GEO
Reviews matter more now that AI reads them
· 2 min read · Oddbird
Ask an AI assistant for a recommendation and you often get a summary of what customers write about a business. Reviews are no longer just stars next to your name, but a source that gets read and quoted.
What you take away
- A model reads the text of reviews, not just the score. What customers mention specifically comes back in the answer.
- Recent reviews say more than old ones. A profile that has not grown in a year looks quiet.
- Ask at the right moment, make it easy and respond to what is written. Fake reviews and rewards are prohibited and do not work either.
Reviews used to matter mainly when someone looked at your profile: four point eight stars, that looks good. Now something happens earlier in the process. Someone asks an AI assistant which physiotherapist nearby is good, and the answer contains a sentence like: customers mainly mention the short waiting time and the clear explanations. That sentence comes from somewhere. From your reviews, or from someone else's.
What a model takes from reviews
An assistant that searches the web reads reviews as text. The score counts, but the content does the work: which words keep coming back, what people are happy about, what they complain about. From that it builds a short description of your business, and that description ends up in the answer.
- Content over stars: a review with two sentences about what went well tells a model more than ten five-star ratings without text.
- Repetition: when several customers independently mention the same thing, it becomes a trait of your business.
- Recency: recent experiences weigh more, especially in sectors where people and offerings change.
- Replies: how you respond to criticism is also text that gets read.
Where it usually goes wrong
For most businesses we look at, the problem is not that customers are unhappy. The problem is that happy customers write nothing. The reviews that are there are old, short, or from that one customer who was genuinely angry. The picture a model forms of you is then thin, and sometimes skewed.
More reviews, honestly
- Ask when things went well. Right after the appointment, the delivery or the solved problem, not in a newsletter three weeks later.
- Make it one tap. A short link or QR code to the place where the review should go, without anyone having to search.
- Help with the content, not with the verdict. A question like "what stood out to you?" produces useful text. A sample text to copy does not.
- Reply to every review, including the positive ones. Short, personal and without a stock phrase.
- Choose the places that matter. For a local business that is almost always the Google profile, plus the platforms your sector is rated on.
You do not need to invent reviews. You only need to ask the happy customers you already have, at the right moment.
What not to do
Buying or writing fake reviews has been prohibited by law in the Netherlands since 2022, and rewarding a review with a discount or a gift is against Google's rules. More importantly, it backfires. A row of empty five-star reviews on the same day is exactly the pattern platforms recognise, and a model learns nothing useful about your business from it.
Also avoid approaching only happy customers and deliberately skipping the rest. Google does not allow selective asking. Ask everyone; the few critical reviews that produces make the whole more credible.