What AI Overviews Did to Local Search Traffic
An HVAC owner in Easton showed me his analytics on a tablet in his office in March. Sessions down about 30% year over year. He had read three articles saying Google’s AI Overviews were eating everyone’s traffic, and he wanted to know what we were going to do about it.
What we did was look at where the sessions went. Eleven of his top fifteen lost pages were blog posts his last agency wrote in 2021 about furnace maintenance. His service pages were flat. His calls were up.
He was not losing money. He was losing readers he was never going to sell anything to. That distinction is worth about $40,000 a year in panic-driven marketing spend, so it is worth getting right.
Here is what I think AI Overviews actually did to local search traffic, what the honest evidence says, and the four numbers I would watch instead of the one everybody is watching. This is an opinion post. I will mark the places where I am guessing.
The claim everybody is repeating
The version going around goes like this. Google put an AI-written paragraph at the top of the results page. People read the paragraph and stop. Clicks to websites collapse. Every business that depends on search is in trouble, and local businesses are next.
The first half of that is measured and true. The last sentence is the part nobody has checked.
The Pew Research Center published the cleanest study on this in 2025. They watched the actual browsing behavior of around 900 US adults. On search result pages that included an AI summary, people clicked a link on 8% of visits. On pages without one, 15%. Roughly half the click-through, gone.
That is a real number from a real sample and I am not going to argue with it. I am going to argue about who it applies to.
What that number actually measured
Pew was looking at general web searches. The kind of question that has a text answer: what causes a furnace to short cycle, how long does a roof last, what is a heat pump.
Those are exactly the searches an AI paragraph is good at eating. The answer fits in four sentences, nobody wanted a website, they wanted the four sentences.
Now think about what a homeowner in Nazareth types when the water heater is leaking onto the basement floor at 7am. “Plumber nazareth pa.” That query does not produce a tidy paragraph. It produces a map with three businesses, phone numbers, and review counts. Google has been showing that map pack for over a decade and it is still the thing people tap.
I am not claiming AI Overviews never appear on local queries. They do, especially on the ones with a question in them. I am claiming that the query that produces a phone call for a contractor is the query least suited to being answered by a paragraph, because the answer is a business, not a fact.
Why most local contractors here were mostly spared
There is a second reason, and it is less flattering.
Most home services companies in the Lehigh Valley and the Poconos never had informational traffic to lose. Of the 402 regional businesses we audited, 34 have no website on their Google listing at all. Dozens more run a five-page template site with no blog, no cost pages, no FAQ. There was nothing on those sites for an AI Overview to replace.
You cannot lose a traffic channel you were never in.
The companies that did get hit are the ones that spent 2019 to 2022 paying an agency for a monthly blog post. Those posts were mostly generic, they mostly ranked for questions rather than for buyers, and the questions are what got absorbed. My honest read is that most of that content was never producing jobs in the first place, and AI Overviews just made the uselessness visible.
That is a mild position and I will defend it. If your traffic dropped and your calls did not, you did not lose customers. You lost an audience.
The part that really did go away
I do not want to talk anybody into comfort here, because two things genuinely changed and both matter.
First, the research step moved. A homeowner in Brodheadsville deciding between a boiler repair and a replacement used to read four contractor blog posts before calling anyone. Now she reads one AI paragraph. She still calls somebody. But she arrives at the call already anchored to a price range and a recommendation that you had no part in writing.
That is a real loss, and it is not visible in your analytics at all. It shows up as more people asking you to match a number they got somewhere else.
Second, the cost pages. “How much does a new roof cost in Pennsylvania” is exactly the kind of query an AI Overview swallows whole. It is also one of the few informational queries that really does precede a purchase. If you had a page ranking for that, you probably lost real traffic, not vanity traffic.
Which is an argument for writing cost pages so specifically that a general paragraph cannot replace them. A national average is easy to summarize. “A tear-off and replacement on a 1,900-square-foot Lehighton ranch, two layers, ran $11,400 last October” is not, because it does not exist anywhere else.
You cannot prove any of this at 900 sessions a month
Here is the limit on everything above, mine included.
A contractor site in this region gets somewhere between 300 and 3,000 sessions a month. At that scale, at least five things move your traffic line more than AI Overviews do:
- Season. HVAC searching in this region swings by more than half between February and April. Roofing follows storms, not months.
- Google core updates, which happen several times a year and are not announced with a ranking report for your site.
- Analytics changes. Consent banners, tracking blockers, a new GA4 setting your developer flipped.
- Bot filtering, which quietly removed a lot of fake traffic from a lot of small sites over the past few years.
- One competitor two towns over finally claiming their profile.
Anybody who tells you they isolated the AI Overview effect on your specific 1,100-session site is doing arithmetic on noise. We cannot do it either. I would rather say that plainly than sell you a report about it.
Four numbers I would watch instead
Sessions is the wrong top-line number for a local service business, and it was the wrong number before any of this started.
Calls, by source, by month. Not form fills. Calls. If you do not have call tracking, this whole conversation is guesswork and no analytics change will fix that.
Map-pack position, town by town. This is the one that actually holds still. If you hold the pack in Lehighton and lose it in Palmerton, that is a fact you can act on this week, and it does not move because an assistant felt different today.
Branded versus unbranded impressions in Search Console. If people are searching your company name more, your marketing is working regardless of what the traffic line does. If branded is flat and unbranded is down, that is a real problem.
Booked jobs per month against last year. The only number that pays for anything. Everything above is a leading indicator of it.
None of those four are affected much by whether Google writes a paragraph at the top of a page. That is the point. Pick metrics that survive a layout change.
Where I could be wrong
Two ways, and I think one of them is likely.
The likely one: Google’s AI Mode, the full conversational tab it rolled out in 2025, is a different animal from the Overview box. If that becomes the default way people search rather than a tab most people never open, local answers get compressed into a conversation and the map pack loses its position of privilege. I do not think that happens in 2026. I would not bet the business on 2028.
The less likely one: the informational traffic I just called vanity might matter more than I think, as a trust signal rather than as a lead source. If a homeowner reads your furnace article in November and calls you in February, my framework scores that article as worthless and it was not. I have no way to measure that and neither does anybody else.
Both of those argue for the same move, which is the convenient kind of conclusion I am normally suspicious of: keep publishing specific, local, priced, dated content, and stop measuring it with sessions. The broader AI search picture and the ChatGPT checklist both come out at the same place.
What to do next
Open your analytics tonight and sort your lost pages by traffic. If the top of that list is blog posts, relax. If it is your service pages or your cost pages, that is a different problem and it is not an AI problem.
Then count your calls against last April. That is the number that decides whether anything here matters to you.
What you cannot see from your own screen is your map-pack position across every town you serve. That is the number I would replace sessions with, and it is the one thing in this post you cannot check yourself in ten minutes.
Our Local Visibility Scan is free and it does exactly that. A grid of where you show up across the Lehigh Valley and the Poconos, your reviews against your top three competitors in each town, your site speed, and the biggest leak we can see. One page. No call to book, and your details are not for sale. If you want the full version with competitor and review analysis, that is the $399 Strategy Session, and you keep the report either way.
