This study examines how the linguistic subjectivity of online reviews distorts the aggregate merchant rating, the system-level trust signal on which buyers in complex digital markets increasingly rely. It asks whether subjectivity biases the rating, through what mechanism, and under what conditions the bias is stronger or weaker.
Drawing on positive-negative asymmetry theory and a second-order, cybernetic reading of review platforms as trust-generating systems, the study analyses 17,349 French-language reviews from Trustpilot for three merchants (Showroom, Amazon and Veepee). Natural language processing (NLP) extracts the subjectivity and polarity of each review, a moderated-mediation model is estimated with Hayes' PROCESS macro, and the analysis is replicated across merchants.
Subjectivity is not valence-neutral. It drives reviews toward negative polarity and, through that negativity, lowers the merchant rating. The distortion is amplified for longer reviews and for high polarity self-selection reviewers (occasional contributors), and dampened for prolific, better-calibrated reviewers.
Platforms and merchants can treat artificial-intelligence sentiment tools as monitoring instruments and read ratings in light of review length and reviewer history rather than at face value.
The paper reframes the merchant rating as a cybernetic trust signal and shows subjectivity to be a systematic, conditional source of distortion in that signal, extending review research from the helpfulness of single reviews to the integrity of digital trust.
