High in Google, and still less reach

A marketing manager at a mid-market company watches his organic traffic drop quarter after quarter. The rankings barely move. The content is there. The technical side is sound. And yet he reaches fewer people. His agency points at the algorithm. The problem lies somewhere else: his website is no longer being asked to take part.

The search engine no longer hands out lists

Type a question into ChatGPT, Perplexity or Google AI Overviews today and you get an answer, not a list of ten blue links. That answer leans on two or three sources. For that user, at that moment, the rest of the web does not exist.

Google AI Overviews now reach 1.5 billion users a month. ChatGPT counts 900 million weekly active users. In 2025, 60% of all Google searches end without a click to a website. In Google AI Mode that rises to 93%. This is not a temporary dip. It is a structural shift in how people search for information.

Rankings and citations follow different logic

The easy assumption is that whoever ranks well in Google is automatically cited by AI search engines too. The data contradicts that directly. Research by BrightEdge shows that only 17% of the sources Google AI Overviews cite also appear in the organic top 10. Five out of six AI citations come from content that is not on page 1.

At the same time, research by Seer Interactive shows that brands that are cited in an AI Overview earn 35% more organic clicks than brands that are not cited, in a market where the overall CTR is falling by 61%. Citation does not replace SEO. It is a separate discipline that most organisations do not yet have.

HubSpot lost half of its traffic

The most instructive example comes from one of the strongest content marketing teams in the world. Between November 2024 and early 2025, HubSpot lost more than half of its organic traffic: from roughly 13.5 million to 6 million monthly visits. The rankings barely moved. HubSpot CEO Yamini Rangan acknowledged it publicly: "AI overviews are giving answers, and fewer people are clicking through."

If this happens to a company that built its entire growth on content marketing, it is naive to think that mid-market organisations without a dedicated SEO team are immune.

The channel disappears inward

What is happening here is, at its core, an old problem: how do you reach your customer? Every organisation has a channel for that, and the underlying question is simple: does the route to your customer run through an intermediary, or via your own direct path?

Twenty years of SEO made that question invisible. Google was the channel. You published, Google indexed, the customer came by. The intermediary was so reliable that it did not feel like an intermediary. That is over. The AI layer is a new intermediary, one that is more selective, changes faster and does not necessarily pass your source on to the customer. Anyone who exists only through that channel exists less and less.

For a mid-market organisation, this means an honest question. What percentage of your customer contact runs through channels you control yourself (a mailing list, a community, a customer portal, a direct relationship), and what percentage depends on a third party that changes the rules? Anyone leaning on that second pile is building on sand.

Thin pages get skipped

Citation-worthiness is binary. A page is either chosen by an AI as a source or it is not, and volume does not compensate for quality. Better one strong page a week than ten thin ones.

Research by Princeton, Georgia Tech, Allen AI and IIT Delhi, presented at ACM SIGKDD 2024, analysed 10,000 searches and their accompanying AI answers. Content with statistics, cited sources and direct quotes increased visibility in AI search engines by up to 40%. Pages at position 5 that were optimised this way saw a visibility increase of 115%, while pages at position 1 showed barely any change.

The implication is not subtle: a well-written, well-structured page from an unknown brand can dominate an AI answer over a page from a market leader with thousands of backlinks. GEO democratises visibility for those who use the right structure. Schema.org markup with explicit author, publisher and date signals is decisive here: pages with attribute-rich structured data are cited at a citation rate of 61.7%, against 41.6% for pages with generic schema.

The question beneath the technology

Filling in Schema.org fields is work that a good developer finishes in an afternoon. Setting up your own newsletter takes a week. The real work lies underneath.

An AI does not cite a page that is a little bit about everything. It cites a brand that stands for something. That presupposes a choice: where are we authoritative, and what do we explicitly stay away from? That is not a content question, it is a positioning question. Anyone who has not yet answered that question is building GEO content on loose ground, and the AI smells it. The citations go to the brand that consistently publishes from one angle, not to the brand that is present in general.

Beneath that lies a second question, one that is older still. To whom are you indispensable, and who will happily walk off to someone else? A brand that has this clearly in focus writes pages that take a stance more easily. A brand that does not have it in focus writes pages that stay cautious, and cautious pages do not get cited. Schwung works from the inside out: first the position and the behaviour beneath it, then the imagery and the technology on top. In that order they hold each other up. The other way round, they both disappear.

What you can do on Monday

The practical step is not to overhaul your entire website. It is learning to think in two things at once: citable units, and a channel of your own that does not depend on Google.

Formats that AI search engines cite most often: definition pages (what exactly is X?), FAQ pages with schema markup, and articles with explicit, signed claims. Not the long, keyword-rich pages that SEO rewarded for years. Write every page with one clear claim in the first paragraph. Back every claim with a datable source. Add schema.org markup with author and publication date. And start with an inventory of your own channels: what do you own, what do you rent, and what happens if the landlord changes the rules?

The marketing manager from the first paragraph does not have an algorithm problem. He has a channel problem, a positioning problem, and a writing problem that presents itself as a technical question. Anyone who tackles those in that order does not disappear from view. Anyone who waits for the rankings to come back on their own does.