Review and Rating Signals AI Can Verify
Why a rating without a review count gets ignored, how to reconcile conflicting numbers, where review markup belongs, and what you must never do.
Why reviews carry more weight in AI answers than in search
In traditional search, a review profile mostly affects how your listing looks. In an AI answer it affects whether you are named at all. When a model is asked to recommend a provider, an independently corroborated review record is one of the few signals it can verify without trusting your own site. That is the distinction that matters: everything on your website is a claim you make, and a review profile is evidence someone else produced. Models weight the second kind far more heavily.
Where reviews actually live
The platforms that matter are the ones a model can read and cross-check: Google Business Profile, TripAdvisor, Facebook, Trustpilot, Yelp where it applies to your trade, and any industry-specific directory that carries ratings. A review on your own website is worth having for conversion, but it is not independent evidence, because you control whether it appears. Concentrate effort where the ratings are visible to strangers and verifiable by machines.
Publishing a rating: the rule that gets you ignored
If you publish aggregateRating, then ratingValue and reviewCount must both be present and both must be true. A rating without a count is not quotable and will be ignored. Never estimate the count, and never round up — the first thing a checking model does is compare your structured rating against the figure on the third-party page it can actually read. A 4.8 declared as 5.0 does not read as optimism; it reads as a claim that failed verification, which is worse than having made no claim.
Your visible copy and your markup must agree
Before adding any review markup, make the page state one set of numbers. A page that says "5-star across TripAdvisor, Google and Facebook" in the hero, "4.5/5 based on 80+ reviews" in a badge, and "50+ reviews" elsewhere has contradicted itself three times. A model reading all three has to choose which to believe, and the usual outcome is that it says nothing rather than risk repeating something wrong. Reconcile the copy first; the structured data then simply restates what is already visible.
Getting more reviews, legitimately
Ask every satisfied customer, at the moment the work is finished, with a direct link that goes straight to the review form. Make the ask specific and the friction near zero — a one-tap link in a follow-up message converts far better than a request to "find us on Google". Follow up once, not repeatedly. Respond to every review, including the poor ones, because a considered reply to a bad review is itself evidence of how you operate.
What you must not do
Do not buy reviews, do not write them yourself or have staff write them, and do not offer incentives, discounts or entries into a draw in exchange for a rating — every major platform prohibits it and all of them can detect it. Do not gate reviews by asking only happy customers, because selective collection is itself a violation and it makes your rating less believable, not more. Do not fabricate a review count in structured data. These tactics do not merely risk a penalty: they create exactly the mismatch a model is looking for when deciding whether your other claims are trustworthy.
Where review markup goes
Put aggregateRating on the LocalBusiness node, alongside name and areaServed. If you also publish individual reviews as Review nodes, each one needs an author and a reviewRating, and the reviews must be visible on the page — marking up a review a visitor cannot read is a guidelines violation. For a service-area business with no storefront, this is still valid: the rating belongs to the business entity, not to a premises. If you have no reviews yet, add nothing and wait. An absent rating costs you less than a wrong one.
How to verify it worked
Check the rendered page source to confirm the block is present and contains no placeholder values. Run it through Google's Rich Results Test and the Schema Markup Validator — the first tells you whether the markup is usable, the second whether it is structurally valid. Then open the third-party profile you took the numbers from and confirm they still match. Review counts change, and markup that was accurate in March is quietly wrong by June. This is one of the few things on a site that needs periodic re-checking rather than a one-time fix.
Don't want to DIY?
Review collection, profile consistency and correctly structured rating markup are part of every Click Elevate plan. We verify each figure against its source before publishing it, because a rating that cannot be corroborated damages every other signal on the page.
This guide resolves these report findings
- Review schema
Run the free AEO score to see which of these apply to your site.
Did this guide answer your question?
See how visible your business is to AI
Get your free 0–100 AEO score across ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews.