For twenty years media buyers planned against one search surface. Similarweb's 2026 Generative AI Landscape report, published in July 2026, shows that surface has already split: ChatGPT's share of generative AI website visits fell from roughly 76% in June 2025 to about 53% by May 2026, while Gemini climbed from under 9% to around 27%. Attention fragmented in twelve months. The ad layer did not follow it.

  • Generative AI websites averaged 9.5 billion monthly visits between June 2025 and May 2026, up 70% year over year, but the growth is redistributing across engines rather than consolidating on one.
  • ChatGPT is the only engine selling ads at meaningful scale, so demand is concentrating on the exact surface where audience share is thinning.
  • Optimizing content for every engine is a production strategy: it gives a brand no control over how much reach it gets, or when it arrives.
  • This industry has solved fragmented supply once before, with an aggregation layer, not with better content operations.

What AI search fragmentation actually means

AI search fragmentation is the splitting of AI-mediated discovery across multiple engines, each with its own audience, citation behaviour and monetisation model. It is not audience decline. The category is compounding fast; what changed is that no single engine carries most of it any more.

That distinction decides who owns the problem internally. A shrinking channel is an SEO question. A channel that grows while splintering is a media question, and it lands on the buying side of the house.

One engine to rank on, half a dozen engines to be cited in

The open-web economy had one dominant index, one auction and one attribution model. Ranking was portable because there was only one place to rank, and reach was buyable because there was only one place to buy. Both conditions have now gone.

Similarweb's worldwide panel shows Claude moving from barely 2% of generative AI website visits to close to 9% in a year, the largest proportional gain of any platform tracked, while Meta AI and Google's AI Mode sit outside chatbot-domain measurement entirely. The closest historical analogue is not a Google algorithm update. It is display in the early 2000s: supply fragmented across thousands of sites, buyers could not transact site by site, and networks then exchanges aggregated the mess into something a media plan could actually purchase.

What Similarweb reported, and what we checked

The report puts generative AI websites at 9.5 billion average monthly visits, 655 million monthly unique visitors, up 57%, and 4.4 billion app downloads, up 58%. One number needs correcting in circulation: The Next Web's write-up placed AI Overviews on nearly four in ten US searches, while the report itself, as covered by Artificial Intelligence News on 29 July 2026, puts them at 43%, up from 15% a year earlier. We use the latter.

On advertising, Similarweb measured 26% of US desktop ChatGPT chats carrying an ad in June 2026, up from 14% in May, with 66.3% appearing after the second prompt and a click-through rate of 0.50%. Independent tracking by Cloro measured roughly 51% of US replies carrying an ad in the week to 3 July 2026. The panels measure different things. The direction is identical.

Why multi-engine optimisation is not a media plan

The strongest case for optimising everywhere at once

Aleyda Solis, who contributed to the report, argues that AI search cannot be measured with one KPI, one platform or one page type, and that brands should strengthen the deep pages feeding answers while making entry pages convert. Goodie's May 2026 traffic report makes the operational version of the same case: multi-surface by default, each engine with its own workflow. Both are right that authority is portable. One strong asset can be cited by several engines at once, and no ad budget is required to earn it.

Reach you cannot buy, pace or guarantee

Portable authority is not inventory. It has no throughput control: a brand cannot buy more of it, pace it across a launch window, or guarantee delivery in a quarter. Similarweb found only 6.8% of US ChatGPT answers included an external link as of May 2026, up more than fivefold in a year and still leaving roughly 93 answers in 100 that point nowhere at all.

The accounting breaks too. Similarweb reports 58.8% of AI referrals landing on homepages while the pages ChatGPT cites sit two or three folders deep. The asset that wins the citation is not the asset that receives the visit, which is precisely the seam a last-click media plan cannot measure across.

Discovery is no longer tied to a single destination.

— Baruch Toledano, Similarweb

What this means for brands and for publishers

For CMOs, media buyers and agencies: plan cross-engine reach, not per-engine tactics

Buying ChatGPT ads is a line item, not an AI strategy, and a 0.50% click-through rate is the tell: this is a mention business, not a click business. The report found users visited an AI-recommended brand two to four times as often as a rival that was not recommended. Plan for share of citation across the engines your category actually gets answered in, and hold the per-engine ad products to the share of attention they genuinely represent.

For publishers: fragmentation is leverage only once it is priced

News was the only major web category to shrink over the year, down 5%, while AI chatbots grew 57%. Six engines are harder for any one platform to dictate terms across than one, and that is real negotiating leverage. But leverage expires unpriced. The move is to sell citation exposure as inventory across engines, rather than wait for a single licensing counterparty to set the market.

Three things that decide the next eighteen months

Watch three signals. Whether Google opens native paid inventory inside AI Mode, which would reconsolidate demand overnight. Whether the fast-growing second tier, Claude, Perplexity, Grok, ships any ad product at all, or stays unmonetised attention. And whether the ageing audience pulls mainstream budget in: under-35s fell from 61% of generative AI users in May 2024 to 50% by May 2026, with growth shifting to the over-45s. Mainstream demographics are what make a surface buyable for large advertisers.

Conclusion

Hold on to this: AI search fragmentation is a reach problem, and reach problems in this industry have never been solved by producing better content, only by building a layer that aggregates fragmented supply into something buyable. That layer is Generative Engine Advertising, native advertising built for AI answers that gives brands cross-engine citation reach and pays the publishers whose content those answers depend on. What to watch next: whether Google opens paid inventory inside AI Mode, because the day it does, every buyer will discover how little of the category one engine now represents.