Local SEO and AI search: two systems, two different jobs

The map pack did not go away. The research did.

There is a version of this story where AI Overviews ate local search and the map pack is finished. The numbers do not support it, and what they do support is more useful.

Whitespark ran 540 queries across Houston, Phoenix and Denver, covering plumbers, personal injury lawyers, dentists, optometrists, medical clinics and real estate agents, and published the results on 12 May 2025. Averaged across everything, AI Overviews showed on 68% of the searches and local packs on 39%. Read only that line and you would conclude the AI layer has taken over.

Split it by intent and it inverts. For explicitly local queries, the plumber near me kind, local packs appeared 93% of the time and AI Overviews 15%. For informational queries, AI Overviews appeared 92% of the time and local packs 6%. For hybrid queries, the ones carrying both a service and a city, AI Overviews hit 97%.

So the AI layer has not taken the transaction. It has taken everything that happens before the transaction. Someone working out whether the pain they have warrants a dentist, what a root canal should cost, and how long one takes now gets all three answers without clicking anything. By the time they search for a dentist in their suburb, the shortlist was formed somewhere you were not.

That is worse than losing the map pack, not better. Losing the map pack is visible on a rank tracker. Losing the research stage is invisible. Your positions hold, your calls fall, and nothing on the dashboard explains the gap. It is the same diagnostic problem we describe in why is my traffic down, one layer earlier in the funnel.

One caveat, stated plainly because the number gets quoted as law: this is a single study of 540 queries, six industries, three American cities, collected in the first half of 2025. It is the largest published sample we could find on the question. Directionally it is strong. It is not a physical constant, and anyone quoting the 68% at you without the intent split is quoting the least useful figure in the report.

The map pack is a geography machine

Google publishes what it uses, in plain language. Relevance is how well a Business Profile matches what someone is searching for. Distance is how far each business is from the customer who is searching. Prominence is how well known a business is. Three factors, and one of them is a map coordinate.

Whitespark's 2026 Local Search Ranking Factors report, published 6 November 2025, asked 47 local search specialists to weight 187 factors against each other. The top of the local pack list ran: primary Business Profile category, proximity of the address to the point of search, keywords in the business title, a physical address in the city of search, and whether the business is open at the time of the search.

Four of those five are about where you are and what you called yourself. The fifth is opening hours. Nothing in the top five is about your content.

That is the honest reason serious local work is a setup job rather than an ongoing campaign. Most of the winnable ground is configuration, and the single strongest lever is the primary category, which the average business sets once, on the day it claims the profile, to whatever generic label sounded closest. Everything after that is proximity, which you cannot change without moving, and prominence, which is the slow half. If you want that configuration done properly rather than guessed at, that is what Local SEO Setup is for.

Your profile is not a citation

Here is where the two systems separate, and it is the part almost nobody says out loud.

In the same Whitespark study, the citations inside AI Overviews for local queries split roughly 60/40. Sixty percent pointed at third party publishers and forty percent at individual local businesses. In the plumber example the cited sources included Reddit, Quora, Yelp, Thumbtack, HomeGuide, Indeed and ZipRecruiter.

Your Business Profile is not on that list, and structurally it cannot be. A profile is a record inside Google's database. A citation is a URL that a model retrieved and quoted back to a reader. They are different objects, and improving the first does not produce the second.

Which means the work that wins the AI half of local search is not profile work at all. It is being a nameable, verifiable, consistently described entity on the pages that do get cited: the review platforms, the best-in-city roundups, the industry directories, the local press, the community threads where people ask for recommendations by name. That is entity work, it is slower than a profile audit, and it is a separate line in the budget. Agencies that sell you one and describe it as the other are not lying so much as failing to notice the difference.

Nobody can tell you which sources each engine weights

Ask the internet where ChatGPT gets local business data and most answers trace back to one place: a Yext post from 13 June 2024 that asked each model to name its own sources. The models obliged with tidy lists naming Yelp, Bing, Facebook, Tripadvisor, and roughly fifteen smaller directories including Hotfrog, Zomato and the BBB.

That is not evidence. Asking a language model to describe its own retrieval is asking it to produce a plausible sentence about itself, which it will do whether or not it has any access to the answer. Those lists now circulate as though they were documentation. They are self-report, and self-report from a system with no privileged view of its own plumbing.

What can be said with a source is narrower and more useful. Whitespark's guide to Google AI Mode, updated 22 May 2026, describes AI Mode assembling a business answer from Business Profile data, review content from sites like Yelp, and unstructured citations from blogs, news and community pages, and doing it through query fan-out, where one user question triggers several underlying searches before anything is shown. Business Profile data is an input. It is not the input.

If you want to know which sources matter for your business, in your category, in your city, the only honest method is to go and measure it: fix a prompt set, run it, log who gets cited, repeat on a schedule. The method is in measuring AI visibility. Everything else is somebody's guess wearing a chart.

Inconsistency costs more than it used to

Consistent name, address and phone details across listings has been local SEO hygiene advice for fifteen years, and it was always slightly dull, because the worst case was a duplicate listing and a mildly confused crawler.

Query fan-out changes the arithmetic. When a model assembles one answer out of your profile, your website, two review sites and a directory, and those five sources disagree about your hours, your suite number, or whether you still take emergency callouts, it does not flag the conflict for anybody. It resolves it, silently, and it can resolve it in favour of the stale source, because nothing in that pipeline knows which of your five records is the current one.

You do not get an error message. You get a confident, wrong recommendation with your name attached, delivered to somebody who will not see the contradiction because they only ever saw one version. That is the failure mode that has genuinely changed, and it is the argument for treating a citation sweep as visibility work rather than housekeeping.

Access is the other half, and it fails separately. If the AI crawlers cannot reach your site, none of the above applies to you at all, and a firewall rule can end the conversation before it starts. The list of who to let in is in the AI crawlers guide.

The AI answer is a shortlist, not a destination

BrightLocal surveyed 1,002 American consumers and published on 10 March 2026. Forty-five percent had used AI tools to find a local business, up from 6% a year earlier. Among active users, 63% said they trust the recommendations. AI now sits third behind Google and Facebook as a way of finding local businesses, ahead of Yelp and Tripadvisor. ChatGPT led on 31%, Google AI Mode on 23%.

The figure worth building around is a different one. Eighty-eight percent of AI users said they verify the recommendation before acting on it, checking whether the business is real or where the answer came from.

So the AI answer does not replace the profile. It creates traffic to it. Getting named in the answer and then failing the check, because the hours are wrong, or the newest review is from 2019, or the pin drops on the wrong street, wastes the hardest-won part of the sequence. Both halves have to be true at the same time, and that is the actual shape of the job now.

What this changes about the work

Not as much as the panic suggests, and not in the direction most people assume. The configuration work got more valuable, not less, because it is now the thing that has to survive a verification click rather than just win a ranking. Set the primary category deliberately against real search volume instead of instinct. Keep hours, including holiday hours, actually current, because Google names being open at the time of search as a top-five local pack factor and a model reading stale hours will state them as fact. Publish a real page for each service you sell, which was already the strongest local organic factor in Whitespark's survey and now doubles as the retrievable, quotable version of what you do.

Then add the half that is new. Audit every place your details appear and reconcile them, treating each mismatch as a fact you are handing to a model rather than a tidiness problem. Work out which third party sources actually get cited for your category and city, and go and earn a place on those specific ones instead of buying a directory blast. And measure, on a schedule, with a frozen prompt set, so that when the answer changes you can tell whether it was you or the model.

None of that is exotic. It is the same discipline applied to a second retrieval system that happens to have different inputs. The mistake worth avoiding is assuming the first system's work carries over, because the evidence says it mostly does not.

Sources

Google's three named local ranking factors come from Tips to improve your local ranking on Google in the Business Profile help centre. The prevalence figures, the intent split and the 60/40 citation split come from Whitespark's research on AI Overviews for local business searches by Miriam Ellis, 12 May 2025, based on 540 queries across three cities and six industries. The ranking factor weights come from the 2026 Local Search Ranking Factors report, published 6 November 2025, in which 47 specialists scored 187 factors. The description of query fan-out and of Business Profile data as one input among several comes from Whitespark's guide to Google AI Mode for local businesses, updated 22 May 2026. The consumer adoption, trust and verification figures come from BrightLocal's research on AI and local business recommendations, published 10 March 2026, surveying 1,002 US consumers of whom 455 had used AI tools. The self-reported source lists we argue against are from Yext, 13 June 2024, whose stated method was asking each model to name its own sources. Three of those six sources are Whitespark, which is a concentration worth knowing about: they are the only party publishing primary local research at this scale, and no larger independent study of AI Overview prevalence in local search has been released.

Common questions