AI assistants now summarize your reviews before they recommend you

Your reviews are the raw copy an AI publishes about you

"Very professional, would use again." That's what a lot of businesses hand an AI assistant when a customer asks who to call. ChatGPT, Perplexity and Google's AI tools read the whole text of your reviews as language, then write their own short summary and show it to the person asking. The machine gets its wording from the sentences underneath your star average.

So whatever your customers typed is the copy a machine now reads out on your behalf. Almost nobody has ever told those customers what to write about. I pull up review pages in audits most weeks, and that gap shows up on nearly every one. Here's what this post is: one question to add to your review request, and one habit when you reply.

Why the summary gets trusted before the reviews do

BrightLocal ran its Local Consumer Review Survey in early 2026 across a panel of 1,002 US adults. The share of people using AI to find a local business went from 6% a year ago to 45% today. That puts AI third for local recommendations, behind Google and Facebook, ahead of Yelp and Tripadvisor. Over the same year, use of Google reviews slipped from 83% to 71%.

The same study asked the AI users what they do with the answer they get. Of the people who ask an AI for a local recommendation, 82% read the AI's written summary of a business's reviews. And 23% say they'd decide on that summary alone, without opening a single real review. Both numbers describe the people already using AI this way, which was 455 of the 1,002 surveyed.

That 23% is the one I keep coming back to. Those are the customers a machine is describing you to.

Some of that review data now reaches these tools through licensing deals. Trade coverage has reported that Yelp and Tripadvisor both announced arrangements that feed their review data straight into these tools, Yelp into ChatGPT and Perplexity and Tripadvisor into Perplexity. That is separate from anything the assistants pick up by reading the open web.

Your Google reviews get read too. When a machine writes a paragraph about your business, it's working from full sentences your customers wrote, pulled off several sites at once.

The customer is doing a version of the same thing by hand. BrightLocal found people now check an average of six different review sites before they pick someone. Six sites. For one plumber. "Consumers are looking for information in more places, more often," Myles Anderson, BrightLocal's CEO, told Street Fight in March 2026. The AI summary is now one of those places, and for a lot of people it's the first one they see.

Worth knowing what the customer does after that. Among the AI users in the survey, 88% check whether a review the AI quoted is real or go looking for where it came from, and 97% at least sometimes double-check an AI recommendation against the reviews themselves. So the summary sets what they expect before they ever land on your page, and then they read your actual reviews with that expectation in hand.

What a machine does with fifteen reviews that agree

The survey asked what makes a review believable to a reader. The strongest thing a review has going for it is other reviews saying something similar. Agreement across several reviews, recent dates, an average that looks human, and an owner who bothered to answer. BrightLocal reported that one without a percentage attached, so treat it as a direction.

A machine ends up in roughly the same place, for a plainer reason. It has to produce four or five sentences about you, and it can only use what your page actually contains. The star average gives it one number, and every business on the list has one of those. The sentences underneath are the only part that tells it what you're like to hire.

So it goes looking for what your reviews have in common. A detail that shows up in most of them is a safe thing to repeat back to a stranger, and safe is what gets written down. Something one customer mentioned and nobody else did usually gets left out, because there's nothing on the page backing it up.

That's the whole reason fifteen reviews saying the tech got there the same day are worth more than fifty saying you were great. The fifteen hand the assistant a sentence it can commit to. It can tell the next person you show up same day, and your own page is the proof.

When there's nothing in common to find, the assistant falls back on the only material left, which is your rating and how many reviews you have. What the customer reads is something like: rated 4.7 with about 60 reviews, seems well liked. That's true. It also describes half the businesses on the list, and it gives nobody a reason to call you first.

When I pull up a client's review page in an audit, I read it the way the assistant does. I look for the thing several reviews say in common, because that's the line that gets repeated. The pattern is the same nearly every time: a solid star average, plenty of reviews, and almost no sentence worth quoting.

I used to tell owners to just go get more reviews. That was thin advice. Volume with no detail in it leaves the assistant holding a number.

Ask one question, and answer with a detail

Here's the change, and it takes five minutes. Open the text or email you send when you ask for a review. Add one question to it, in place of the generic please-leave-us-a-review line you have now. A question with a real answer, like: what is one thing we did that made the appointment easier?

A dental office can ask what part of the visit the patient was most worried about beforehand. An HVAC company can ask what broke and how long the wait was. The point is the same in both. Ask something specific and people write you a sentence you can use.

You won't get a good answer every time, and that's fine. Two or three a month changes what the page looks like by the end of a quarter, and it keeps the recent stuff coming. That second part matters more than it sounds. In the same survey, 74% of consumers said they only count reviews written in the last three months. Anything older reads as history to the person on your page, and an assistant writing a summary is working off that same dated material.

A customer who writes that the tech called ahead and had the part on the truck has handed a machine a full sentence about how you run the job. That sentence can come back out of ChatGPT almost word for word when the next person asks who to call.

Then answer every review, and put a specific in the answer. Your replies sit on the same page and get read along with everything else, which means they get summarized too. Naming the job you finished gives the assistant one more real sentence to work from.

Do the two things together and you get a page that works on both audiences. Someone scrolling it sees several people describe the same kind of work. So does the model writing the summary, and it quotes them. Build the question into the template once and the page keeps refilling itself without you thinking about it again.

All this takes is a line of text in a template you already send, and 30 seconds on each reply. That's the whole job. As of 2026, it's also the part of your business an AI is most likely to read before it says your name to somebody.

The reviews you already have are doing this work right now, one way or another. Go read your last ten as if you'd never heard of your business. If they all say some version of the same empty phrase, you know what a machine has to work with.