How to Win Friends: What the Research Actually Supports

Every claim here is checked against the primary research. How we check.

TL;DR

Carnegie’s central bet — that liking and cooperation are produced by learnable behaviours, not by charisma — is the rare self-help claim that later research mostly confirmed.
The strongest vindication is his least intuitive rule. In three preregistered field experiments, 230 canvassers talking to 6,869 voters found that conversations built on arguments alone changed nothing, while otherwise identical conversations that swapped stories non-judgementally produced attitude change lasting four months (Kalla and Broockman, American Political Science Review, 2020).
«Never criticise» is wrong as written but right in its mechanism: across 607 effect sizes and 23,663 observations, feedback raised performance on average (d = .41) yet lowered it in over a third of cases, and it failed most reliably when it pointed at the person rather than the task (Kluger and DeNisi, Psychological Bulletin, 1996).
The two techniques everyone quotes are the two with the thinnest support. The name evidence rests on 42 dining parties and 22 students; the Ben Franklin favour rests on one 1969 study that PubMed shows has never been directly replicated.
The sincerity Carnegie demanded turns out to be optional for the effect and essential for the outcome: ingratiation works, but its payoff collapses when the target sees the motive (Gordon, Journal of Personality and Social Psychology, 1996).

How to Win Friends and Influence People by Dale Carnegie — cover

Verdict

Read our notes instead — but treat the advice as recommended. This page previously carried a «read it» tag while its own text told you to keep the tools and skip the book. The tag was wrong. The evidence supports Carnegie’s behaviours, not Carnegie’s 250 pages.

That is an unusual split for this site. Most books we check are wrong. «How to Win Friends and Influence People» is largely right and almost entirely unevidenced — a 1936 sales trainer’s field notes that happened to anticipate findings nobody would run for another sixty years. The behaviours survive testing. The book that carries them is anecdote, survivorship and 1930s department-store scenery, and every rule worth having fits on one page. See our other book reviews for the same triage applied to books that fared worse.

The claim on trial

Strip the stories and Carnegie makes one falsifiable claim: a short list of learnable behaviours — remembering and using a person’s name, showing genuine interest, letting the other person talk, refusing to criticise, admitting your own errors first, avoiding arguments — reliably increases how much people like you, how much they cooperate, and how far you can influence them.

That is testable three ways. Does each behaviour, manipulated on its own, move liking or cooperation? Does it work outside a laboratory? And does it still work when the target can see you are doing it on purpose? Carnegie ran none of those tests. He proved his points with stories about salesmen who closed and managers who turned people around, and you never meet the ones it failed on. His book contains no citations, because in 1936 there was nothing to cite.

There is now. Below, each rule is checked against the strongest study we could find and verify.

What the research backs

Be genuinely interested; ask questions. This is the best-supported rule in the book. Across three studies of live dyadic conversation, people who asked more questions — especially follow-up questions — were better liked by their partners, and the effect ran through perceived responsiveness (Huang et al., Journal of Personality and Social Psychology, 2017). In the face-to-face speed-dating study, a trained follow-up-question detector showed that daters who asked more of them were more likely to get agreement to a second date. The same paper found that people do not expect this to work, which is exactly why a book had to tell them.

Let the other person do the talking. Carnegie says people would rather talk about themselves than hear about you. Neuroimaging supports the premise: self-disclosure engages the brain’s mesolimbic reward regions, and participants gave up money for the chance to disclose about themselves rather than answer questions about others (Tamir and Mitchell, PNAS, 2012). Talking about yourself is paid work for the listener and a reward for the speaker.

Listen instead of preparing your rebuttal. A registered systematic review pooled 664 effect sizes across 400,020 observations and found perceived listening correlated with work outcomes at r̄ = .39, with the effect on relationship quality (r̄ = .51) larger than on performance (r̄ = .36) (Kluger et al., Journal of Business and Psychology, 2023). The authors flag the obvious weakness themselves: 75% of that literature is self-report. The experimental work is cleaner. Four experiments manipulating listening quality found that high-quality listening reduced actual attitude extremity during disagreements, via less defensive self-reflection (Itzchakov et al., Journal of Personality and Social Psychology, 2024), and an earlier set showed listening raised speakers’ attitude clarity (Itzchakov et al., Personality and Social Psychology Bulletin, 2018).

Avoid the argument. Carnegie’s most-quoted line — «the only way to get the best of an argument is to avoid it» — is a slogan, and it has now been tested at scale in the hardest possible setting. Kalla and Broockman ran three preregistered field experiments on immigration and transgender attitudes: 230 canvassers, 6,869 voters, seven US locations. Face-to-face conversations using arguments alone had no effect on either policy views or prejudice. Identical conversations that also included a non-judgemental exchange of personal narratives durably reduced exclusionary attitudes for at least four months (d = 0.08). An earlier randomised trial in the same programme sent 56 canvassers to 501 voters for roughly ten minutes each and found reductions in transphobia that persisted three months and survived exposure to counterarguments (Broockman and Kalla, Science, 2016). Carnegie had no data. He had the mechanism.

Give honest appreciation. Four experiments, including a field study with university fundraisers, found that a brief expression of gratitude increased helpers’ subsequent prosocial behaviour — toward the person who thanked them and toward an unrelated third party — mediated by feeling socially valued rather than by feeling competent (Grant and Gino, Journal of Personality and Social Psychology, 2010). Carnegie’s «hunger to feel important» is a crude phrase for a real mediator.

Admit your error quickly and emphatically. A meta-analysis of 175 studies and 26,006 participants found apology among the strongest situational correlates of forgiveness (r = .42), behind state empathy (r = .51) (Fehr, Gelfand and Nag, Psychological Bulletin, 2010). Four further experiments found that apologising for things you did not cause — the rain, the traffic — still increased trust in the apologiser, via empathic concern (Brooks, Dai and Schweitzer, Social Psychological and Personality Science, 2014).

Where the evidence is thin

The name. Carnegie calls a person’s name «the sweetest sound in any language», and the phrase has outlived every study behind it. The one robust finding is adjacent, not identical: people prefer the letters of their own name to other letters, replicated across twelve European languages (Nuttin, European Journal of Social Psychology, 1987). The famous extension of that — that people disproportionately marry, move and take jobs matching their initials — was re-run by Simonsohn, who reproduced the original patterns and then showed they are explained by cohort, geographic and ethnic confounds plus reverse causality (Simonsohn, Journal of Personality and Social Psychology, 2011). The original implicit-egotism papers (Pelham, Mirenberg and Jones, Journal of Personality and Social Psychology, 2002) sat in textbooks for a decade first.

The direct tests of using someone’s name are small. A restaurant study randomised 42 two-person parties and found the tipping rate rose from 15.0% to 23.4% when the server introduced herself by name (Garrity and Degelman, Journal of Applied Social Psychology, 1990). A compliance study ran on 22 students (Howard and Gengler, Psychological Reports, 1995). The best of the set is three experiments showing that remembering a name increased compliance with a purchase request — but forgetting a name did not reduce it, and the mediator was the target feeling complimented, not the sound of the name (Howard, Gengler and Jain, Journal of Consumer Research, 1995). The behaviour has modest support. Carnegie’s explanation for it does not.

The Ben Franklin favour. The idea that asking someone for a small favour makes them like you traces to a single experiment (Jecker and Landy, Human Relations, 1969). Searching Crossref and PubMed on 10 August 2026 for «Benjamin Franklin effect» returns two PubMed records, neither of them about this at all, and no direct replication of the 1969 design. The closest modern test found that Japanese and American participants asked for help by a confederate did increase their liking of that confederate relative to baseline — but participants who helped without being asked did not (Niiya, The Journal of Social Psychology, 2016). One supportive study is not a refutation. It is also not a result you should build a networking strategy on. Untested is not the same as false.

The era’s evidence base. Carnegie wrote alongside the human-relations movement, which took its authority from the Hawthorne studies. When Levitt and List recovered the original illumination-experiment data — long thought destroyed — they found that «existing descriptions of supposedly remarkable data patterns prove to be entirely fictional», though subtler effects survived (Levitt and List, American Economic Journal: Applied Economics, 2011). The intellectual scaffolding around the book was weaker than the book.

Never criticise is the one rule that breaks

Carnegie opens with «don’t criticise, condemn, or complain» and never qualifies it. The feedback literature does not support a blanket ban, and the way it fails to support it is the most useful thing on this page.

Kluger and DeNisi meta-analysed 607 effect sizes over 23,663 observations. Feedback interventions raised performance on average, d = .41 — but more than a third of them made performance worse, and that could not be explained by sampling error or by whether the feedback was positive or negative. Their proposed mechanism is the part Carnegie would have recognised: feedback loses effectiveness as it draws attention up toward the self and away from the task.

So the rule should read: criticism aimed at the person degrades performance; correction aimed at the work does not. Carnegie got the direction right and the scope wrong. Following his version literally means withholding task feedback that would have helped — an error the book has no way to detect, because it never tracked outcomes.

The sincerity problem

Carnegie insists appreciation must be genuine and then supplies a toolkit that works without it. That tension is now measurable.

A meta-analysis of ingratiation research found a small positive effect on performance evaluations and a substantially larger one on liking — both qualified by the perceived transparency of the tactic, along with the direction of the influence attempt and whether the perceiver was the target or a bystander (Gordon, Journal of Personality and Social Psychology, 1996). Transparency is the whole ballgame. The techniques pay when the motive is invisible and stop paying when it is not.

Worse for Carnegie: they do not fully stop. Flattery with an obvious ulterior motive still leaves a favourable implicit attitude intact alongside the consciously discounted explicit one, and the implicit attitude drove downstream choices (Chan and Sengupta, Journal of Marketing Research, 2010). The tools work on people who have already seen through you. That is the honest case for calling the book manipulative — not that its advice is bad, but that its advice is effective independent of the sincerity it demands, and the demand has no enforcement mechanism.

The one caveat worth holding: none of this research randomly assigns sincerity. It manipulates whether the motive is visible. Whether an internally sincere listener outperforms a well-trained instrumental one, over months rather than minutes, has not been tested that we can find.

Who should actually read it

Read the whole book if you are early in a career that depends on people, tend to lead with correction, and have never been shown that asking a second question is a skill. The repetition that irritates everyone else is what makes it stick when you are starting from zero.

Skip it if you already listen well, or if you are reading it to get better at closing. In that second case the book is a competence upgrade with no ethical brake, and the ingratiation evidence says the brake is not decorative.

Anyone reading it for management should read Kluger and DeNisi alongside it. Carnegie’s «never criticise» applied inside a team is a way of making people comfortable and worse at their jobs.

One thing to try

Ask one follow-up question — a question about the thing they just said, not a new topic — and do not add your own related story. That single behaviour is what the trained detector in Huang et al. picked up on, and it is the one that predicted second dates.

If you want the harder version: next time you disagree with someone, spend the first five minutes finding out how they arrived at the position, and say nothing that starts an argument. That is the canvassing protocol, minus the clipboard.

Get the book

Find «How to Win Friends and Influence People» on Amazon — as an Amazon Associate, The Boring Work earns from qualifying purchases (disclosure).

The boring bottom line

«How to Win Friends and Influence People» is the most-vindicated unevidenced book in the self-improvement canon. Six of its rules now have real support, one of them from a 6,869-voter field experiment Carnegie could not have imagined. Two of the rules people quote most — the name and the Franklin favour — rest on 42 dining parties, 22 students and one 1969 experiment nobody has repeated. One rule, «never criticise», is wrong as stated and right about the mechanism.

The book itself adds nothing to that ledger. It is anecdote all the way down, and the anecdotes are the parts that aged. Take the six behaviours, ignore the sales stories, and hold on to the finding Carnegie would have hated: the techniques work on people who can see you using them, which means the only thing keeping this out of manipulation is you.

How to Win Friends: The Evidence, Technique by Technique