is probably not someone you know.

Even a strong network doesn't solve this. A great network gets you 20 or 30 people who will take your call, and they cluster around the companies you've worked at and the level you're currently near. You can't assemble a panel with the right expert for every question you'll have over a career. An AI has read an enormous amount of what those people have written and said, and it will answer as one of them if you ask it to. It won't match the real expert, and for most of your questions the real expert was never going to be free on a Tuesday night anyway.

You can do it twenty times

Getting better at any kind of conversation takes reps, and people will only give you so many. Nobody is going to sit through your difficult conversation with your director eight times while you try different openings. An AI will, and the eighth attempt is usually much better than the first. If the thing you're rehearsing is a promotion ask, the promotion playbook gives you the structure to practice against.

You can give it the whole context

Explaining your full situation to a person is slow and socially awkward. You can't spend fifteen minutes of a mentor's half hour on the history of your relationship with your skip-level, so you compress it, and they advise on the compressed version. With an AI you can paste the whole thing: the doc, the org chart as you understand it, what your manager said in March, what you're worried about. The advice gets better in direct proportion to how much you tell it, and there's no social cost to telling it everything.

Reid Robinson at Zapier did a version of this for his wife's job search, feeding an AI the careers page, the job description and the competitive landscape before each interview.

The AI job search playbook goes deeper on that kind of preparation if you're interviewing right now.

Hearing it is much easier

Hard feedback is difficult to take mostly because of the person delivering it. You're managing your face, worrying what they now think of you, and deciding whether to defend yourself, all while trying to absorb the content. None of that applies to a machine. There's nobody to disappoint, no relationship to repair afterwards, and you can dismiss the whole thing with no consequences. That last part sounds like a weakness and it's actually why it works. Because walking away costs you nothing, you'll ask for the harsh version far more often than you ever would from a person, and you end up hearing a lot more.

If you're going into a review where you expect to hear something hard, the promotion denial playbook has the questions to ask in the moment.

Where AI feedback goes wrong

An off-the-shelf assistant will tell you your work is good. It will do this almost regardless of whether your work is good. If you take that at face value, you'll walk into the meeting confidently wrong, which is worse than walking in unsure.

Hilary Gridley, Head of Core Product at Whoop, has run into this enough that she changed how she asks.

Make that swap your default. Asking "is this good?" invites agreement, and you get agreement. Asking it to restate your argument back to you produces something you can check against what you meant. If the restatement is fuzzy in the same place your draft is fuzzy, you've found your problem without anyone having to be critical of you.

AI also smooths things out instead of pushing back on them. Tomasz Tunguz noticed it in his own writing.

It does the same thing to your thinking. Give it a mediocre plan and it will hand you back a tidier mediocre plan. It won't tell you the plan is the problem unless you make it.

How to ask so you get something useful

Every fix here comes down to taking away the option to agree with you. Never ask a yes or no question about quality, because "yes" is always available and always cheap.

The same question, asked two ways

Say you're up for a promotion and you've written the self-assessment. Here's how the two versions play out.

The second prompt works because it gives the model a job it can't do by agreeing with you.

How to get more out of the humans too

None of the above replaces people, because an AI only knows what you've told it. It doesn't know who actually decides your promotion, what your skip-level says about you when you're not there, or that the reorg in six weeks makes your project irrelevant. If your read on the situation is wrong, everything it tells you is built on that wrong read and it will sound just as sure of itself.

The bigger gap is that it can't do anything on your behalf. It can't put your name forward in the meeting you're not in, introduce you to the hiring manager, or spend its own credibility backing you. Carla Harris spent decades at Morgan Stanley on the deciding side of those meetings.

So you still need people, and your job is to make it cheaper for them to tell you things. Vague requests do the opposite, because "any feedback for me?" makes the other person choose the topic, the severity and the wording, all on the spot.

Then make sure they don't regret having told you. Mochary's five A's cover this: acknowledge what you heard by saying it back, say thank you without arguing, say plainly whether you're acting on it, and then tell the person later what you did. Whether someone gives you real feedback a second time depends almost entirely on how the first time went.

How to take it so people keep telling you

Asking well gets you one round. What you do in the ninety seconds after someone answers decides whether you ever get a second one. Mochary's five A's give you a method for that, and they're also how you avoid getting branded as someone who can't take feedback.

One tactic from that doc is worth stealing on its own. When someone is clearly holding something back, Mochary borrows a move from Lachy Groom and asks them not to say it yet: what's the thought you're afraid to tell me because you think it'll hurt my feelings? Don't say it, just think it. Do you have it? Now say it. Naming the fear first makes the sentence much easier to get out.

Sorting the signal from the noise

More feedback means more contradictions, and some of the advice will cancel out. Here are a few rules that hold up.

If the feedback you keep getting is about how you show up rather than what you produce, the meetings playbook covers that ground specifically.

Frequently asked questions

Why doesn't my manager give me feedback?

Usually a mix of three things. They're worried the honest version will demoralize you or provoke a reaction they'll spend the next month managing. They've been trained to save it for the review cycle, which means you hear it once a year and it feels like an ambush. And they may believe they've already told you, in language so softened that you heard it as a suggestion. Ask them directly what would move your rating, and ask what you'd need to stop doing.

Is it a bad sign that I get no critical feedback?

It isn't proof of anything on its own, but silence tells you nothing about how you're doing. Kim Scott's Bob had no idea, and everyone around him did. If you've had no critical feedback in six months, treat that as a gap in your information rather than a good report.

Can I really use AI as a career coach?

For rehearsal, structure, perspective and reps, yes, and it beats the alternative, which is practicing on the real meeting. For anything that requires knowing your organization, no. The thing to watch is that a general-purpose assistant defaults to agreeing with you, so if you ask it whether your plan is good you'll hear yes and learn nothing.

How do I stop AI from just agreeing with me?

Change the shape of the question. Ask it to restate your argument rather than rate it, ask for the weakest part rather than whether it's strong, ask it to argue against you, and give it a role that has a reason to object. Also tell it explicitly not to compliment the work and to skip the summary of what you did well.

Isn't an AI coaching company always going to say AI is the answer?

Fair question, so here's our actual position. A general-purpose assistant is a reasonable first step and it will flatter you, which is a real problem when the whole point is to hear something you don't want to hear. A tool built for this specific job does better. It can hold its own view of what good looks like instead of adopting yours, and you can set how direct you want it to be. Neither one knows your organization's politics or can advocate for you. Both of those stay human.

What if I ask for feedback and get nothing useful?

Narrow the question and shorten the time gap. "How am I doing?" is nearly unanswerable. "In yesterday's review, what would you have changed about how I opened?" gets a real answer, because it's about one moment and the person doesn't have to render a verdict on you.

The checklist