Local visibility has stopped being something you buy
For twenty years a local business bought clicks and leads. AI assistants sell neither: they say names. What the Angi and Yelp filings show, what Pew and BrightLocal measured, and what our 872 US answers add.
8% against 15%
searches followed by a click to a website, with and without an AI answer (Pew Research Center, March 2025)
In this article
Short answer: a local business's acquisition channel is shifting from a market you buy into to an asset you hold. A click could be bought and a lead billed one at a time. An AI assistant sells neither to the business it names. It reads a corpus and says three names. This shift is not a forecast: it is already legible in the public filings of the intermediaries, and in what customers do.
45% against 6%
US consumers using AI to find a local business, one year apart (BrightLocal, 2026)
plus 98%
growth in Yelp's "other" revenue, which carries the licences granted to assistants (Q2 2026)
The old market sold access to the customer. The new one sells nothing to the business: it buys data from whoever owns it.
The lead market is not dying of AI
This is the easy story of the year: assistants killed the directories. The filings say otherwise. In the first quarter of 2026 Angi's revenue fell 3%, to 238.2 million dollars. The drop sits on one line: Network revenue down 56%, Network service requests down 55%, Network leads down 54%. The cause is written plainly in the release, and it is not artificial intelligence: it is "the implementation of homeowner choice in January 2025", a decision the company took itself.
The detail that kills the easy story sits two lines below: over the same period total US service requests returned to growth, up 5%. Household demand did not evaporate. What contracted is the lead-resale model, and it contracted because its owner changed the rules.
Yelp is not being replaced, it is being bought from
The other large intermediary tells the second half of the story. In the second quarter of 2026 Yelp's revenue grew 1%, to 376 million dollars. One line stands out: "other" revenue, which carries licensing, subscriptions and transaction fees, jumped 98%, to 33 million dollars. The release says why: Yelp's trusted content is "powering local discovery for ChatGPT and other AI partners".
In other words the assistant does not replace the directory: it buys its corpus. Value is not leaving the chain, it is changing seats. It leaves the resale of a contact to a professional and moves to the ownership of the data the machine consults before answering.
The customer has already moved
Two independent measurements frame the shift. The Pew Research Center followed the actual browsing of 900 US adults through March 2025, covering 68,879 Google searches. When a generated answer appears, 8% of visits end in a click to a standard result, against 15% when it does not. A click on a source cited inside the answer happens in just 1% of cases.
On the demand side, the BrightLocal 2026 survey of 1,002 US adults measures the swing in usage: 45% say they use AI to find a local business, against 6% a year earlier. Among those who use it, 63% say they trust its recommendations. One figure tempers the picture, and it matters: 97% of users at least sometimes check the answer against real reviews.
| Signal | Measurement | Period | Source |
|---|---|---|---|
| Clicks to a website when an AI answer appears | 8%, against 15% without | March 2025 | Pew Research Center, 68,879 searches |
| Clicks on a source cited inside the AI answer | 1% | March 2025 | Pew Research Center |
| Consumers using AI to find a local business | 45%, against 6% a year earlier | 2026 | BrightLocal, 1,002 US adults |
| Angi Network service requests (US) | minus 55% | Q1 2026 | Angi, 8-K filed 5 May 2026 |
| Total Angi service requests (US) | plus 5% | Q1 2026 | Angi, 8-K filed 5 May 2026 |
| Yelp other revenue, which includes AI licensing | 33 million dollars, plus 98% | Q2 2026 | Yelp, 8-K filing |
What the new layer gives, once measured
If visibility can no longer be bought, what is the spot worth? We put buyer questions from 15 trades in 13 US cities to ChatGPT and Gemini, 872 answers collected on 13 September 2026. For each trade and city we take the ten most-cited businesses and look at which ones both engines name.
The result says two things. First, there is no single AI ranking but two: only 4.6% of the most-cited names are named by both engines, and 139 trade and city pairs out of 195 share no name at all. Second, agreement concentrates exactly where a national brand exists: Robert Half is named by both engines in nine cities, Kforce in five, PwC in three. Where the trade is held by independents, agreement collapses: among attorneys, not one shared name out of 130 examined, in none of the thirteen cities.
The economic consequence is the point. In a local trade the position is unoccupied: nobody holds it yet, neither a national network nor the firm down the road. And it is not for sale: there is no counter at which to buy it. What decides it is what a machine manages to read about you in the corpora it consults, including, now, the ones it pays to read.
What this reading does not prove
That AI pushed Angi down.
The release attributes the fall to homeowner choice, an internal decision taken in January 2025. We report that cause as written, and nothing licenses us to substitute our own.
That directories are doomed.
One of them sells its corpus to the layer that was meant to erase it, and that revenue line doubles in a year. An intermediary that becomes a data supplier is not a dead intermediary.
That classic search optimisation is now useless.
A generated answer cites pages; the same foundations, business profile, reviews, a readable site, serve both channels. Pew measured clicks, not the usefulness of pages.
That 45% of customers buy through AI.
The survey measures a declared use for finding, at one moment, in one country. Finding is not ordering, and a declaration is not a log.
What it changes for a local business
Stop reasoning in cost per lead alone.
That cost still exists, but it now describes only part of the channel. The other part is not billed: it is observed, by reading what assistants answer about your trade and your city.
Treat your presence in the corpora as an asset.
Business profile, reviews, mentions in the local press, pages readable without JavaScript: that is the material the machine buys and re-reads. An asset is checked, maintained and defended.
Measure both engines, not "AI".
In a local trade their lists almost never overlap. A report that queries only one describes half the market.
Do not copy the national brands.
They are named because they are national, not because they found a recipe. Your advantage is the opposite of theirs: an empty spot, in one city, that nobody holds.