Google has confirmed that AI Overviews now expand to their full state automatically for some queries, pushing the ordinary search results well below the fold. In the same fortnight, a dataset covering thousands of Google Business Profiles showed US calls and website clicks still falling year over year while direction requests keep climbing. The two stories are the same story. The click is no longer a reliable proxy for the outcome, and reporting built on it is measuring a shrinking slice of what happens.
What changed in AI Overviews
Chris Long spotted it first and posted on X on 27 August 2026:
Holy moly SEOs: Google is now revealing the AI Mode prompt box BY DEFAULT for some queries. This is how Google continues to push AI Mode as the default search experience.
Barry Schwartz replicated it and got Google on the record. The statement, given to Search Engine Roundtable and Search Engine Land, was this:
For some queries, AI Overviews may dynamically expand for topics where our systems determine it’s most useful for people. Our research has shown that with this dynamic experience, users find Search more helpful and engage deeper in follow-up exploration.
Mechanically: the AI Overview used to load as a snippet with a “Show more” button. For some queries it now loads in full, with the “Ask anything” follow-up box already open. The ten blue links go a long way down the page. Robby Stein of Google added one detail worth knowing, which is that if a user has already started scrolling, the expansion is cancelled so they do not lose their place.
Schwartz’s assessment was blunt:
This is not good for publishers, not good at all. But I think we all expected this to happen.
Google has not said what share of queries this affects, or whether it will spread. Nobody who has watched the last two years would bet on it staying narrow. The direction has been visible since Bing first shipped conversational search in early 2023; what changed this month is that it became the default rather than a tab.
What the local data shows
The parallel evidence comes from Michel van Luijtelaar, writing at Search Engine Land with GMBapi’s Google Business Profile data. The piece is unusually honest about its own history. In April the same team declared local SEO dying on stage at BrightonSEO, based on Q1 numbers that appeared to show actions and impressions falling by roughly half. Then Q1 was revised. Their own verdict on that: “We were wrong about the scale.”
The corrected picture, year over year in the US:
- Q1 2026: website clicks and calls each down 15.8%; direction requests up 31.3%; desktop Search impressions up 12.3%; mobile Search impressions down 20.6%; desktop Maps impressions down 17.9%.
- Q2 2026: calls down 11.9%; website clicks down 12.5%; direction requests up 21.1%. Desktop Maps reversed from −17.9% to +3.2%. Mobile Maps up 30.4%. Desktop Search up 13.9%. Mobile Search down 20.1%.
Two things stand out. The declines are real but decelerating, not the collapse the Q1 read suggested. And the direction of travel is consistent: fewer calls and clicks, more direction requests, more Maps impressions. The reading offered is that US customers are increasingly completing the whole journey inside Google Maps rather than starting on a results page, and the desktop reversal suggests that is no longer only a mobile behaviour.
The picture outside the US is genuinely different rather than delayed. In the EU, desktop Maps impressions fell faster in Q2, from −16.5% to −34.7%, while direction request growth slowed from +21.7% to +13.1%. The UK’s smaller dataset swung the other way, with mobile Maps impressions going from +22.4% in Q1 to −70.8% in Q2 even as direction requests and website clicks kept rising. Anyone assuming their market is simply a year behind the US should look at their own numbers before relying on that.
Why rank tracking stopped answering the question
Sterling Sky and Jepto analysed 179 Google Business Profiles and found that AI-powered local packs often show two businesses rather than three, frequently without a click-to-call button, and surface only 32% as many unique businesses as the traditional Map Pack. Most rank trackers could not see any of it.
That is the crux. Rankings held steady while performance declined, which is the pattern Joy Hawkins, Claudia Tomina and Matt McGee were each reporting independently. A position report that says nothing changed can be perfectly accurate and completely misleading at the same time.
The clearest framing of why comes from the same Search Engine Land piece:
One system determines whether you appear on the map. The other determines whether Google’s AI trusts what it knows about your business enough to answer a customer’s question directly, without a click.
Traditional Maps ranking still runs on proximity, relevance, engagement and prominence. The signals behind AI Mode and Gemini add web context, entity matching, brand authority and review sentiment on top of the profile. You can hold your position in the first system while losing ground in the second, and no rank tracker will tell you.
So what do you measure instead?
Give direction requests equal billing with calls. If your monthly report leads with call volume, it is describing the part of the journey that is shrinking and ignoring the part that is growing. Directions are not a soft metric for a business people physically visit; they are frequently the last step before someone arrives. The same reasoning applies to lead type in Local Services Ads, where we found a charged phone lead costs roughly 1.8 times a charged text lead while both arrive in near-equal numbers.
Report keyword by keyword in position bands, not site averages. This is how we report for every client and it matters more now, not less. Counts of distinct keywords in positions 1–3, 4–10, 11–20, 21–50 and 51+, tracked across quarters, show the wave moving. A site-level average hides it. Split branded from non-branded, always, because brand contaminates every aggregate.
Pull Search Console in three-month windows. Longer windows lift more of the tail above the privacy threshold, which is where the movement usually starts.
Separate free product listings from organic. Use the searchAppearance dimension. Merchant listings sit at position one by definition and will drag any organic average down until you split them out.
Add the AI impressions column, and label it honestly. Search Console’s generative AI performance report went worldwide on 31 August. It gives you impressions from AI Overviews, AI Mode and AI features in Discover, with no click data and no query data. Track it as presence, separately, and resist folding it into a traffic chart. There is more on what that report does and does not contain.
Watch profile completeness as a performance input. The finding in the GMBapi data is that businesses with the most complete and accurate profile data capture more of the remaining clicks, calls and directions. When the pool shrinks, share of the pool is what is left to compete on.
Is this the death of local SEO?
No, and the people closest to the data have walked that claim back themselves. What the numbers describe is a channel where the easy proxy stopped working. The van Luijtelaar piece closes on a line we would happily borrow: local SEO has grown up, and it is starting to look a lot more like SEO. Entity consistency, third-party corroboration, review quality and content that answers the actual question are now the work, in the same way they became the work in organic search several years ago.
The uncomfortable part is the reporting conversation, not the strategy. A client who has been shown call volume every month for four years will read a 12% decline as a failure regardless of what else moved. Changing what you measure is easier before that conversation than during it.
Frequently asked questions
Are AI Overviews expanding for every query?
No. Google says they expand dynamically “for topics where our systems determine it’s most useful for people,” and has not said what proportion of queries that covers or whether it will widen.
Why are my rankings stable while calls fall?
Because two systems are now in play. Map Pack placement runs on proximity, relevance, engagement and prominence, while AI Mode and Gemini layer on web context, entity matching, brand authority and review sentiment. Holding position in the first does not guarantee inclusion in the second, and most rank trackers cannot see AI-powered local packs at all.
Should we report direction requests as conversions?
For a business with a physical location, they are closer to the outcome than a website click is. At minimum they belong in the report with the same prominence as calls, rather than in a footnote.
Did local performance really halve in early 2026?
No. That reading came from Q1 data that was later revised. The corrected year-over-year figures show US calls and website clicks down 15.8% in Q1 and down around 12% in Q2, with direction requests rising in both.
The short version
Google is answering more of the question on its own page, and customers are finishing more of the journey inside Maps. Both trends remove clicks without removing customers. Reporting that still treats the click as the outcome will keep showing a decline that the business is not actually experiencing, and will miss the decline that it is.
