What is actually changing in search?
Something real is happening, and pretending otherwise would make the rest of this page worthless. Start with the best-documented finding.
Pew Research Center tracked the browsing of 900 US adults during March 2025 and found that when a Google results page carried an AI summary, users clicked a traditional search result in 8% of visits, against 15% when no summary appeared. They clicked a link inside the summary itself in just 1% of visits. Read the method before you panic: 900 people, one month, Google only, and Pew re-collected the results pages weeks after the searches, so the AI-summary flag is a reconstruction. Nothing in that dataset is local-business specific. It measures general web search behavior.
The wider zero-click trend points the same way. SparkToro, using Similarweb panel data for January to April 2026, put US Google searches ending without a click at 68.01%, up from 60.45% in 2024. Rand Fishkin is candid that the long trend line stitches together different data providers and that the mobile split is estimated, so treat the direction as solid and the decimal places as decoration.
Google's own account is different. In August 2025, Search head Liz Reid wrote that "total organic click volume from Google Search to websites has been relatively stable year-over-year" and that click quality has risen. Weigh that fairly. Google published no data, no date range and no method behind it, while Pew published all three. Also note what Google actually claimed: aggregate stability, which is perfectly compatible with your particular site losing traffic while others gain. Google says as much in the same post.
On the other side of the ledger, AI assistants send very little traffic so far. Ahrefs, studying around 82,000 sites in June 2025, reported that Google sent roughly 210 times more traffic than the three main AI platforms combined, with AI accounting for about 0.25% of the average site's traffic. Vendor data, self-selected sample. Ahrefs also flags two caveats that cut in opposite directions: some AI referrals get logged as direct traffic, so the number is a floor, and in that same period the average site's search traffic fell about 21%.
So the fair summary is narrow. Informational queries leak clicks. AI does not yet replace those clicks with its own referrals. And no published, verifiable dataset breaks any of this out for local service queries, which is exactly the part that matters to you. Anyone quoting a precise percentage of "local searches lost to AI" is quoting something that does not exist in checkable form.
What has not changed for a local business?
Four things, and together they carry most of the revenue.
The local pack still decides the call. Google's own guidance still describes local results as being "mainly based on relevance, distance, and popularity", and the same page names reviews and links to your business as inputs. That is Google saying, on the record, that filling out your profile and earning reviews moves your ranking. No AI answer replaces the map, the hours, the directions button or the tap-to-call.
The intent is different. A person searching "emergency dentist open now" is not researching, they are acting in the next twenty minutes. Zero-click statistics count the searches where nobody needed a website. Booking an appointment is not one of those. This is why Fishkin, the loudest voice in the zero-click argument, still writes in that same study that "there are still some categories that benefit from SEO, including branded searches, local businesses, and high-intent transactional or tactical queries".
Search is still where the journey starts. BrightLocal's 2026 consumer survey (vendor-reported, 1,227 US consumers, self-reported recall) found 52% began their most recent local business search on Google Search and another 9% on Google Maps, with 71% touching Google Search somewhere in the journey. The same firm reports 45% of consumers now using AI tools for local recommendations, up from 6% a year earlier, with 88% saying they fact-check what the AI tells them. Both numbers are survey recall from an interested vendor, so hold them loosely. The pattern they describe, AI as a fast-growing first opinion that people then verify on Google and Maps, matches what we see in our own tracking.
The machines eat the same food. This is the load-bearing point of the whole page. OpenAI runs a dedicated search crawler, and its documentation says "OAI-SearchBot is used to surface websites in search results in ChatGPT's search features". Google is blunter still: for AI Overviews and AI Mode there are "no additional requirements... nor other special optimizations necessary", and the same page tells you to keep your Business Profile up to date. There is no second internet for AI. An assistant recommending a gym is reading indexed pages, listings and reviews, which is to say the output of ordinary local SEO. Do the base work badly and you are invisible to both surfaces at once.
So where do you lose if you ignore AI entirely?
In one specific place: recommendation queries, where the answer names one or two businesses instead of listing ten.
In our own tracking (July 2026) we froze a panel of 15 buying prompts for local health, fitness and aesthetics services and ran each of them multiple times, 90 runs in total. Two things stood out. A Google AI Overview fired in 42 of 45 runs, so the synthesized paragraph is now the default first thing a buyer reads. And the answers named a handful of providers, never a list of ten. That is the structural change. In the ten blue links, being fourth was a bad day. In an AI answer, being fourth is being absent. If you run a gym and that is exactly what you are seeing, we walked through the usual reasons a business gets left out of that paragraph in why your gym doesn't show up in Google AI Overviews.
Scarcity is also opportunity, because almost nobody has claimed those seats. SOCi's 2026 Local Visibility Index looked at more than 350,000 locations across 2,751 multi-location brands and found ChatGPT recommending 1.2% of those locations, against 35.9% appearing in Google's local 3-pack (vendor-reported, directional). Two things that number is not: it is not a measure of independent local businesses, and it is not your ceiling. SOCi's own reading is that "LLMs often favor single-location businesses" because their data, reputation and activity signals are simpler and more unified. Read the gap as a map, then: the volume today is in the pack, the empty room is in the AI answer, and a single-location business is better placed to walk into it than the chains that study measured.
The work that wins those seats is not exotic. The academic paper that named the field (Aggarwal et al., KDD 2024) measured visibility gains of up to 40% in generative answers from adding quotable statements, citations and statistics to source pages. Academic benchmark, older GPT-3.5-class engines, so directional rather than a promise. But the direction is the whole argument: clear, quotable, corroborated pages are what a model repeats, and those are the same pages that convert a human reader. You are not choosing between two disciplines. You are choosing whether your service pages are worth quoting.
Which queries are worth what, in 2026?
The useful move is to stop arguing about "SEO versus AI" and sort your queries by type. Different types are won on different surfaces, and only one of them genuinely lost value.
| Query type | What wins it today | Where you should invest |
|---|---|---|
| "dentist near me", "gym in [suburb]" | The map pack. Proximity, category, reviews and profile completeness. | Business Profile filled to 100%, exact NAP everywhere, steady reviews. Highest return per hour of anything on this list. |
| "emergency dentist open now" | The map pack plus live hours and a working tap-to-call. | Accurate hours including holidays, phone answered in business hours, a booking path that works on a phone. |
| "best gym in [city]", "top med spa for [treatment]" | AI answers and third-party roundups, alongside the pack. | Corroboration off your site: legitimate local roundups, directories, community threads, review text that names your services. |
| "how much does [treatment] cost" | An AI Overview or assistant answer. Fewer clicks reach anyone. | One honest pricing page with real local ranges, answer-first. Expect fewer visits, and value being the source that gets quoted. |
| "does [treatment] hurt", "is [treatment] safe" | Synthesis across several sources. Rarely a single click. | Short, quotable, dated, author-attributed answers. Worth writing once, not worth a monthly content mill. |
| "[your business name]" | Your own site and profile. It was never contested. | Almost nothing. If an agency is charging you to rank for your own name, that line item is filler. |
Surface assignments reflect our own run logs from the July 2026 prompt panel plus the sources cited above. Engines change retrieval behavior without notice, which is why the last section is about measuring instead of assuming.
How do you decide with your numbers instead of somebody's average?
Every statistic on this page is somebody else's business. Yours is knowable in about an hour, and it beats any benchmark. Pull four things.
- Search Console, brand versus non-brand. Last 16 months, split queries containing your business name from those that do not. Brand clicks tell you how well your existing demand converts. Non-brand clicks are the ones SEO is actually buying. If non-brand clicks are near zero, you are not losing to AI, you never had the traffic.
- Impressions against clicks on non-brand. If impressions hold or rise while clicks fall, that is the AI compression pattern, and it will show up first on informational queries. If both fall together, it is a ranking problem, which is an ordinary SEO problem with an ordinary fix.
- Calls and bookings, not sessions. Business Profile calls, form fills, and booked appointments, monthly, for the last year. A local business can lose website sessions and gain patients in the same quarter. Traffic is a proxy. Appointments are the business.
- Ask every new customer how they found you. One question at intake, logged in a spreadsheet. Thirty answers beat every study cited on this page, because they are about your city and your business. Watch for "ChatGPT said" appearing in that column, and watch how often it does not.
Then do the arithmetic with your own variables. We are not going to invent an industry average lifetime value for you, because those numbers vary so much by treatment mix and city that the average is meaningless.
Monthly spend ÷ (your average first-year value of a new customer × your close rate on inbound leads) = new leads per month needed to break even
Fill in your own two numbers. A practice where a new patient is worth $600 in year one and that closes half its inbound calls needs roughly five extra leads a month to break even on $1,500. A med spa whose average client is worth $3,000 needs one. Same invoice, completely different decision, which is why nobody can answer this for you from a blog post.
Then compare against the alternative uses of the same money: paid ads, a receptionist who answers the phone during lunch, or a better booking flow. If the phone rings and nobody picks up, fix that first. It is cheaper and it converts demand you already paid for.
When is the honest answer "do not hire anyone yet"?
Sometimes it is, and saying so costs us business. The test is not how good your search presence looks today. It is whether the free work is finished, and whether finishing it was enough. A retainer only earns its keep in the band between those two points, and there are two quite different ways to sit outside it.
Outside it from below: the free work is still undone. Nobody should be charging you a monthly fee yet.
- You have no website worth optimizing. Retainers cannot fix a site that does not say what you do or take a booking. That is a build problem, not a search problem.
- Referrals carry you and you have never checked whether anyone searches for your service in your city. Check first. It is free.
Outside it from above: the free work is done and it is still carrying you. The highest-value work in local search is unpaid, and when it is genuinely finished there may be nothing left worth two thousand dollars a month.
- Your book is full. More demand you cannot serve is not an asset. Fix capacity or raise prices first.
- You are the only real option in a small market. Thin competition means the pack ranks you almost by default. A complete profile and steady reviews may be the entire job.
- Your profile and reviews are already in good shape and stay that way without you thinking about it, and your name comes back when you test your own money prompts in ChatGPT and Google.
Hiring makes sense in between: the free work is done, and it stopped being enough. The two hours a week version is real and it is written down (profile, reviews, hours, one solid page per money service), and our do it yourself or hire an agency guide is the task-by-task breakdown. It also carries the full list of questions to put to any agency in writing before you sign, which is the part worth reading twice if you decide you are in the middle band.
What we would actually do first
If you take one thing from this page, take the order of operations. It has not changed as much as the headlines suggest.
- Business Profile to 100%, exact and consistent name, address and phone everywhere. Free, and it feeds both surfaces.
- A review habit that runs every week. Reviews are a named ranking factor and they are also the text an AI reads to describe you.
- One quotable page per money service, answer-first, with real local price ranges, dated and attributed to a named person.
- Check the machines can read you: server-rendered content, no accidental blocks, crawler access decided on purpose rather than by a security plugin's default.
- Freeze a panel of the prompts your buyers actually use, run each several times, and re-measure monthly against that baseline. Our guide on how to track whether ChatGPT recommends your business walks through it, and you can do it without paying anyone.
Notice that steps one through four are what a competent local SEO has always done. Step five is the only genuinely new one, and it is measurement, not a new channel.
If you would rather hand it over, this is what we do for dental practices and for gyms and studios. The deep audit is $1,500 and credits 100% toward your first month, and retainers run $1,500, $2,500 and $5,000 a month on the pricing page. On Growth and above the risk sits with the measurement: if you are not cited inside 90 days, judged against the panel we froze at the start, we carry on at no charge. We also sign only one business per city, since a seat that an answer gives to one or two providers cannot honestly be sold twice. Still deciding between a specialist and the SEO firm you already pay? GEO agency vs SEO agency covers what genuinely differs between them.