AI-generated review summarization (AIGRS), a form of AI-generated content extracted from User-Generated Content (UGC) and presented in the same format at the top of review sections, have been adopted by major e-commerce platforms. This feature exhibits key characteristics of AI-generated content, including low emotional intensity and information neutrality. Does the presence of AIGRS influence subsequent user reviews? Will the review sentiment intensity align with that of AI-generated content ?
We collected two datasets from the online platform: store information and review data. By linking these datasets via store IDs, we aggregated them into a store-week level dataset (N = 19,526). To preprocess the review content, we used the Jieba library for Chinese word segmentation and part-of-speech tagging in the review content. To robustly establish the causal effect of AIGRS, we combined propensity score matching with a difference-in-differences (DID) framework to rigorously establish the causal effect of AIGRS.
The findings indicate that introducing AIGRS creates a sentiment convergence effect on subsequent reviews, which is moderated by store rating distribution type and product type.
The conclusions highlight the systemic impact of AI technology on interactive marketing, enrich research on AIGC and UGC, and offer strategic insights for better AI technology utilization for platforms.
This study reveals that AIGRS trigger a sentiment convergence effect, systematically reducing emotional intensity in subsequent user reviews through cognitive anchoring and normative signaling.
The effect is governed by a dual-path mechanism where AIGRS sets an emotion-compressed cognitive anchor while users adapt their expression to this perceived platform norm.
Its strength is context-dependent: weakened in high-rating-cluster and healthy food stores (linked to ethical concerns and identity construction), but more pronounced in low-rating-cluster and unhealthy food stores.
We advise platforms to adopt dynamic AIGRS designs—adapting to community rating patterns and preserving emotional cues for identity-relevant products—to balance informational efficiency with authentic user voice.
