AI shopping answers are starting to separate seller ratings from product ratings

Recent shopping queries across ChatGPT and Gemini show a meaningful shift in how AI assistants handle marketplace reputation: seller ratings are being surfaced as a distinct piece of evidence, separate from the product's own review score. Instead of a single blended star rating, answers are increasingly citing "4.8 from 12,000 merchant ratings" alongside the item's own review count and average.

• Two-tier trust data. Assistants now read aggregateRating at both the Offer/merchant level and the Product level, and cite each where it matters — seller reputation on price-and-service comparisons, product ratings on quality comparisons.
• Seller history is a filter. Summaries are more likely to surface a merchant rated below a threshold only when no better option exists, which effectively folds store reputation into the "should I trust this source" decision.
• Halos are breaking. A strong product rating no longer fully masks an average seller score, and vice versa. The two are treated as independent signals rather than one blended number.

For marketplace sellers, this is the difference between losing a comparison answer because of shipping or service data you can improve, and losing it for reasons you never see. If an AI assistant is quietly penalizing your page for seller-level metrics you are not tracking, you will watch competitors get cited without understanding why. The fix is to make seller reputation visible and structured in the same disciplined way as product reviews: expose merchant ratings in schema, keep them current, and ensure store-level trust data is consistent across marketplaces.

Reputation is becoming multi-dimensional citation data, so merchandisers need to audit it the way they audit reviews and schema. Visora's free scan at geovisora.com/audit shows, for a single product URL, which trust and rating fields an AI assistant can actually read — including whether your seller-level signals are extractable or silently missing. If ratings are splitting into two earned citations, the merchants who structure both layers will be the ones plotted back in.