Google’s AI Now Answers Nearly 9 in 10 Buyer-Stage Searches. Here’s How a Small Business Gets Picked.

Google’s AI summary now shows up on roughly nine in ten buyer-stage searches in a new Semrush study, and a small business gets picked by answering the buyer’s question in plain words, early on the page, with real facts and a real price. Pages that bury the answer, hide the price or dodge the question get skipped in favor of someone who didn’t.

I recently wrote that independent blogs get about 1% of AI Overview citations. The easy objection is: “Fine, but that’s blog stuff. My buyers don’t search that way.” Then this study came out, and it’s about the searches that do send buyers.

What the study actually found

Semrush tracked the same 400 commercial keywords every month from September 2025 to August 2026. These were searches with a decision-stage modifier attached: “best X for Y,” “reviews,” “price,” “alternatives,” “X vs Y.” In September 2025, an AI Overview appeared on 45.75% of them. By August 2026, it was 88%. Semrush published the study and Search Engine Land covered it on October 22.

Weighted by search volume, the number was 97.08%. “Reviews” queries went from 30.14% to 89.04%, and “best X for Y” finished highest at 96.12%.

Now the fine print, because I’d rather you hear it from me. The panel was 400 keywords in banking, software and ecommerce, on U.S. desktop. It was not a local-search study, so I can’t tell you what happens to “plumber near me” from this data. And it measured whether the summary appeared, not what happened to clicks. PPC Land’s breakdown also points out that Semrush sells AI visibility tracking, so it has a horse in this race. Still, a direction that clear is worth taking seriously.

Picture this: before and after

This is an illustration, not a screenshot. Say a marketing manager searches “best managed IT provider for a 40-person law firm.”

A couple of years ago, she’d see ten blue links and a few ads. She’d open three tabs, skim, and call two firms. Your page had a fair shot just by ranking.

Today she sees a written summary at the top: what to look for, a few typical price ranges, a couple of names. The links sit underneath it. If she’s satisfied, she calls the names in the summary. If your page never gave the summary a clean sentence to quote, you weren’t in the tab pile. You weren’t in the room.

The playbook (it’s not magic)

Most business sites, mine included at various points, are written to impress peers instead of answer buyers. Here’s what I’d do about it.

1. Open with the answer. The first one or two sentences under a heading should answer it outright. Story and context come after. (Yes, I did that to you in this post’s first paragraph.)
2. Make headings real questions. “How much does a website redesign cost?” beats “Investment.” Buyers type questions, and so do the machines.
3. Put extractable facts on the page. Who it’s for, how long it takes, what’s included, what isn’t. Short, specific, quotable. A table or tight list works.
4. Show real pricing. Even a range. “Contact us for a quote” gives a summary nothing to repeat, and it gives your competitor’s number the stage. We try to practice this ourselves with what WordPress maintenance should actually cost, price ranges and all.
5. Check what’s actually being said about you. The Semrush write-up shows a summary where one brand’s own page made its case but a third-party site made the case for its rival. Your reviews and directory listings feed the answer too.

If you want to see how you’re doing right now, the 5-minute AI visibility self-check is a decent place to start. And I wrote up what I’m doing about it on my own site, warts included.

Why a one-time fix won’t hold

Here’s the part that should bother you. A separate study, from a company called Envoyra, asked ChatGPT, Claude, Gemini and Perplexity 170 buyer-style questions a week for four weeks (2,720 answers total). Search Engine Land’s write-up says 79% of the question-and-assistant combinations changed their recommended brands at least once. Only 144 of 680 stayed identical all four weeks.

Caveats again: Envoyra didn’t publish its questions, and it tested API models rather than the consumer apps. Take it as a signal, not gospel. But the signal matches what I see: who gets named shifts by platform and by week.

You can’t “do AI SEO” once in October and frame it. Pages go stale, prices change, competitors publish, and the answer machines reshuffle. It’s closer to maintaining a site than building one: somebody has to keep checking, keep fixing, keep the facts current. That’s why I think of it as an ongoing job, which is how our AI SEO service works.

And if Google’s AI is already calling businesses to ask their prices, a page that states them clearly is no longer a nice-to-have.

Where to start this week

Pick your single most valuable service page. Rewrite its first two sentences so they answer the buyer’s question, add a price range, and turn one vague heading into a real question. That’s an hour, tops.

If you’d rather see where you stand first, our SEO Snapshot is the low-pressure way to find out.

Have you searched your own money keywords lately? What did the AI summary say about you, if anything? I’d like to hear it.