The Machine Answers First Now
A field note from the SEO work
I have done search work long enough to remember when the whole job was ten blue links and an argument about their order. That world is gone, and I do not think it is coming back. When somebody searches for a business like yours now, the first thing they meet is a paragraph, written by a machine, in a confident tone, before a single link appears.
The numbers caught up with what I was already seeing. Search Engine Land, reporting Semrush's tracking of 400 US desktop commercial keywords, found Google's AI Overviews on 88% of them in August 2026. That is nearly double the share those answers held in September 2025. In the space of a year, the machine's answer went from an experiment to the front of the line.
What I keep seeing in the work
The answer on top is assembled from parts. The machine gathers what it can find about a business—its website, its Google Business Profile, its directory listings, reviews, stray mentions—and it trusts whatever agrees with itself. Then it summarizes. Here is what that looks like from where I sit:
A multi-location retailer I work with needs to be the named answer in four different towns, for four different stores, and the facts for those stores have to be right town by town. A therapy practice I audited had its name, address, and phone number disagreeing with each other across the very sources the machines read first, so the machines hedged, and hedging means somebody else gets named. A media company found that its own name was being patiently explained to the world by everyone except the company itself.
Bad websites were missing from that list; the websites were mostly fine. The problem is that a business no longer speaks with one voice in one place. It speaks in fragments, scattered across a decade of listings, profiles, and pages, and the machine is the first reader who ever tried to hear all of it at once. It believes the fragments that agree. It discards the ones that don't. Sometimes the fragment it discards is your phone number.
You do not get to opt out of being summarized
Owners tell me they want nothing to do with AI. I understand the sentiment. The AI writes about them anyway. The answer about their business exists today, in production, in front of their customers, whether or not anyone at the business has ever read it. Being summarized has become a condition of being findable at all.
The only choice on the table is whether the summary is accurate—whether it names you, describes what you actually do, and cites your pages as the source. That is a choice you make by tending the record, or one that gets made for you by whoever's record is tidier.
Claim it, then measure it
The fix is unglamorous. First, you claim your entity: one canonical account of what the business is, stated on your own site, tied together with entity schema and sameAs links, an llms.txt file for the machines that read those, and a page that answers "what is this business" so plainly that a summarizer can quote it instead of improvising. Second, you fix the disagreements: the listings, the profile, and the site telling one story. Third—and this is the part almost nobody does—you measure. AI answers drift. A competitor publishes, a directory changes, and the answer that named you in the spring names someone else by summer. So the money questions get re-asked every month, and the changes get written down.
That is the work I sell, and I sell it in pieces so you can stop wherever it stops making sense: an AI Visibility Audit for $750 that tells you exactly what the answers say about you today, a Claim Sprint for $1,500 that fixes the layer underneath, and AI Presence, Measured at $500 a month for the receipt that keeps it true.
Let's have a chat about it
If you have never actually read what the machine says about your business, you are flying on instruments somebody else set. Let's have a chat about it. I can help. Thirty minutes, bring your website and the searches you wish you won: book a 30 minute call.
Updated