Most AI answers about your business link to somebody else's page

Count the source links, then count how many are yours

Three times out of four, the page speaking for your business inside an AI answer is one you didn't write. Omniscient Digital got to that number by sorting more than 23,000 of the links AI assistants list under their answers, the ones that show where the answer came from. On questions where a person named a business, pages on that business's own website came to about 23 percent. The rest split two ways: articles other people wrote at roughly 48 percent, and pages built to sell something, whoever published them, at roughly 30 percent. (All three are rounded, so they add up to a little past 100.) Search Engine Journal wrote the study up on July 22, 2026.

The reason for this is dull. A customer's question usually covers more than one business at a time: who does emergency furnace work nearby, or which medspa does lip filler under 600 dollars. One best-of article answers that whole question on a single page, which is why the assistant reaches for it. Your own site speaks for one business out of the several a customer is weighing.

Omniscient counted brands of every size and kind, so the split says nothing on its own about a dental practice or a heating company. Read it only as a rough direction. Then go and look at your own category, because that part you can do today.

What that looked like on a real results page

On August 12, 2026, Joy Hawkins, a local search analyst, posted a screenshot of Google doing this about as crudely as it gets. Google had written a short list of law firms and put it above the normal blue links. Here is how Hawkins described it: "Google is quoting a single lawyer's self-promoting listicle for every law firm listed here and is using it as the primary source for this entire list, linking to it instead of the law firms' websites."

Read that from the customer's side. She searched for a lawyer. Google handed her a write-up of several firms, every one of them drawn from one competing lawyer's best-of article, and the links sent her to his article.

Barry Schwartz of Search Engine Roundtable wrote it up the same day. He says he reproduced the result himself after several tries on a few different browsers. His verdict was short: "This falls far short of quality and fails to make Google's AI results look helpful; it is simply a shame."

Nobody suggested the firms had done anything wrong, and there's no sign they had. They were described on a live results page by an article they had nothing to do with, written by a man who wanted their customers.

That's one documented case that a second person reproduced, so nobody should read it as proof that Google behaves this way everywhere. Still, that is what the 23 percent looks like when it reaches somebody about to spend money.

The other counts are thinner, and they point the same way

Whitespark ran 540 searches by hand for a 2025 study. The answer Google writes at the top of the results page, above the links, showed up on 68 percent of local business searches. The little map with three businesses under it showed up on 39 percent. So on most local searches there is now a written answer sitting above the map, and the links under that answer matter as much as your spot on it.

Google listed 23 sources across nine plumber questions, the kind that ask what a job costs and who nearby does it. Whitespark again. About 60 percent of those pointed at other people's websites (Yelp, Thumbtack, HomeGuide, Reddit, Quora and two large job boards), which left about 40 percent for the plumbers themselves. Whitespark collected those searches from outside the areas they were about, so what they saw is roughly what a stranger to your town sees. That's 23 links from nine questions, and I wouldn't build a year's plan on it.

Page one of Google turns out to be a poor guide to what gets quoted. The count I keep going back to is Bill Widmer's at Orbit Media. He logged 13,184 source links from ChatGPT, Claude, Perplexity and Gemini through August 23, 2026. Only about 10 to 30 percent of the websites they used also ranked in Google's top ten for the same question, and the assistants were using best-of lists and review sites instead. His sample is 72 prompts across three business-software brands (he says himself it only points a direction), so it's not your market either.

Four partial views, each one with a hole in it, all pointing the same way. That's worth five minutes of your week.

The five-minute check

You need an ordinary web browser, and five minutes for the first question.

1. Write down the sentence a customer would actually type. "How much is a crown without insurance" beats "restorative pricing options".
2. Type it into Google. Read the answer at the top and write down every link underneath it.
3. Type the same sentence into ChatGPT. Write down its source links too.
4. Cross off anything that goes to your own website. What's left is the list.
5. Open the first two and read what they say about your business, or what they say in place of it.

Step 5 is where it pays off. Sit an owner down in front of that list and the page that stops them is one nobody at the business had ever heard of, doing the talking for them. A best-of article that names four competitors and leaves you out is a page you can email the writer about. Your profile on a review site might still show the hours you changed two years ago (that one is a ten-minute fix).

That is five minutes spent on one question. Run it on the three questions that bring in the most money, closer to fifteen minutes, and you end up with maybe eight or ten pages. A few of them will turn up on every question. Those repeats are your real competitive set for that work. Often they belong to names you'd never have put on a competitor list.

Widmer's own advice, after logging all those source links, is the same move. He puts it like this: "List every third-party page cited instead of you, count repeats, and pitch the ones that keep appearing." Going after them is slow work. It means emailing the writer of the best-of article, or writing a page of your own that answers the question in the customer's own words. None of that lands this week.

Put it back on the calendar for three months from now. The answers move around, and a page that turns up on both checks has earned your attention. One that showed up only the first time may have caught a bad afternoon at the search engine. Two passes on one question, three months apart, cost you ten minutes and tell you which pages are sticking around.

The first five minutes land today. Most operators have never once looked at which pages are answering for them, and that's a strange thing to leave unchecked.