Why do AI assistants pay attention to reviews?
Because reviews are independent, specific and plentiful, which is exactly the kind of source a model trusts. When ChatGPT or Gemini writes a sentence like this firm is well regarded for boiler repairs in Huddersfield, it is usually paraphrasing reviews it has read, on Google, on a directory or on your own site. Reviews also carry the words customers actually use, so they match the way questions are asked. A model deciding between two similar firms tends to lean on the one whose reviews describe the job in question, because that is the closest thing it has to a reference.
Which reviews actually help GEO?
The ones that read like a short case study. What the job was, where, roughly when, what the customer was worried about, and how it went. Fay Dolphin’s review of our own work says what was built, that questions were answered, and that she would recommend us, which is more useful to a model than a row of stars. A review that said, say, new boiler fitted in Lindley on the day promised, tidy and clearly explained, would give an assistant three facts to match: the service, the place and the manner.
Recent reviews matter more than old ones, and a steady trickle matters more than a burst. Twenty reviews in one week followed by silence looks, to a model and to Google, like a campaign.
How do I ask Huddersfield customers for reviews that say something?
Ask at the moment the job finishes, in person, and ask a specific question. Would you mind leaving a review that says what we did and where? gets a usable answer far more often than a link buried in an invoice. Send the direct link to your Google review form the same day. For trades, a card left with the customer with a QR code on it works well, because the work is fresh. Do not offer anything in return, because Google’s rules forbid it and a model can spot the pattern. And reply to every review, good or bad, in a sentence that repeats the job and the place, because your replies get read too.
Do reviews on my own website count?
Yes, if they are real and marked up as such. Quoting Google reviews on your site with the customer’s name and the job puts the evidence next to the service it relates to, which is where a model reading the page wants it. The case studies on this page carry the client’s own Google review for that reason. Structured data can mark them as reviews, and that helps a machine parse them. What does not work is inventing them, or copying the same three onto every page; an assistant that reads the same quote in six places starts to discount it.
Do reviews on other platforms help AI visibility?
Yes, and the mix matters more for AI answers than it does for the map pack. ChatGPT and Perplexity cannot read Google’s listing directly, so reviews on Checkatrade, Trustpilot, Facebook, a trade association site or a supplier’s page are often what they see. For a Huddersfield trade, Checkatrade and Google between them cover most of it. For a B2B firm like a flooring contractor, a written testimonial on a client’s or supplier’s own site is often worth more than a dozen consumer reviews, because it is the kind of source a model treats as authoritative.
What should I do if the reviews are old or there are hardly any?
Start this week, and start with the customers you know were pleased. Five specific reviews in the next two months change what an assistant can say about you more than fifty vague ones from 2021. Reply to the old ones too, so the profile looks attended to. And put the two or three best on the relevant service pages of your site, with names. It is slow, ordinary work, and it is the part of GEO with the best return for a small firm, because nobody can buy it.
