Something quietly strange has been happening to local businesses for about two years now. The rankings hold. The Google Business Profile still sits in the top three for the searches that actually matter. Every dashboard is green. And the phone rings a little less every quarter.
Joy Hawkins of Sterling Sky put numbers to it. She asked the team at Jepto to compile two years of click data from 179 Google Business Profiles across 34 law firms, and the pattern was unmistakable: clicks to call were sliding steadily downward, even for profiles that never lost a ranking position. Website clicks, meanwhile, held steady. That asymmetry is the entire story in one chart. Website icons appear on desktop. Call buttons appear on mobile. The decline wasn’t happening to businesses — it was happening to a button.
This is the defining condition of local SEO 2026: visibility and ranking have come apart. You can hold position two and still be functionally invisible to the person standing on the sidewalk outside your building. The rank tracker will keep telling you everything is fine, right up until you’re staring at a quiet reception desk wondering what changed.
Nothing changed about your business. What changed is what a ranking now buys you.
The three-pack is becoming a one-pack
For roughly a decade, local search ran on a simple, legible bargain. Rank in the local pack, appear alongside two competitors, and get a click, a call, or a direction request. Three businesses got the shortlist. Everyone understood the game.
Google is now rolling out AI-powered local packs, and they break that bargain in three specific ways. Hawkins’s team, tracking with Places Scout, documented them in detail:
They show fewer businesses. AI local packs typically surface one or two businesses instead of three. Across the reports Places Scout analyzed, AI local packs featured 5,943 unique businesses where traditional three-packs featured 18,330 — meaning AI packs surface only about 32% as many businesses as the format they’re replacing. In 322 markets examined, 88% showed fewer unique businesses in the AI pack than in the classic one.
They drop the call button. Many AI local pack formats remove click-to-call entirely. For a clinic, a studio, or a trades business, that is not a cosmetic change — it removes the single highest-intent action a mobile searcher can take.
They feature different businesses. This is the part most owners miss. The AI pack doesn’t simply truncate the three-pack to its top two. It frequently surfaces businesses that weren’t in the three-pack at all. Your position in one is not a claim on the other.
Layer onto that a second, quieter change: Google has been replacing call buttons with images in several verticals. Sterling Sky tracked the swap in the handyman category between October and November 2025, and had documented the same move years earlier in dentistry, where it produced a measurable drop in call-button clicks.
Your ranking didn’t fall. The number of chairs in the room did. Local search went from a shortlist of three to a shortlist of one — and stopped handing out the phone.
Two-thirds of the businesses that used to make the cut simply don’t appear. That is not an optimization problem. It’s a structural one.
Your rank tracker is telling you a comfortable lie
Here is the operational trap. Most rank-tracking software still can’t see AI local packs. It reports on the traditional three-pack, finds you sitting at position two, and renders a reassuring green number.
Meanwhile the customer on their phone never saw that three-pack.
Sterling Sky calls this rank blindness, and it is the reason so many owners describe the same disorienting experience: the SEO report looks great, the lead volume doesn’t. If your marketing reporting has not changed since 2023, it is measuring a version of Google that your customers are increasingly not using.
The AI packs are still early — showing on roughly 7–8% of tracked keywords as of the most recent counts, concentrated on mobile and observed first in the United States. But the direction of travel is not ambiguous, and “small share of keywords” is cold comfort when those keywords are the ones with buying intent.
What’s driving it: the answer layer above the map
Above the local pack now sits a generative answer layer, and it is not a niche feature. When Whitespark studied AI Overviews in local search, it found them appearing on an average of 68% of local queries. A GatherUp survey of more than 1,000 US consumers found that 48% had already used a conversational AI tool while looking for information about a local business.
Google AI Mode compounds this. It’s conversational and multi-turn: a searcher can ask for “physio near Port Credit,” then follow with “one that takes direct billing,” then “and has evening hours,” without ever restarting the search. Under the hood, Google runs what it calls query fan-out — generating a set of related sub-queries and assembling an answer from many sources at once. Your business isn’t competing for one keyword anymore. It’s competing to be a credible answer to a cloud of questions you never bid on.
Two consequences follow, and both are uncomfortable.
First, top-ten organic rankings do not guarantee inclusion. A study by Ziptie, discussed by Michael King and Rand Fishkin and cited in Whitespark’s AI Mode guide, found that ranking in Google’s top ten gives a page roughly a 25% chance of appearing in an AI Overview. Ranking is now a qualifying round, not a result.
Second, results vary by person. Two people running the same query from the same room can get materially different AI Mode outputs. There is currently no practical way to know what any given prospect is being shown about you.
And the machine gets things wrong. Whitespark documented AI Mode confidently telling a user that a restaurant was dog-friendly — a conclusion it had assembled from the policy of a business that previously occupied the address and an Instagram video for an entirely different restaurant. For a gym, that’s an annoyed prospect. For a regulated health clinic quoting the wrong service, hours, or practitioner credentials, it’s closer to a liability. The same GatherUp research found 67% of consumers aren’t rigorously fact-checking the sources AI cites.
The counterintuitive good news
If the story ended there it would just be doom. It doesn’t.
Seer Interactive analyzed roughly 10,000 informational-intent keywords ranking in the top twenty and found exactly what you’d expect: when AI Overviews appeared, organic click-through rate fell from 1.41% to 0.64% year over year. Brutal.
But the same analysis found something else. When a brand appeared inside the AI Overview, its click-through rate went up — organic from 0.74% to 1.02%, and paid from 7.89% all the way to 11%. And on queries where no AI Overview appeared at all, organic CTR actually rose.
Read those two findings together and the strategy writes itself. The AI layer isn’t destroying clicks evenly. It’s concentrating them. It is taking traffic away from everyone who isn’t named and handing a premium to everyone who is.
Being cited is the new ranking. The businesses named inside the answer are getting more than they used to. Everyone else is splitting a smaller pot.
That’s a harder game than the old one, but it’s a game with better odds for a well-run local business than for a content farm — because the things that earn a citation are things a real business genuinely has and a spam operation can’t fake.
What Google actually said — and what it debunked
On May 15, 2026, Google published its first consolidated official guidance on this, Optimizing your website for generative AI features on Google Search. Anyone currently being pitched an “AI search optimization” retainer should read it before signing anything.
The headline: from Google’s perspective, optimizing for generative AI search is still SEO, because these features run on the same core ranking and quality systems as everything else. The guidance also includes a mythbusting section, and Google’s own documentation on AI features states it plainly — you don’t need to create new machine-readable files, AI text files, or special markup, and there’s no particular schema.org structured data required to appear.
So the llms.txt file someone wants to sell you does not appear in Google’s guidance as something Google uses. Neither does chunking your content into AI-friendly blocks, nor maintaining Markdown mirrors of your pages. That doesn’t make every emerging tactic worthless across every platform — but it does mean that if a proposal leads with acronyms instead of fundamentals, you’re being sold novelty.
While we’re puncturing things: ChatGPT is probably not the one taking your traffic. Hawkins reported that for one of Sterling Sky’s largest multi-location clients, ChatGPT referrals grew from 0.1% of Google’s volume to 2% in a year — real growth, still a rounding error, and still only about 22% of what Bing sends. Google’s own AI layer is doing the damage.
The local 2026 local playbook
None of what follows is exotic. That’s the point — Google said the fundamentals still govern, and the fundamentals are where most local businesses are quietly weakest.
1. Fix the entity layer before anything else. Name, address, phone, hours, categories and service descriptions must be identical everywhere they appear — your site, your Google Business Profile, directories, social profiles, booking platforms. AI systems assemble your identity from all of these as a single data stream, and every contradiction is an invitation to hallucinate. This is unglamorous, cheap, and the highest-leverage hour you will spend this quarter.
2. Treat reviews as source material, not social proof. Reviews have become one of the primary inputs generative answers draw on. Volume alone no longer carries it — recency, velocity and your response rate all matter, and the content of reviews matters most, because specific language (“direct billing,” “evening appointments,” “parking on site”) is what makes you retrievable for specific questions. Regulated professionals in Ontario should be careful here: your college’s advertising and testimonial rules constrain how you can solicit and use patient feedback, and generic agency advice will walk you into a compliance problem.
3. Earn mentions you don’t own. Generative answers lean heavily on third-party sources — local news, community pages, association directories, industry publications, sponsorship listings, partner sites. A page on your own website is one input among many. A mention on the local business association’s site is corroboration. Real community involvement is now a search strategy, which is a rare case of the honest thing also being the effective thing.
4. Answer the questions people actually ask, in their words. Start logging the questions your front desk fields every day, verbatim. Those are your content briefs. Query fan-out means Google is decomposing broad intents into narrow ones; the businesses that have already published clear, specific answers to narrow questions are the ones it finds.
5. Rebuild the conversion path that no longer starts with a call button. If the tap-to-call shortcut is disappearing, more people will arrive at your website as the first real point of contact — and they’ll arrive on a phone, mid-decision, with less patience than ever. That makes mobile speed and a frictionless booking flow revenue infrastructure, not polish. We’ve written about why website speed is a revenue problem and the conversion mistakes that quietly cost sales; both matter considerably more in 2026 than they did in 2024.
6. Change what you measure. Ranking position is no longer a proxy for anything. Track calls and bookings as primary metrics, segment AI referral sources in GA4, and use Search Console for what it can still tell you — impressions and clicks from AI features are folded into overall web search reporting. If your monthly report leads with rankings, it’s measuring the wrong decade.
7. Be honest about paid. The uncomfortable finding in Sterling Sky’s data is that local pack ads went from appearing on 1% of their tracked mobile reports in early 2025 to nearly 22% by December, and Local Services Ads climbed from around 11% to 31% in the same window — while ads kept the call buttons that organic listings are losing. We would rather tell you that than pretend organic alone still covers the ground it used to. For many local businesses the right answer in 2026 is a smaller, sharper paid budget defending high-intent queries while organic does the compounding work underneath.
The Toronto advantage nobody is talking about
Here’s the part that should change what you do this month.
AI local packs have so far been observed primarily on mobile and primarily in the United States. Canadian businesses are, for once, downstream of a rollout rather than in front of it.
That is a window, and it is closing. Everything on the list above — entity consistency, review depth, earned local mentions, documented answers to real questions, a booking flow that works on a phone — takes months to compound. It cannot be bought in a panic the week the format arrives in Mississauga or Scarborough. The businesses that will own the one or two slots in a GTA AI local pack in 2027 are the ones building the evidence base for it right now, while their competitors are still admiring a green ranking report.
The shortlist is getting shorter. The only durable strategy is to be so clearly, verifiably good at one specific thing in one specific place that no summarizer can leave you out of the answer.
Local search is not dead, and anyone telling you otherwise is selling something. But the mechanism has changed underneath a lot of businesses without their noticing, and the reporting most of them rely on is structurally incapable of showing it. If your calls are down and your rankings aren’t, you now know why — and you know it isn’t your fault, which also means it isn’t going to fix itself.
If you’d like an outside read on where your business actually stands across search, site and reputation, that’s precisely what our Digital Posture Audit is built to answer.
Frequently asked questions
What is an AI local pack?
An AI local pack is a generative, AI-assembled version of Google’s traditional local three-pack. Instead of listing three nearby businesses with standard actions, it typically surfaces one or two businesses inside an AI-generated summary, frequently without a click-to-call button, and often featuring different businesses than the ones ranking in the classic three-pack.
Is local SEO still worth doing in 2026?
Yes — arguably more than before, but the target has moved. Google’s own May 2026 guidance confirms its generative features run on the same core ranking systems as traditional search, so SEO fundamentals still determine whether you’re eligible to be surfaced at all. The difference is that ranking is now a qualifying condition rather than the outcome; the goal is to be the business named inside the answer.
Do I need an llms.txt file or special AI schema to appear in AI Overviews?
According to Google’s documentation, no. Google states that you don’t need to create machine-readable AI files, AI-specific markup, or special structured data to appear in AI Overviews or AI Mode. Some of these tactics are still debated for other AI platforms, but for Google they are not a requirement — and any proposal built primarily on them deserves scrutiny.
Why are my Google Business Profile calls down when my rankings haven’t moved?
Most often because the call button itself is disappearing from mobile results — replaced by images in some verticals and absent from many AI local pack formats — while paid placements retain it. Two years of profile data analyzed by Sterling Sky and Jepto show call clicks declining even for profiles holding their positions, with website clicks (which appear on desktop) staying flat.
How do I find out whether AI search is mentioning my business?
Start manually: take the ten questions your customers ask most often, run them through Google AI Mode and other assistants, and record whether you appear and which sources get cited. That list of cited sources is your outreach target list. Pair it with GA4 segmentation for AI referral traffic and Search Console for overall search performance. Results vary by user, so treat this as directional evidence rather than a rank report.
