Do reviewers write about objective product quality or personal experience?
This explores whether online reviews actually report something true about the product, or mostly broadcast the reviewer's own situation, social posture, and feelings — and the corpus comes down hard on the 'personal experience' side.
This explores whether online reviews actually report something true about the product, or mostly broadcast the reviewer's own situation, social posture, and feelings. The collection's answer is unsettling: reviews are far more about the reviewer than the thing being reviewed. Before a single word gets written, two selection filters have already bent the data — only people who expected to be satisfied buy in the first place, and only some of them bother to review, so the aggregate measures self-selected preferences rather than objective quality Do online reviews actually measure product quality or just buyer preferences?. Participation cost sharpens this: small frictions mean only people with strong opinions show up, producing U-shaped distributions where lukewarm-but-honest middle experiences simply vanish Why do people bother writing online ratings at all?.
Even the words that do get written aren't a clean read of personal experience. Reviewers perform for an audience. One striking finding: people lower their public ratings after reading negative reviews — even when their own experience was positive — because negative reviewers come across as more intelligent, and writers want to look smart. Private raters, with no audience to impress, show no such shift Why do online reviewers publish negative ratings despite positive experiences?. So a 'review' is partly a self-presentation move, calibrated to the room. The ratings themselves also drag each other around over time: prior ratings measurably shape later ones, and that social-dynamics influence compounds through future reviews rather than washing out Do online ratings actually reflect independent customer opinions?.
Here's the lateral turn you might not expect — the *type of product network* a review lives in changes what it says. The same item rated inside a 'frequently bought together' network versus a 'co-viewed' network converges differently, because each network funnels a different audience with different prior expectations to the product Do different recommender types shape opinion convergence differently?. Quality isn't being measured against a fixed yardstick; the yardstick shifts with who's holding it.
The AI-review work makes the personal-experience dependency concrete by showing what it takes to *recover* it. Off-the-shelf models trained with RLHF are too polite to write the honest negative review a dissatisfied user would — you have to feed in the user's behavioral history and explicit satisfaction signals before the model will produce an authentically critical review matching that person Can user history override an LLM's politeness bias in reviews?. The signal lives in the individual's history, not in any neutral assessment of the product. If you want something closer to grounded evaluation, the corpus points away from isolated star-ratings entirely: comparative explanations that reference other items carry more decision-relevant information, because that's how people actually judge things — relative to alternatives, not against an absolute scale Do comparisons help users evaluate items better than isolated descriptions?.
So the honest summary: reviewers mostly write personal experience dressed as objective quality, and even the 'personal' part is shaped by who's watching, what others said first, and which crowd the product attracts. If you want to know something true about quality, the more reliable move is to read reviews *comparatively* and treat any single aggregate rating as a portrait of a self-selected, audience-aware crowd rather than a measurement of the product.
Sources 7 notes
Only consumers expecting satisfaction purchase and review, creating two selection filters. Research shows early reviewers shape later perceptions, altruism affects learnability, and summary statistics can actually slow quality discovery. Observed ratings misrepresent the satisfaction distribution of all potential buyers.
Lafky's experiments show raters care about both buyers and sellers rather than purely one or the other. Small participation costs create U-shaped distributions where only strong-opinion raters engage, biasing average ratings away from true quality.
Posters systematically reduce their ratings in public when exposed to negative reviews, even with positive personal experience—because negative reviewers appear more intelligent. Private raters show no such shift, revealing a self-presentational mechanism tied to multiple-audience communication.
Moe and Trusov decomposed ratings into baseline quality, social-dynamics influence, and error, finding that prior ratings meaningfully affect subsequent ones. These effects have both immediate sales impact and long-term compounding effects through future ratings, though high opinion variance can eventually dampen the distortion.
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.
Review-LLM defeats the politeness bias inherent in RLHF-trained models by aggregating user behavior sequences (prior reviews, item ratings) in the prompt and fine-tuning on these contextualized examples. This dual intervention—personalized context plus explicit satisfaction signals—allows the model to generate authentically negative reviews matching user dissatisfaction.
Relational explanations that compare items carry more decision-relevant information than isolated evaluations because they match how humans naturally assess products. A system extracting aspects from reviews and generating aspect-controlled comparisons produces sentences rated as both accurate and useful for purchase decisions.