For close to fifteen years, the geometry of local search was a comfortable certainty. A customer types dentist near me or plumber in Catania and Google produces a small map with three businesses arranged neatly below it. The famous three-pack. Anyone in the SEO world knew that if you could get into one of those three positions, the phone would ring. If you could not, you would survive on the organic results below, which carried perhaps a third of the click traffic of the pack.
This three-pack was, in a way, the heart of local SEO. The Google Business Profile optimisation industry — and there is one, a substantial one — exists almost entirely to chase those three spots. The reviews, the photos, the categories, the proximity tuning. All of it was designed for a three-result world.
And then, quietly, over the second half of 2025 and into the spring of 2026, that world began to change.
Not for every query. Not in every city. But for an increasing fraction of high-intent local searches, the three-pack is no longer what Google shows. Instead, the AI Overview at the top of the page features a single recommendation — one business, named, with a brief explanation of why — followed by a short list of two or three alternatives in much smaller type. The traditional map pack, when it appears at all, sits further down the page and gets perhaps a third of the engagement it used to.
I have been watching this trend across the roughly two hundred local-business sites we audit on a rolling basis, and I want to share what I am seeing, because I think the implications are not yet widely understood — and the action items, for a small business in a competitive city, are urgent.
What is actually changing
The change appears to be driven by Google's Search Generative Experience evolving into AI Mode for an ever-larger share of queries — including the local queries that used to trigger the three-pack reliably. When AI Mode handles a local query, it tends to produce a confident, singular recommendation rather than an enumerated list. The model has been trained to give an answer, not to give options.
Behind the scenes, the answer — the one business that gets the named recommendation — is selected from a much smaller candidate set than the three-pack used to be drawn from. Where the three-pack was a fairly mechanical ranking exercise based on proximity, prominence, and relevance, the singular recommendation appears to require a kind of confidence threshold the model only reaches when one business is, in its judgement, substantially better-signalled than its competitors on the dimensions the query implies.
This is, naturally, a different game. In the three-pack world, being in the top five or six was good enough most of the time, because the proximity and personalisation factors meant any of those candidates might cycle into one of the three slots on a given search. In the singular-recommendation world, being in the top five or six is essentially the same as being invisible. Only the named recommendation is read.
Who is winning the singular slot
When you look at the businesses that are now consistently winning the named recommendation in a given category and city, three patterns emerge with striking regularity.
The first pattern is review depth, not just review count. Businesses with two hundred reviews that include detailed, specific, recent text are outperforming businesses with eight hundred shorter reviews. The AI model appears to read the reviews — not just count them — and to extract specific descriptors it can use in its recommendation. The detail in the review becomes, in a way, the recommendation's evidence.
The second pattern is category-specific photographic depth. Not the number of photos, but the coverage. A restaurant that has photos of the dining room, the kitchen, the menu, the front door at night, the patio in summer, and the chef at work tends to win. A restaurant with thirty photos all of the food, no matter how appealing, tends to lose. The model uses the photographic completeness as a confidence signal that this is, in fact, a real and well-documented business.
The third pattern is cross-platform consistency. The business that wins the named recommendation almost always has its name, address, phone, and hours appearing identically across Google Business Profile, Apple Business Connect, Yelp, the official website's contact page, and at least one local directory of standing. Inconsistencies — even minor ones like an old phone number on one platform — appear to suppress the confidence threshold the model needs to commit to a single recommendation.
There is a fourth pattern, which I mention with some caution because the data is thinner. Businesses that have been mentioned in genuine local press — a city paper, a regional magazine, a respected blog — within the past eighteen months appear to enjoy a noticeable lift. Whether the model is reading those mentions directly or whether they correlate with other unmeasured authority signals, I cannot say with confidence. But the pattern is too consistent to ignore.
What this means for everyone else
If you are not currently winning the singular recommendation in your category and city, you are in a position that is structurally weaker than the second or third spot in the old three-pack. The traffic implications are real. Anecdotally — and I want to stress that this is anecdotal, because the dashboards do not yet measure this cleanly — businesses that used to sit comfortably in the number-two or number-three slot of the three-pack and have now slipped out of the singular recommendation are reporting click and call declines in the range of forty to sixty percent on the affected queries.
The instinct, naturally, is to throw more effort at the traditional levers. More reviews. More photos. More Google posts. This is not wrong, exactly, but it is incomplete. The traditional levers were calibrated for a three-pack world where being broadly competitive was sufficient. The singular-recommendation world requires being specifically excellent on the dimensions the AI is reading.
The question I would ask first, if you are in this position, is which competitor in your city is winning the named recommendation. Not as an exercise in envy, but as a study. Read their reviews. Note the descriptors. Look at their photo set. Check their consistency across platforms. The pattern that put them in the spotlight will, almost always, show itself within thirty minutes of careful examination.
Then, and this is the harder part, ask yourself whether your business can credibly claim a different angle the AI could anchor on. Not a generic angle — we are family-owned and we have been serving the community since 1987 are not anchors any more; every competitor says them. A specific angle. We are the only English-speaking dental practice on this side of the river. We are the only veterinary clinic in the city with weekend emergency hours. We are the only barber who still does straight-razor shaves on appointment.
The specific angle is what the AI needs in order to reach the confidence threshold for a different singular recommendation, in a different sub-context. The model, in many cities, is now producing different singular recommendations for slightly different phrasings of the same underlying intent. Best dentist in Catania and best English-speaking dentist in Catania will, for example, produce different named recommendations. The angle creates the sub-context in which you can win.
The defensive moves
Beyond the offensive work of finding your angle, there are defensive moves that protect what you already have.
Audit your cross-platform consistency this week. Not in a month — this week. Use a free tool, or simply do it by hand: search your business name on Google, Apple Maps, Yelp, Bing Places, and the two or three local directories that dominate in your city. Any inconsistency you find, fix immediately. If the model loses confidence in which business you are, it will not recommend you, no matter how strong your other signals.
Refresh your photographic coverage. Not by adding more food photos or more interior shots — by adding the photos that are missing from the coverage. The exterior at night. The car park. The waiting area. The seasonal decoration. The thing a customer would want to verify before walking through the door. The model uses photographic completeness as a confidence proxy, and most businesses are weakest in the unglamorous photos.
Solicit longer reviews. Not more reviews — longer ones. A polite note to your three best customers each month, asking them to share what specifically made their experience good, will produce a slow accumulation of detailed review text the AI can extract from. This is, in my opinion, the single most undervalued intervention in local SEO right now.
A word about the dashboards
The shrinking of the local pack is, predictably, not yet captured well by the dashboards. Most local-SEO tools still report your position in the three-pack even when the three-pack is no longer being shown for the query they are measuring. The position number they give you is therefore, in those cases, fictional. You are not at position two in a three-pack that does not exist; you are absent from a singular recommendation that does.
The practical implication is that you cannot trust the dashboards alone. You must, again, do some of the work manually — run your top ten local queries from a clean profile in the city you serve, and note what actually appears at the top of the result. The number you compute this way is the only number that reflects what your customers are actually seeing.
The closing thought
There is a temptation, when describing a structural change like this one, to frame it as a crisis. I want to resist that framing.
The truth is, the businesses that adapt to the singular-recommendation world will benefit disproportionately. When only one business in a category gets the named recommendation, that business captures a far larger share of the local intent than any single business used to capture in the three-pack era. The winners win bigger. The losers lose more. The middle ground is shrinking.
For a small business with the discipline to find its specific angle, do the unglamorous consistency work, and play the long game on review depth and photographic coverage, this is not a crisis. It is an opportunity. The window in which the rules are still being figured out is the window in which the most movement is possible. Six months from now, the patterns will be more entrenched.
So do the work now. Read your competitor's reviews. Take the boring photos. Fix the inconsistent phone number. Find your angle. The pack is shrinking. The spotlight remains.