What Did AI Overviews Actually Do to My Local Pages?

What do I run, and what was I afraid of?
I am not an agency with a hundred clients. I am one person running a small portfolio of local lead-generation sites — a roofing site for Houston, an attic and mold site for Pasadena, a burst-pipe site for Brighton, Colorado, water-damage sites, a salary guide for the UAE, and a construction-cost site for Türkiye. Every one of them lives or dies on the same thing: showing up when somebody nearby searches for help.
So when Google started putting AI Overviews at the top of local searches — a written answer sitting above the map and the blue links — my honest first reaction was fear. If Google answers the question itself, who clicks? I did not trust the panic posts or the "nothing has changed" posts. I went and looked at my own Search Console data instead, site by site, page by page.
Which of my pages lost ground first?
The pattern was not random, and it was not everything at once. The pages that faded were my thinnest ones: broad pages built around a keyword first and a reader second. A page that tries to answer "roofing services" for a whole city, in general terms, with nothing specific to say — that page now competes with an AI summary that says the same general things faster.
I think of it this way now: the AI Overview is very good at being average. If your page is also average — same facts, same order, same vagueness as everyone else's — the summary can replace it completely, and the click simply never happens. My generic pages were borrowing their rankings from a world where Google had nothing better to show. That world ended.
What did my Houston numbers actually look like?
Let me give you the least glamorous number I own, because it is a real one. In early October, my Houston roofing site's Search Console showed an average position of about 58.9 for "roofing contractor houston" — deep on page six. I am not sharing that to impress anyone. I am sharing it because it taught me where the floor is.
That broad city-keyword page is exactly the kind of page this new search has no use for. It targets everyone in a huge city and answers nobody's specific question. A homeowner does not search like a keyword tool. They search for their street, their suburb, their problem, their price fear. The pages built for the keyword were built for a reader who does not exist.
Why did the specific local pages hold?
The other half of the split showed up on my UAE salary site. Through late October its Search Console showed the opposite behaviour: pages indexed, impressions growing — 52 impressions at an average position around 37 on 26 October, modest numbers, but moving the right way. And it was not the homepage or the generic pages doing the work. It was the specific ones: one profession, one city, one exact question about what a job pays in a place.
Those pages hold because they contain something the summary cannot invent: a specific answer, for a specific place, built from real published figures with sources and dates. An AI Overview can summarise the average of the internet. It cannot summarise a page that is the only place a particular local answer exists in that form. Specific is not a style choice any more. It is the moat.

Was it just my sites, or did other people see it too?
I tested the pattern in public. I wrote a post here called "Is SEO Dead in the AI Era?" and shared it on Threads, expecting an argument. What I got instead was recognition. Other SEOs replied with their own Search Console stories, unprompted, and they matched mine almost word for word: the pages that lost traffic were thin, top-of-funnel pages that AI Overviews absorb, and the pages answering specific local queries held or climbed.
One reply put it better than I had: AI did not kill SEO, it killed the shortcut version of it. That is the sentence I would hand to any worried site owner. Different people, different niches, different countries — same split. When independent data keeps landing in the same place, it stops being an opinion.
What do I check before I publish anything now?
I run one test on every page before it goes live, and it takes ten seconds. I read the page and ask: could an AI summary quote this page with confidence? If the page answers one question completely — with facts, specifics, and the local detail a real person in that town actually needs — it passes. If it reads like it could belong to any city and any company, it fails, and I rewrite it until it could not belong anywhere else.
Under that test, most of my old habits died. No page built around a keyword instead of a question. No city name swapped into the same template twenty times. No paragraph that says nothing the summary could not say. What survives is slower to write and harder to copy — which, I have come to believe, is precisely the point.
So what should a local site owner actually do?
Do not panic, and do not chase the summary. Audit your own pages the way I did: sort them into the specific ones that answer a real local question completely, and the generic ones that were borrowing rankings. Fix or retire the second pile. Then put your effort where the clicks still exist — the suburb, the service, the situation, the question nobody else bothered to answer properly.
If you run a local business and you are not sure which pile your pages fall into, that is exactly what my free 5-point SEO check and my Google Business Profile audits are for — the same sorting test I run on my own sites, applied to yours. AI Overviews did not end local SEO. They ended the version of it that never deserved the ranking in the first place.
But I would rather hear your data than only show you mine — so tell me: what did AI Overviews do to your local pages, and did your specific ones hold too?
Want SEO built on proof, not promises?
I build and rank websites for home service businesses — and I test everything on my own sites first.