Beyond the Feedback Sandwich: Evidence-Based Ways to Give Feedback

I've worked for a dozen bosses across twenty-plus years, and one of them ran every one-on-one off the same template: a compliment to open, "one thing to work on" in the middle, a compliment on my way out the door. It took about two months for the conditioning to set in. Praise from him stopped registering as praise. It registered as a countdown.

That's the feedback sandwich, and it fails for reasons the research explains well: wrapping criticism in praise turns the praise into a threat cue, buries the one message that needed to land, and teaches people to discount your appreciation even when you mean it. The evidence-backed alternative isn't harsher – it's more honest. Name the shared goal. Describe specific behavior and its impact, not character. State your care in plain words instead of padding. Close with who will do what by when. The GAIN framework packages those moves into something you can remember mid-conversation, and the rest of this article walks through it, with real feedback rewritten line by line.

Some context on where this comes from: when Adam White and I built Conflict Campaign, we synthesized more than 200 studies on conflict and communication skills, and feedback delivery kept surfacing as the skill with the widest gap between how often it's needed and how well it goes. Adam's the licensed professional counselor of the two of us; I'm a product manager who's been on both ends of a lot of bad feedback. Feedback is also one of the five trainable skills in our map of the 15 types of conflict on product teams. This piece covers the delivery half: what to actually say.

Why does the feedback sandwich fail?

Three mechanisms, and they compound.

Praise becomes a threat cue. Neuroscience research on feedback is blunt about the starting condition: the brain treats criticism as social risk – a threat to your standing in the group – and a threat state narrows thinking exactly when you need it open. The sandwich doesn't calm that system. It trains it. Once someone learns that your compliments precede corrections, every compliment starts the brace. You haven't softened the criticism. You've booby-trapped the praise.

The criticism gets lost. Kluger and DeNisi's 1996 meta-analysis of feedback research found that feedback improved performance on average – and that more than one-third of feedback interventions actually made performance worse. Their explanation, Feedback Intervention Theory, is that feedback fails as attention shifts away from the task and toward the self. The sandwich is a self-attention machine. It opens by judging you (favorably), pivots to judging you (unfavorably), then judges you once more on the way out. The single actionable sentence in the middle competes with three verdicts about your worth, and the verdicts win. People walk out remembering the bread.

Trust erodes. Even the bread is stale. Carol Dweck's mindset research shows that person-praise – "you're a rockstar" – pushes people toward protecting the image rather than taking learning risks. Executive coach Jack Cohen calls these "flattering judgments" and puts them in the same family as critical ones. So the sandwich pays out twice in the wrong currency: praise that undermines and criticism that doesn't land. And once appreciation becomes a delivery device, people discount yours even when it's genuine.

To be fair to the sandwich, nobody ran a controlled trial on it by name. What the research tested are its ingredients – evaluative tone, person-focus, mixed messages – and each ingredient fails on its own.

What should you do instead of the feedback sandwich?

Separate appreciation from correction, then structure the correction so it points at behavior and forward motion. Five moves, each with evidence behind it:

1. Lead with the shared goal. John Gottman's marriage research found that lasting couples open conflict by naming the "dream within the conflict" – what they want more of, and why. Watching just three minutes of a conflict conversation, Gottman predicted which couples would be happy, distressed, or divorced six years later with 93% accuracy; most people, including couples therapists, average about 50%. Openings decide conversations.
2. Describe actions and impacts, not character. "Response times to client emails averaged 4.2 days last month" lands where "you're being unprofessional" detonates. The Center for Creative Leadership's SBI model (Situation-Behavior-Impact) is the classic structure, and studies on it show lower anxiety for givers and less defensiveness in receivers.
3. Say the care out loud. One 19-word sentence from a Stanford study more than quadrupled how many students acted on feedback. More on it below.
4. Check your readiness before you book the meeting. If you're walking in to win, wait a day.
5. Close with who will do what by when. Feedback without a next action is just commentary.

How does the GAIN framework work?

The best packaging of this evidence I've found is the GAIN framework, published by Jack Cohen in Lenny's Newsletter in 2025. Cohen spent 11 years coaching hundreds of founders, executives, and managers, and GAIN is his synthesis of research spanning couples therapy, decision-making, and workplace collaboration. The letters are the structure: Goal (what both parties ultimately care about), Actions (what needs to change to get there), Impacts (why), and Next actions (who will do what by when). If you want a single feedback sandwich alternative to remember, this is it.

The philosophy under the acronym matters more than the letters. Cohen's core claim is that it's profoundly more effective to frame feedback around what you both want to move toward – the gain – than around what you want to move away from – the pain. He isn't guessing. In one study by Amos Tversky and colleagues, doctors given identical data made the most effective treatment decision 50% of the time when it was framed as "the mortality rate is 10%" and 84% of the time when framed as "the survival rate is 90%." Same facts, different frame – and the threat frame degraded expert judgment. A feedback conversation opens under a heavier threat load than a statistics problem.

Here's Cohen's worked example, an engineering lead frustrated by sloppy handoffs: "I want us to reduce handoff friction so we can ship faster. The last handoff meeting we had, you didn't share any background customer research, which I failed to flag at the time. Lacking that context led to multiple rounds of engineers pinging you with questions" – followed by a closing ask to include the research next time. Goal, observed action with the giver owning their own part, impact, next action. No bread anywhere.

Two details from Cohen's write-up stick with me. First: in over 50 feedback workshops he's run, there has never been one where everyone agreed on whether a given example was an observation or a judgment. Telling them apart is genuinely hard, which is why it's a skill and not a tip. Second, on the close: a meta-analysis of 52 studies found that adding "feel free to say no" to a request doubled the rate at which people said yes. Autonomy isn't the enemy of accountability. It's the price of a real yes.

Worth noting, too, that Cohen's centerpiece case is a VP using GAIN on her own founder – the structure survives a power gradient. That situation gets its own treatment in how to disagree with your boss.

Can you be direct and still be kind?

The research says directness and care aren't a trade-off; the sandwich just implements care badly. In a randomized, double-blind Stanford study, teachers added one sentence to essay feedback: "I'm giving you these comments because I have very high expectations and I know that you can reach them." Compared with a placebo note, that sentence more than quadrupled the number of students who revised their work, and the revisions were on average 26% better. Nineteen words. That's the entire care layer the sandwich spends three paragraphs failing to deliver – stated once, honestly, and attached to a standard instead of to flattery.

Kim Scott's Radical Candor makes the same point structurally: care personally and challenge directly. Drop the challenge and you get what she calls ruinous empathy, which is the sandwich's natural habitat. Directness with care also shows up in how you open the conversation – the Gottman soft startup is the companion skill here, and if you want exact opening language, we keep a working list of phrases that open conversations.

Are you ready to give this feedback?

Frameworks fail when the person running them is still angry. In Conflict Campaign training we use a feedback-readiness check before anyone practices delivery – the Engaged Feedback Checklist, adapted from Brené Brown's engaged-feedback work into our kits. The full version runs eleven items; the four I lean on hardest are questions you can ask in the hallway before the meeting. Am I ready to sit next to you rather than across from you? Am I willing to put the problem in front of us rather than between us? Am I open to owning my part? Can I acknowledge what you do well instead of just picking apart your mistakes?

If the honest answer to any of those is no, the conversation can wait a day. It can't wait a quarter. A Bravely study found 70% of employees avoid difficult conversations, and avoidance runs its own tab – we've written up the cost of conflict avoidance at work separately. Readiness is a delay of hours, not months.

What does constructive feedback sound like? Three before-and-after examples

The sandwich, rebuilt. Before: "You're doing an awesome job with candidates, they love you. It'd be great if interviews got scheduled a little faster. Anyway, thanks for everything you do on hiring." After, from Cohen's article: "When you don't schedule interviews in the time frame you said you would, it slows down our hiring process. Can you schedule those this week and let me know if there are any blockers?" The rewrite names the action, the impact, and a dated next step – and the blocker question keeps it a dialogue instead of a verdict.

The vague judgment, grounded. Before: "Honestly, you're being unprofessional with clients." After: "Response times to client emails averaged 4.2 days last month. Slow replies put renewals at risk. What would get us to same-day responses – and is anything on my side creating the backlog?" The observation carries a number the judgment never could, and asking about your own contribution widens the levers for change. Cohen's phrase for this: more levers, more leverage.

The empty praise, made useful. Before: "You're a rockstar. You crushed that launch." After: "Your approach to gathering stakeholder input on the launch was thorough – support and engineering were in the room before the spec was written, and the review had zero surprises. Do that on every project." Process-praise keeps people in a learning orientation; person-praise turns their next failure into an identity threat. Positive feedback deserves structure too, and this is how praise gets its meaning back after years of sandwich duty.

Reading a framework and running one while your pulse is up are different skills, and only reps close that distance – that gap is the reason our two pilot runs of Operation Aetherfall were built around practicing these conversations rather than hearing about them. But the frameworks are where the reps start.

Skip the bread. Name the goal, describe the behavior, say the true thing – and let your praise mean something again.