How to get quoted by an assistant

We have written before about search growing a second front, where the customer asks an assistant a question and never sees a results page. That piece argued the shift was real. This one is about the practical question underneath it, which is what actually decides whether your store is the one named in the answer.

First, the size of the change, measured rather than asserted. Pew Research followed the real browsing of 900 US adults and found that when an AI summary appeared, people clicked a traditional result in 8 percent of searches, against 15 percent when no summary appeared. Clicks on the links inside the summary itself happened in 1 percent of visits. Being read to is replacing being clicked through to.

Ranking is no longer the ticket in

The instinct is to assume the answer is drawn from whatever ranks first, so the old game still applies. That was roughly true a year ago and is not true now. Ahrefs looked at 863,000 keyword result pages and 4 million cited URLs and found only about 38 percent of the pages cited in AI Overviews also ranked in the top ten. In their earlier study, that figure was around 76 percent.

In other words, the share of citations that came from the visible first page has halved in under a year. Roughly a third of cited pages do not rank in the top hundred for the query at all. The model goes looking sideways, asking related questions and pulling from wherever the clearest answer sits. That is bad news if you were counting on position one to carry you, and unusually good news for a small store, because it means the ranking hierarchy you were never going to beat is not the gate anymore.

What correlates instead

Ahrefs also ran the correlations across 75,000 brands, testing which measurable factors line up with being mentioned by ChatGPT, AI Mode and AI Overviews. The strongest by a wide margin was branded web mentions, meaning the number of places on the web where other people write your name. It scored 0.664 for ChatGPT, 0.709 for AI Mode and 0.656 for AI Overviews.

Backlinks, the currency of twenty years of SEO, sat between roughly 0.19 and 0.25. Domain rating landed between 0.266 and 0.326. Correlation is not proof of cause, and Ahrefs are careful to say so. But the direction is consistent across three different systems built by different companies, and it matches how these models are trained. They learn what the web says about a thing, not who linked to it.

For a store, that reframes the work. Getting quoted is less about engineering your own page and more about being described, consistently and in plain language, in enough places that a model reading the web arrives at a stable idea of what you sell and who you are.

A model does not count your links. It notices what everyone else already said about you.

What a small store can actually do

Be describable in the words a buyer would use. If your product is a soy candle that burns forty hours and costs under thirty dollars, that sentence should exist somewhere on your site in that form. Poetic collection names are fine as a layer on top and useless as the only layer.

Be consistent. The same product name, the same category words, the same claims across your site, your marketplace listings, your social profiles and anywhere you are written about. Every variation is one more thing a model has to guess is the same object, and guessing is where it gives up and cites someone else.

Be somewhere other than your own site. A local write-up, a niche roundup, a supplier page, a marketplace profile. These have always been slow, unglamorous marketing. They now happen to be the input with the strongest measured relationship to being named in an answer.

And keep it current, because all of it decays. Prices move, products sell out, seasons turn, and a description written eighteen months ago is a description of a store that no longer exists. This is the getting-found work inside Vantage. Listings written the way people actually search, kept accurate as the catalogue changes, structured so both a crawler and an assistant can read them without guessing, with every change waiting on your approval before it ships.

None of this is a trick, and anyone selling you one is selling the 2026 version of a meta-keywords package. The whole of it is being unusually clear about what you sell, in more than one place, on purpose, over and over.