A 3.3-star car wash just won the AI answer. Annie Jackson of GatherUp asked Google for a no-touch car wash that fits an SUV in Norfolk, Virginia, and Google returned exactly one business / clearance height and 24/7 hours answered inline, above the star rating. The rating was 3.3. It did not matter. The business matched the question, and the question is what AI answers now.
That one search result is the whole shift in miniature: AI tools assemble a description of your business from reviews, listings, and public web mentions, then repeat that description to customers who never reach your website. Here is what feeds that answer, what quietly stopped mattering, and what to do about both.
Customers stopped searching. They started asking.
The query was not "car wash near me." It was a full sentence with constraints in it: no-touch, SUV, Norfolk. Customers now ask complete questions and accept the summarized answer. In consumer data GatherUp collected in fall 2025, 55% of consumers had consulted Google or Bing AI summaries, 48% had asked ChatGPT about a local business, and 31% had asked more than once.
And the machines remember who is asking. As GatherUp's Jason Wertham put it, even the time of day you run a query in Google Maps can change which businesses come back. An assistant that knows you drive an SUV or own a large dog applies that context to every future local question, whether you restate it or not. The answer your customer sees is personalized, assembled, and increasingly final / most people never click past it.
"Google answered my questions, but this business is actually showing up as a 3.3 star," Jackson said. "It's surfaced the context of my query above the star rating."
Your reviews are invisible to AI / until you move them
Here is the part almost nobody has internalized. Google, Yelp, and the other major directories block LLM crawlers from reading the review content on your business profile. The reviews sitting on your Google listing still power your local rankings and convert the humans who read them there. But ChatGPT and its peers cannot see them, which is why AI answers do not cite specific reviews from those platforms.
The same reviews become readable the moment you republish them. Post them to your public social channels, embed them in a review widget on your own site, and / in Wertham's words / "now they're fair game for the LLM tools to be pulling in."
That single fact decides which questions you can win. When a customer asks an AI for the "popular" or "highly reviewed" option, the model goes looking for review text it can actually access. Review content locked inside a directory contributes nothing to that answer. If you are relying on the review platforms to carry your reputation into AI results, Wertham was blunt: it is not going to be enough.
The quiet demotion of the star rating
None of the AI answers in GatherUp's audit examples cited an average star rating. Every one cited review content. The consumer data points the same direction: 45% of users prioritize review recency over the star rating, 60% trust detailed written reviews over rating-only reviews, and 70% want the review request within 72 hours of the transaction.
Consumers already behave this way on their own / they routinely flip Google's review sort from "most relevant" to "newest," because the most recent review is the best predictor of the experience they are about to have. A gaudy average built on years-old reviews carries less weight than a current, steady stream of detailed ones.
"I'd rather go to a business with 1,000 reviews and a 3.9 or 4.2 than 30 reviews and a 5.0," Wertham said. So would the machines.
The slot machine problem
One more adjustment to make: stop treating any single AI answer as the truth about your visibility. Jackson cited SparkToro research in which different people asked LLMs the same question across devices and accounts, and the results never came back in the same order. "Asking AI a question is kind of like a slot machine," she said.
That kills "position" as the metric. What predicts whether you appear at all is citation breadth / how many crawlable sources feed the answer with your name in them. Your brand can miss one device's answer entirely and lead the next. The response is not to chase a ranking; it is to widen the base of machine-readable material the answer is assembled from.
One warning on how not to widen it: Google updated its guidance on AI features this month and now actively detects low-value AI-generated content. Generic AI blog posts and scraped-together FAQ pages have moved from "ignored" to "penalized." Feeding the machines slop now costs you.
The playbook
This is the same discipline we laid out in The State of Google Search, applied to the local layer: own assets the machines can read, and keep your own scoreboard.
Audit what AI already says about you. Ask ChatGPT and Google's AI surfaces about your business the way a customer would / full questions, real constraints. Run the prompts in incognito or temporary-chat mode so your own stored context stops flattering the results, and re-run them monthly so you are measuring movement, not a one-time slot-machine pull.
Fix the boring facts first. Consistent name, hours, phone, and services across every platform you are on. Small facts move into AI answers fast; positioning takes two weeks to a month or longer. GatherUp told the story of a restaurant whose Facebook page listed the owner's personal cell number / he never knew why the calls kept coming. That is the caliber of error these systems repeat verbatim.
Republish your reviews where machines can read them. Social posts, an on-site review section or widget, with the business reply carried alongside. This is the single highest-leverage move on the list, because it is the only way your review content enters the answer at all.
Trade vanity rating for velocity. Ask for the review within 72 hours of the transaction, respond within 72 hours of receiving it, and prize detailed written reviews over rating-only taps. Recency and volume are what hold relevancy over time / and policy-violating reviews stay disputable at any age, so defend the rating you actually earned.
Announce changes on your own site first. A new offering has to appear on your own channels before anywhere else, because your reviews will not announce it for you and the AI cannot cite what nobody wrote down.
Our position
The star rating was a scoreboard you could watch. The AI answer is a composite you have to feed. It is assembled from whatever crawlable, current, specific material exists about your business / and if that material is thin, the machines will happily hand your customer to the 3.3-star competitor who matched the question.
This is not a reputation problem. It is an infrastructure problem, and infrastructure is buildable. If you do not know what ChatGPT says about your business today / or whether your thousand hard-earned reviews are sitting somewhere no machine can read them / that is the first conversation to have.