Why a five-star review isn't evidence an agent can use.

A rating is a summary of opinions.

A shopper wants waterproof hiking boots for a wet weekend in the Lake District, and needs to be able to send them back easily if the fit is wrong. Their agent finds two listings: one carries a 4.9-star rating from six hundred reviews, the other states plainly that the boots are waterproof to the ankle, with returns accepted free within 60 days. The agent cannot act on the star rating the way it can act on the second listing, and that gap says something real about what counts as evidence, not a flaw in the review score.

Six hundred people rating something 4.9 stars tells the agent that most buyers were satisfied, which correlates with quality but is not itself a fact about this specific pair of boots, this specific return window, or whether they will keep this shopper's feet dry in this weekend's weather. It is evidence about the population of past buyers, giving a general impression rather than a checkable claim about the product in front of the agent right now.

It cannot be scoped to the one thing that matters here.

The stated return policy applies directly and specifically: 60 days, free postage, no argument required. The star rating cannot be scoped that precisely. A boot rated 4.9 stars overall might still run narrow in the toe for exactly this shopper's foot shape, and nothing in an aggregate score says so either way. An agent checking one specific request needs facts that apply to that request, not an average built from everyone else's different ones.

It has no source an agent can hold accountable.

If the returns policy turns out to be wrong, the business is answerable for it, since it is their own stated term. If the star rating turns out to have been inflated, or quietly gamed by a batch of early reviews, there is no single party an agent can point to the way it can point to a business misstating its own policy. That difference in accountability is a large part of why one counts as usable evidence and the other counts as a softer signal sitting alongside it.

None of this requires the rating to be fake for the point to hold. Six hundred genuinely happy customers is a real, honest number. It simply is not the same kind of thing as a business standing behind a specific, checkable term, and treating the two as interchangeable is where the confusion starts.

This overlaps with a shorter version of the same test this site's own piece on what an agent checks before it trusts a claim sets out in full: source, scope, and an accountable party behind it. A star rating simply never clears the first two of those on its own, however many people left one.

None of this makes reviews worthless.

A high rating is still a reasonable tie-breaker once the hard facts, price, stock, returns terms, are already sound on both options, and a genuinely low rating is still worth an agent flagging even when everything else checks out cleanly. The mistake is reaching for the star rating first, or letting it settle a question that only a specific, sourced, scoped fact can answer.

A business's actual commercial terms deserve to be just as easy to check as the star rating sitting next to them. Selfe connects for exactly that, so a genuinely good return policy is not competing for attention with an average score that was never built to do the same job.

Do good reviews still matter for AI shopping tools?

Yes, as a supporting signal once the specific facts, like stock, price and returns terms, are already confirmed. They are not a substitute for those facts.

What makes a fact "usable evidence" rather than just a good sign?

A source that can be held to it, a clear scope for what it covers, and a way to tell how current it is. A star rating has none of the three in the way a stated policy does.