Why do humans accept recommendations from people they perceive as similar?
This explores the social mechanics of why perceived similarity makes a recommendation persuasive to people — and what the corpus reveals about whether that instinct actually serves us well.
This explores why perceived similarity makes us trust a recommender — and the corpus has a surprising answer: similarity works as a *persuasion signal* far more than as a *quality signal*, and the two come apart in interesting ways. The clearest direct evidence comes from analysis of how humans actually recommend to each other. When researchers studied 1,001 real recommendation conversations, the moves that landed weren't careful questions about your preferences — they were sociable gestures: sharing a personal opinion, offering encouragement, signaling "I'm like you," and making credibility appeals Do recommendation strategies beyond preference questions work better?. Opinion-sharing showed up in 30% of recommendation sentences and experience-sharing in 27%. So similarity isn't just background trust — it's an active rhetorical strategy people deploy, and it works.
Why does it work? One thread in the corpus reframes recommendation as fundamentally *communicative* rather than informational. Expert judgment, on this view, isn't just retrieving the right answer — it's anticipating what an audience will accept as valid Can AI replicate the communicative work experts do?. A similar person has already done that anticipation for you: they share your reference points, so their suggestion arrives pre-fitted to your context. You accept it because it *sounds like it was made for someone like you* — which it was.
But here's the twist the corpus drives home: similarity is a great trust heuristic and a mediocre accuracy heuristic. In algorithmic terms, recommendations built on people *unlike* you often outperform homophily-based ones — friends with different tastes surface the anomalous, outside-your-usual choices that pure taste-matching never would Can friends with different tastes improve recommendations?. The value of a social tie comes from influence on the unexpected, not from confirming what you already lean toward. And when systems try to exploit near-similarity directly, it backfires: profiles that are *almost but not quite* like yours produce the worst personalization errors of all — an uncanny-valley effect where the system confidently applies preferences that are subtly wrong, more damaging than an obvious mismatch Why do similar user profiles produce worse personalization errors?.
This is worth sitting with, because the same dynamic plays out at scale. Recommendation feeds function as persuasion infrastructure, and the type of recommender shapes whether the people it gathers converge or diverge in opinion — each format quietly sorts audiences by shared prior expectations, then reinforces them Do different recommender types shape opinion convergence differently? How do recommendation feeds shape what people see and believe?. The very similarity-acceptance instinct that helps a friend persuade you is the lever that lets systems herd populations toward agreement.
So the honest synthesis: we accept recommendations from similar people because similarity signals shared context, credibility, and the sense that the advice was tailored to us — and that instinct is socially reasonable. But the research suggests it's a comfort heuristic, not a correctness one. The recommendations most likely to teach you something new tend to come from people just different enough to see what you can't.
Sources 6 notes
Analysis of 1,001 human recommendation dialogues shows successful recommendations correlate with personal opinion sharing, encouragement, similarity signals, and credibility appeals—not just preference questions. Opinion and experience sharing appear in 30% and 27% of recommendation sentences respectively.
Expertise requires anticipating audience acceptability and social validity, not just retrieving information. AI lacks the mechanism to perform this communicative work, making its fluent output epistemically misleading despite its confident form.
Social Poisson Factorization uses friends' diverse tastes to recommend items outside users' usual preferences, outperforming methods that pull friends' representations together. Networks add value through influence on anomalous choices, not taste similarity.
PRIME shows a U-shaped error curve where most-similar profile replacements cause steepest performance drops. The model confidently applies wrong preferences when profiles are nearly but not truly matched, an uncanny valley effect more harmful than obvious mismatch.
Research shows that frequently-bought-together and co-viewed recommendation networks produce different opinion convergence patterns. The mechanism: each recommender type attracts different audience segments with different prior expectations, shaping both who sees products together and how they rate them.
Research shows recommendation systems operate as political actors: feed weights influence producer behavior, network topology drives opinion convergence, and automation enables targeted persuasion at population scale. These effects compound through rating contamination and selection biases.