How brands get cited in AI search, and why your own site is rarely the source

When an AI assistant answers a question in your category, it is quoting pages. Most of those pages are not yours, and no amount of work on your own website changes that. The brands that get named are the ones already written about in the places the model reads.

That one fact reorganises the budget. If the answer is assembled out of third-party pages, then getting into it is a media problem before it is a technical one, and the work is commissioning rather than tidying.

An answer is assembled out of pages, and most of them are not yours

Ask ChatGPT, Gemini or Perplexity to compare the options in a category and it does two things. It retrieves a set of pages it judges relevant, then it writes a short answer out of them. The brands that appear in the answer are, in the main, the brands that appeared in those pages. So which pages get retrieved is the whole question.

The largest published count of what gets cited puts the split plainly. Foundation Marketing and AirOps, 5 May 2026, examined 57.2 million citations across 5.1 million AI responses covering 50 brands in seven verticals and found that 90% of the sources AI cites when answering are ones the brand does not control; only 10.15% of citations go to brand-owned domains. The market the sample covers is not stated. Read the study.

Read that as an operating instruction rather than a headline. The paragraph a buyer reads about your brand is written from a set of pages you do not own, have never briefed, and in most cases could not currently name.

Your own pages cannot answer an open question

It is not because the site is badly built. It is because of what the question asks for. A buyer early in a decision asks an open question, and an open question needs an answer that weighs several options against each other. A page whose job is to say that one brand is the right choice cannot do that job. A comparison piece or a category explainer written by a publication with no stake in which brand wins is something a model can quote without contradicting itself.

Your own pages still have real work, and it is different work. They serve the buyer who has already chosen you and wants the practical detail: what is included and how to start. Get those wrong and a model will repeat them wrong. But they are not where a shortlist is decided, and treating them as though they were is how a year of content budget goes into pages that were never eligible for the question you wanted to win. AI visibility is won on other people's pages.

Being the biggest name in the category buys you nothing here

The evidence on this is unusually clean. Surfer, July 2026, took three months of its own tracking across about 5 million unique citation source URLs and 20,000 prompts and found about zero correlation between a domain's authority score and whether AI cites it, at Spearman minus 0.073 to plus 0.026. The market is not stated. Read the analysis. Offline reputation no longer buys a seat in the answer.

For a challenger that is the best news in the category, because the ground is not allocated by size. A brand that is third on market share can be first in the answer if the pages deciding it say so. For the incumbent it is the same finding read the other way, and it is the more urgent reading of the two.

UK shoppers using an AI assistant more than doubled in a year

Censuswide for Adyen, January 2026, in a nationally representative sample of 2,000 UK consumers aged 16 and over fielded from 21 November to 3 December 2025, found that 28% of UK shoppers now use an AI assistant when shopping, more than double the 12% a year earlier. Read the report.

That is not everybody, and we would rather say so than argue from a collapse that has not happened. The click is not the point. The point is that a growing share of people considering a purchase in travel, finance, appliances or telco now read one composed paragraph about the category before they read a word you wrote, and that paragraph is built from sources nobody at your end has briefed.

Two things get a brand into the sources, and most brands have one

Both have to be true at once. The brand needs a claim worth quoting, meaning something specific, attributable and comparable, because a model repeating a claim needs something it can hold up against a rival's. And that claim needs to be carried by publications the model already reads, written by their journalists to their standard, because that independence is what makes the page worth quoting in the first place.

The mechanism has a name and it is not new. Branded content is a publisher's own team writing a piece to a brief, on their site, to their editorial rules. What has changed is what it is worth. The same article that once bought a category's attention for a week now sits among the pages a model reads every time it answers a question in that category.

The first of those two conditions is work a marketing team can do with what it already knows about its own product. The second is where the category stops being helpful to you. Software that measures AI answers can name the publications deciding your category's answers, and then it has to stop, because reading an answer and commissioning the pages behind it are different businesses. Avid commissions content across a network of more than 600 trusted media brands in Australia, the UK and the United States. We buy editorial for a living, so the list of titles a report hands you is a list we can act on, not one to admire.

One article is a data point, four in a window is the category's view

One article in one title is a single data point, which a model can weight or ignore. The same claim carried by several unaffiliated publications inside the same window reads as the settled view of a category rather than one publisher's opinion. That is the shape worth buying, and it is the reason a campaign starts at four publishers rather than one. We then keep the claim current, because a point that stops being restated stops being the one on hand when the question comes round again.

None of that is new thinking about media. Trust sits with the publication and cannot be bought with owned media, an argument that held long before AI answers existed. What AI answers added is a way to see it working. The hard part was always briefing several publishers at once and keeping the standard even across all of them. One brief goes to many partners in a single workspace.

Until you have the source list, every content decision is a guess

Start by finding out which pages decide your category's answers today. A free AI Visibility Report runs the questions your buyers ask across the engines and shows where your brand appears, what gets said about it, and which publications the answer is being built from. The case studies show what the work looks like once it runs. Nobody can promise a model will name you. We can show you the sources deciding it, and we can put your claim in front of the people who write them.