How reviews decide who AI assistants recommend

Why engines lean on reviews

An engine answering a recommendation question needs a reason it can state in one line. Your own website says you are good; a platform with thousands of independent ratings says so in a form the engine can cite. In the answers we measure for local services, the businesses that are named are frequently introduced with exactly that: a rating, a count, and the platform it came from.

This does not mean reviews are the only factor. A business still has to be readable, relevant to the question and consistent across sources. But where several candidates qualify, review evidence is often what separates the one that is named from the ones that are not.

Which review sources matter

The sources depend on the category, and the only reliable way to know is to look at what the engines actually open for your questions. The pattern differs a great deal between sectors:

• Beauty, wellness and other bookable services: booking platforms such as Treatwell, Fresha and Booksy are read heavily.
• Agencies and B2B services: directories such as Clutch and Sortlist, and Trustpilot.
• Software: G2, Capterra and similar comparison sites.
• Almost every category: Google results and community discussion on Reddit.

Your own site is the one source the engine treats as interested. Reviews quoted on your site help less than the same reviews on a platform the engine reads independently.

Volume, recency and consistency

Three properties of review evidence show up in the reasons engines give. Volume, because a rating based on many reviews is stated with more confidence. Recency, because an engine searching the web sees what is current. Consistency, because a business rated well on one platform and absent from the one the engine read for that question does not benefit.

Collecting reviews without breaking the rules

In the UK, the Digital Markets, Competition and Consumers Act 2024 bans fake reviews and misleading review practices, and the platforms enforce their own rules on top. A review programme that helps your visibility is also one that would survive scrutiny:

• Invite every customer, not only the ones you expect to be happy.
• Never write reviews for customers or give them text to post.
• Do not offer anything in return for a review unless the platform allows it and it is disclosed.
• Use the platform's own invitation method where one exists, so reviews are marked as verified.
• Answer criticism in public; a measured reply is also something an engine can read.