The fastest shift in local discovery in a decade happened over twelve months. Most clinics have not changed a thing about what they publish.
Key takeaways
- Use of AI assistants for local business recommendations went from 6% to 45% in a single year, according to BrightLocal's 2026 survey of consumers.
- Assistants answer with one or two names rather than ten links, so the gap between being recommended and being invisible is wider than a ranking gap.
- Assistants describe businesses using what those businesses have published. A clinic with a thin website gives them nothing to say.
Local search does not usually move quickly. Google adjusts, review counts creep up, a competitor redesigns their website. The ground shifts by inches.
Then this happened. In BrightLocal's 2026 consumer survey, the share of people who said they use an AI assistant to find a local business went from 6% to 45% in twelve months. That makes assistants the third most-used channel for local business discovery, behind only Google and Facebook. They passed Yelp. They passed TripAdvisor. They did it in a year.
In the same survey, Google's share as the place people go to check a business fell from 83% to 71%. Not a collapse — but the first real dent in a long time, and it went somewhere specific.
Why this is different from a ranking change
A search results page is a shortlist. Ten links, a map with three clinics, and a person who decides. Being fourth is worse than being first, but fourth is still on the page. Someone scrolling might still click you.
An assistant does not produce a shortlist. Someone types "I need a physiotherapist for knee pain near me" and gets back a paragraph with one or two names in it. There is no fourth position in a paragraph. You are either in the answer or you are not part of the conversation the patient is having.
There is no fourth position in a paragraph. You are in the answer, or you are not in the conversation.
This matters more because people appear to be acting on those answers. In the same research, 42% of consumers said they trust an assistant's recommendation about as much as they trust written reviews, and among people who actively use assistants, 63% said they trust the local recommendations they get. Adoption is highest among 30 to 44 year olds, at 64% — which, for most clinics, is a description of their best patients.
What an assistant is actually doing when it names a clinic
It helps to be unromantic about this. An assistant asked to recommend a chiropractor has no way of knowing who is good at chiropractic. It has no clinical judgment and no experience of being treated. What it has is text: what businesses have published, what directories list, what reviewers wrote.
From that, it has to produce a sentence that sounds confident. So it reaches for the business it can describe. A clinic with a page explaining how it assesses lower back pain in a first appointment hands the assistant a ready-made sentence. A clinic with a homepage, a contact form, and a stock photo of a reception desk does not.
This is why the results can look unfair. A clinic with a 4.6 rating and a deep website gets named ahead of a clinic with a 4.9 and three pages. The assistant is not judging care quality. It is naming the business it has something to say about.
Reviews still matter. They just do a different job now.
None of this makes reviews less valuable. They are still what converts a patient who is deciding between two names, and they still feed how a business is described.
But reviews tell an assistant that people were happy. They do not tell it what you treat, how you work, what a first visit involves, or who you are the right clinic for. Two clinics with identical ratings are indistinguishable to an assistant unless one of them has written something down.
The practical version: reviews are how you get chosen. Content is how you get mentioned in the first place. You need both, and most clinics have been investing in one.
What to do about it, in order
- Write the questions down. Every question a patient asks before booking is a page you do not have. Your front desk fields them all week.
- Be specific about place. Almost every one of these prompts contains a location. Name your city and your neighbourhood in plain language rather than leaving it to a footer address.
- Cover the ground rather than polishing one page. Depth of coverage is what gives an assistant multiple routes to your name. One excellent page is one route.
- Keep going. This is a compounding asset, and it is the part almost everyone underestimates, because it asks for a Tuesday every week rather than a weekend once.
A test you can run in five minutes
Before changing anything, find out where you stand. Open an assistant and ask it the way a patient would — not "best dentist in London Ontario", which is a marketer's phrasing, but the sentence a real person types. "My tooth is sensitive to cold and I have not been to a dentist in three years, who should I see in London Ontario."
Ask it four or five different ways, covering your main services. Write down which businesses get named and, more usefully, what the assistant says about them.
Three outcomes, each meaning something different:
- You are not mentioned. The common result. It usually reflects how little you have published rather than anything about your clinic.
- You are mentioned with nothing specific attached — named in a list, no reason given. You are being found through directory data, not through anything you wrote. Fragile position.
- You are mentioned with a reason. Note the reason. It will almost always trace back to a specific page on your website, and that tells you what kind of page to write more of.
Run the same prompts against your two closest competitors' names in mind. If one of them keeps appearing with a specific reason attached, go and look at what they have published. It is usually not a mystery.
The part that is easy to get wrong
Two failure modes are common enough to name.
The first is writing for the assistant instead of the patient. Pages built out of question headings with thin answers underneath, stuffed with the phrases someone thinks a machine wants. These read badly to humans, and systems have become good at recognising content assembled to be retrieved rather than to be useful. The pages that get cited are pages that would have been worth writing anyway.
The second is treating this as a project with an end. A burst of twelve pages in September, then nothing. The clinics that hold these positions are the ones still publishing in month nine, and it is almost never because they are better writers.
The uncomfortable part of the 6% to 45% number is not that it happened. It is how little most clinic websites changed while it did. The clinics that get named in a year's time are, for the most part, going to be the ones that started publishing this year.