AI search and local intent: what is different

Almost all the alarm you read about AI search comes from publishers, and publishers have a different problem. Their product is the answer. When the answer appears without the click, the business model takes the hit.

A local business sells a haircut, a repair, a table at seven. The answer cannot substitute for that. It can only route the customer, and routing is what local search always did.

What makes a local answer different from an informational one?

Verifiability, mostly.

An informational answer is assembled from sources that broadly agree, and small inaccuracies pass unnoticed. A local answer commits to specifics. This business, at this address, open until this time, this far from you.

Those specifics come mostly from structured local data rather than from prose on your site. Google states that generative responses can include information about local businesses, and that a Business Profile helps your services appear "in both AI responses and other Google Search results." That is the plainest guidance available in this whole subject, and it points at a free asset. The same documentation also points at Business Agent, Google's own layer for handling business information inside these surfaces.

Practitioner observation lines up with it. Whitespark's testing of AI Mode local answers in May 2026 found that star ratings, review volumes, pricing and hours are pulled from Google Business Profiles or comparable sources, and that reviews and unstructured citations, meaning blogs, news and community pages, are the heaviest inputs. That is vendor observation rather than a controlled test, so hold it loosely. It points at the same two assets Google does.

The consequence is a different priority order. For an informational query, your content is the lever. For a local query, your profile, your categories, your hours, your services and your reviews are the levers, and your website supports them.

A local business with a mediocre website and an excellent profile usually outperforms the reverse.

How often does local intent actually produce an AI answer?

This is where two credible studies disagree by a factor of eight, and the disagreement is more useful than either number alone.

Ahrefs analysed 146,122,391 desktop SERPs sampled in September 2025 and found AI Overviews on 7.9% of local searches, against 20.5% of all searches. Sterling Sky, tracking independently with different tooling, observes AI-powered local packs on roughly 7% of tracked keywords. Two vendors, two methods, the same neighbourhood.

Then a Whitespark study of 540 manual queries across six local industries reportedly found AI Overviews on around 68% of local business queries, and 97% for blended queries such as "best plumber for emergency leak in [city]."

Both are right about their own sample. Ahrefs classified local by intent tag across a keyword database dominated by short "[service] [city]" patterns, which almost never trigger an Overview. Whitespark hand-picked long conversational questions, which almost always do.

The synthesis is the operating rule: short transactional local queries are still local pack territory. Long conversational "help me choose" local queries are AI territory. Quoting 68% without that caveat oversells the change. Quoting 7.9% without it undersells the change. Your customers do both, and the second kind is growing.

How do you work out your own exposure?

Does local get hurt by AI answers the way publishing does?

Less, and for a structural reason. But the local contraction is real and it is not the one people are worried about.

Local intent ends in an action that happens off the internet. A call, a booking, a drive, a walk in. The answer can name three businesses, and someone still has to pick one and contact it. The click was never the product.

What does change is the shape of the shortlist, and Sterling Sky's 2026 field data puts numbers on it. AI-powered local packs are mobile only and US only at the time of writing, they show one or two businesses instead of three, and they carry no call buttons. A Places Scout analysis behind that report counted 5,943 unique businesses across AI local packs against 18,330 in regular three-packs, about 32% as many. Across 322 markets, 88% had fewer unique businesses in the AI version.

Fewer businesses get named. That is the actual risk, and it is a slow one.

The same report documents the trends that are already costing local operators money right now. Clicks-to-call from Business Profiles are declining even for profiles holding stable rankings, with Jepto data covering 179 profiles across 34 US law firms over two years. Website clicks did not fall the same way, because website icons show on desktop while call icons dominate mobile. The loss is mobile-specific and it is the call button, which is the metric a restaurant, dentist or plumber actually cares about. Meanwhile local pack ads went from roughly 1% of mobile reports in early 2025 to roughly 22% by December 2025, and Local Services Ads from roughly 11% to 31% of tracked queries. Joy Hawkins's summary: Google is going pay-to-play.

Sterling Sky is a local SEO agency, so this is vendor field data with disclosed tooling. Weigh it against the AI panic and the ranking of urgency changes. Disappearing call buttons and a paid results page are documented and immediate. AI search is a hedge.

What actually decides whether you get named?

The unglamorous inputs, in roughly this order.

A claimed, complete and accurate Business Profile. Primary category is the highest scoring individual local pack factor in Whitespark's 2026 survey of 47 local search experts, at 227 points, with proximity second at 225. Categories that describe what you actually do, not what you wish you sold.

Hours that are right, including holidays. In the same survey, "business is open at time of search" entered at number five with 189 points. A business the system believes is closed is a business it will not recommend.

A steady flow of recent reviews, which the survey scores separately from rating at number eleven. This is the input most within your control and the one most businesses run intermittently. The businesses that keep it going are the ones that built a system rather than a habit, which is why a review system has to survive a staff change to be worth anything.

Consistent details wherever your business is mentioned. Conflicting facts force a system to choose, and it does not always choose you.

Specific claims that are true and written down somewhere indexed. Parking, languages spoken, wheelchair access, late hours, group seating, emergency callouts. These are what the long questions filter on, and most businesses have never published them.

What actively hurts, and this one is new. Whitespark's 2026 survey added suspension-risk factors, and two of them are worth naming here: AI-generated text content on the profile, and AI-generated photos or videos on the profile. The reflex to fill a profile with generated description and generated imagery is now a documented risk to the asset that carries most of your local AI visibility. Also on that list: setting service areas larger than two hours' driving distance.

And what does not decide it: special files, artificial content chunking, content rewritten for machines, or purchased mentions. Google names all four as ineffective.

How should you measure local AI visibility?

By sampling carefully, and by reading your logs.

Sample with prompts written the way a local customer speaks, including the neighbourhood rather than the whole city, and run each prompt repeatedly. A single check tells you nothing, which is the trap a single AI visibility score is built on.

Expect geography to move results. Answers vary with where the person is standing, so a prompt run from your office is not the same as one run from three miles away. That is a feature of local, not an error.

Expect the engine to matter too, and be honest that nobody outside these companies knows which index ChatGPT uses for local. The long-standing view is Bing and Bing Places. Some practitioners argue it shifted toward Google in 2025. Whitespark and Greg Sterling both report not seeing Business Profile data inside ChatGPT, and Whitespark's own conclusion is that there is no firm industry consensus. The practical hedge is cheap: claim Bing Places, keep it accurate, stop there. An afternoon, no downside, non-zero expected value.

Then look at what your server logs say about which AI crawlers reached you, verified by IP range. It is the only fully deterministic measurement in the category, and it tells you whether retrieval is happening at all.

One thing not to do about weak local coverage: build forty near-identical city pages. That reflex predates AI search and it still backfires, since local landing pages start hurting past a certain number. Fan-out rewards specific pages, not templated ones with the city swapped.

What to do this week

1. Open your Business Profile and check hours, holiday hours, categories and services. That is the highest-value hour in this article. 2. Sort one month of your search terms by length and run the four-step arithmetic above. The long quarter of that list is your work queue. 3. Write down five true, specific things about your business that a customer might filter on, and publish them where they are indexed. 4. Remove any generated description or generated imagery from the profile. It is now a named suspension risk, not a shortcut. 5. Write ten prompts in neighbourhood language and run each seven times, which is the minimum the measurement research supports. Record who gets named. 6. Claim Bing Places if you have not. It costs an afternoon and it is the only cheap hedge available on the engine question nobody can settle.

If a partner or a client asks what all this means, what to tell someone who asks about AI search will keep you from overpromising.

Want to know which of your terms carry local intent?

Send me the three phrases customers use to describe what you do, in their words, to eric@seod.com. I will tell you which ones carry local intent, which sit in the long conversational band where AI answers actually appear, and which facts about your business you have not published that those questions filter on. No charge, no pitch.

I offer this because the gap is usually two or three specific facts you have never written down, and that is a fixable afternoon rather than a project.

Otherwise, keep reading the AI search library.