Google's AI is eating the local pack: what to measure now

From one surface to three: organic, local pack and AI Overview citation

For fifteen years local SEO was explained with two boxes. Map with three listings on top, ten blue links underneath. Optimize the business profile to get into the map, optimize the site to get into the ten, done. That drawing is no longer correct. When Google decides to answer a local search with an AI-generated summary, a third box appears above the other two and it plays by its own rules. The relationship between AI Overviews and the local pack is not replacement, it is superimposition — and that difference changes what you have to measure.

Worth being precise about what we are discussing. Google rolled AI Overviews out generally in the United States in May 2024 and extended them through 2024 and 2025 to more than a hundred countries and several languages, Spanish among them. Since then a good share of informational queries — including plenty with local intent — can return a synthesized block of text with links to the sources Google used to build it. That block is not always there: Google turns it on or off depending on the query, the language, the device and the moment. That is where the measurement problem starts, because an intermittent surface cannot be watched with the same logic as a stable position.

The most useful way to think about it: you stopped having one position and started having three simultaneous states. You can be strong on all three, weak on all three, or — most commonly — strong on one and weak on another without noticing, because you never measured them apart. This table sums up what each one is, what feeds it, and what "being there" actually means.

Look at the right-hand column: the three are measured differently because the three are different things. The first two return a number that goes up or down. The third returns a yes or a no that can flip overnight. Cramming them into the same report in the same format is why so many local businesses believe "everything looks the same" while their calls quietly disappear.

There is an uncomfortable practical consequence. If your current tool only reports organic position, you have visibility on one third of the problem. If you only watch the map, you have another third. The full picture of what can be tracked continuously is on the Rankiamos features page, but the idea is tool-independent: if you do not separate the three surfaces, you will not be able to explain why your phone stopped ringing.

What the research says about AI Overviews in local search (and why to read it with tweezers)

Go looking for data on how AI is affecting local search results and you will find a handful of studies plus an enormous pile of articles quoting them secondhand. As of July 2026, the serious research comes almost entirely from Anglophone houses — BrightLocal with its annual local consumer surveys, trade publications like Search Engine Land, and the periodic visibility studies the big SEO suites publish — and it shares a set of traits worth holding in mind before you repeat any percentage in a client meeting.

I am deliberately not quoting a specific figure here. Each study measures a different thing with a different sample, the numbers contradict each other, and they age in weeks. What can be said without risk is the direction: the presence of AI summaries on locally-intended queries has grown steadily since 2024 and has not reversed. What cannot be said is how much, in your language, in your city, in your vertical. Four reasons a number measured in English does not transfer.

The operational conclusion is simple and unglamorous. Use the published studies to convince your boss or your client that measurement has to start, then discard their numbers and keep your own. A local business with twenty keywords can produce its own AI trigger rate in two weeks of daily measurement. That number, however laughably small its sample compared to a published study, describes your reality. The study describes somebody else's.

Ranking in the 3-pack does not get you into the AI Overview: different signals entirely

This is the part that draws the most pushback from an owner who has spent two years working their business profile. The reasoning is airtight inside the old model: "if Google considers me the best dentist in the area and puts me first in the map, obviously the AI will name me." It is not obvious, and the reason sits in the raw material each system consumes.

The local pack is assembled from your business profile and the signals Google associates with that physical entity: category, attributes, reviews, distance from the searcher, name and address consistency, and aggregate user behavior. It is a place-centric system. An AI Overview, by contrast, synthesizes text. It needs written, citable sources on the topic, and it will find them wherever they live: on your site if you have a page that answers the question well, but also — more often than anyone would like — in vertical directories, "best X in [city]" roundups, review portals, local news outlets and even forum threads. It is a text-centric system.

Two situations fall straight out of that mismatch. First: the business that owns the map and appears in none of the sources Google used to build the summary, because its site has three pages and nothing that answers a question. Second, and more infuriating: the business that never cracks the top 3 but does get mentioned inside the summary, because a third party's blog put it on a recommended list. Local businesses in AI Overviews do not get there through the door they used to reach the map.

One technical nuance worth holding so you do not chase ghosts. When the query is purely navigational-local — somebody searching a business by name, or phrasing something with a clear "take me there now" intent — Google tends to resolve with map and profile, no summary. When the query asks for judgment, comparison or explanation, the summary shows up far more readily. That boundary is what you need to map across your own keyword set, and it sits in a different place for every vertical.

How you technically track those citations — the manual methods, the browser extensions and the automated tools, with the real cost of each option — is a subject with enough substance to deserve its own post, and we cover it separately on the blog. What matters here is the consequence for the local business: these are two independent readings and they have to be taken separately, on the same keyword and from the same city, or the comparison is meaningless. And if you look at Google Maps and AI expecting one to explain the other, you will end up with two series that do not correlate and one wrong conclusion.

What this looks like on the ground: dentist, auto shop and lawyer + city

Let us bring this down to three businesses that exist in every city. To be clear before we start: what follows is not a study or a systematic measurement, it is a framework for classifying your own keywords. The real data comes from measuring yours. But the framework holds up, because local searches cluster reasonably well into three intent types, and each type behaves differently against the AI summary.

The pattern across all three is the same: the more a search resembles a question, the more the AI summary takes over and the further down the map falls. The more it resembles an immediate need for a location, the more the local pack rules. Most businesses have keywords of both types mixed into the same list, which is exactly why a single blended average is useless: you are averaging a keyword where the summary is stealing your screen against one where it never even appears, and the resulting number describes neither.

There is one more layer, and it is specific to anyone working Spanish-speaking markets. The 17 Spanish-speaking countries of Latin America plus Spain do not behave alike, not even for the same query translated. Vocabulary shifts — taller, mecánico, servicio automotriz — the directories with authority change country by country, and the AI rollout timing differs. An SEO covering Latin America who measures from a single city is looking at one country and extrapolating to all the others.

That is exactly why serious local measurement happens per city rather than per country. The mechanics of geolocation and what it implies for quota consumption are covered in the guide to tracking rankings across multiple cities, which is the natural companion to this post.

Doing local SEO with artificial intelligence in the picture is not about bolting one new metric onto the same old report. It is about accepting that you now hold three data series that cannot be averaged together and that do not even read at the same tempo. This is the minimum protocol I recommend to any local business that wants to know where it stands, in the order it should be executed.

1. Step 1 · Pull your real query list from Search Console

   Before deciding what to watch, look at how people find you today. The performance report, filtered by country, gives you the real queries with impressions. It is the best raw material available and it is free. One caveat: Search Console does not separate what happened inside an AI summary from what happened in a blue link, so it helps you choose keywords, not confirm citations.

2. Step 2 · Classify each keyword by intent type

   Tag each one as proximity, judgment or complex question, using the framework from the previous section. This five-minute exercise decides which surface matters for each term and stops you chasing AI citations on searches Google will never summarize.

3. Step 3 · Define the cities you actually operate in

   Not the cities you would like to sell in: the ones your customers come from today. Every city you add consumes quota, so discipline here pays for itself. A single-location business in Medellín does not need Bogotá data to know how it is doing.

4. Step 4 · Start measuring before you touch anything

   This step gets skipped every time and it is the most expensive one to skip. Keyword history in Rankiamos starts the day you add the keyword — it is not retroactive. Optimize first and measure later and you have no baseline, which turns every later conclusion into an opinion. Let it run one or two weeks before changing anything.

5. Step 5 · Read each surface on its own cadence

   Local organic moves slowly and tolerates a weekly read. The local pack moves more and deserves closer watching after any profile change or when a competitor arrives. Presence in the AI summary is the most volatile of the three: it changes day to day with nothing on your side having moved.

6. Step 6 · Read all three together monthly, never sooner

   The joint read happens with a month of data behind it, not three days. That is the moment to ask whether the drop in calls came from losing the map, from the summary covering you, or from neither — in which case the problem lives somewhere else entirely.

A note on step 6. The question "why did my calls drop?" almost never has a single-variable answer, which is why the joint read matters so much. You may still be first in the map and the map may have stopped being on the first screen. You may have won the citation in the summary and lost clicks because of it, since the user already had the answer. The only way to tell those scenarios apart is having all three series with the same history and the same city.

One keyword, one city, one quota unit: what it costs to watch all three

The predictable objection to everything above is financial: "if there are three surfaces now, I am going to be billed three times." With most tools on the market that intuition is correct, because the standard model multiplies: keyword by city, by device, sometimes by search engine. In Rankiamos the billing unit is the keyword × city combination on each search engine: the same keyword on Google and on Bing (Bing is Pro only) is two units, but the three local surfaces all live on Google and are measured inside a single unit. Adding surfaces does not consume quota; adding cities does.

There is a second detail that saves a lot and that almost nobody compares when choosing a tool: on the Pro plan, mobile and desktop share the same unit (on Free you pick one device). You do not pay twice to see the same keyword on both devices. With that in hand, the math for a local business is trivial to estimate. Here are three realistic scenarios.

Reference figures, as of September 2026: the Pro plan is 14.99 USD per month, or 119.99 USD per year — which works out to 9.99 a month — and includes 5 projects, 50 quota units and automatic daily checks (a keyword that has not moved in 30 days drops to one check every 3 days until it moves). On both plans you can re-measure any keyword by hand 6 hours after its last check, up to 30 units a day on Pro and 13 on Free. The Free plan is 0 USD with no card, 2 projects, 5 units and a weekly check. Full details live on the pricing page.

Where this really shows is the third row. A business operating across five cities that wants to watch twenty terms in each is using a hundred units: exactly the Pro plan. In a tool that bills per city and per device, those same twenty keywords across five cities and two devices are two hundred billable units, and that is the tier jump that breaks a local business budget. Run that math against whatever tool you are evaluating before you sign, ours included.

One last allocation rule if you have to prioritize on limited quota. Proximity keywords — the urgent ones, the "near me" ones — need the fewest cities, because whoever searches them is already in your area. Judgment and complex-question keywords are the ones that benefit from multi-city measurement, because they compete against national portals that rank everywhere. If you have units to spare, spend them there.

And an honest platform note before you go further: Rankiamos runs on iPhone, iPad, Mac and Apple Watch, and the app interface is in Spanish only. There is no web version and no Android app. If your team works on Windows or Android, we are not your tool today, and I would rather you learn that in the pricing paragraph than after downloading the app.