Where Is Search Going?

Last week I sat in on a webinar built around Similarweb's newly published 2026 Generative AI Landscape report, hosted by Laurie Naspe, Similarweb's Director of Market Insights, alongside SEO practitioners Lily Ray and Kevin Indig. The panel talked through what the data actually shows about how people search now, rather than what any one platform would like us to believe.

The figures below are drawn from the published report itself and the panellists' own public research, not just from notes taken on the day.

The market is fragmenting, not consolidating

AI platform traffic worldwide reached 9.5 billion monthly visits between June 2025 and May 2026, up 70% year on year. That's the headline growth number. The less comfortable one for anyone building a single-platform strategy: ChatGPT's own share of that traffic has slipped from roughly three-quarters a year ago to around half today, as Gemini and Claude both pick up meaningful ground.

ChatGPT still leads in most individual markets and its total audience keeps growing, so this isn't decline in the way a shrinking business experiences it. It's a market growing faster than any single platform can hold onto its slice of.

For brand visibility work, that means treating one AI platform's optimisation rules as good enough for all of them is a riskier bet than it was twelve months ago.

Different platforms, different jobs

People use each AI platform differently, and that should shape what "visibility" means on each one. Claude holds the longest average session length of any major AI assistant, consistent with it being used disproportionately for work tasks rather than quick lookups. Google's AI Mode and Gemini, by contrast, stay closer to short, quick-answer search behaviour. ChatGPT and Perplexity sit somewhere in between.

The practical takeaway: work out which platform your specific audience actually reaches for before assuming one AEO approach covers ChatGPT, Gemini and Claude equally. It probably doesn't.

Google is quietly becoming the AI answer

AI Overviews now appear on 43% of US Google searches, up from around 15% a year earlier. Over the same period, average Google query length crept up from around 3.3 to 3.5 words, with people increasingly typing full, conversational questions instead of short keyword strings. Visits to Google's AI Mode grew from 126 million to 279 million a month across the same window.

None of that looks accidental. Google is routing people from a standard search into an AI Overview and, from there, into a fuller AI Mode conversation, building a search habit that looks a lot more like talking to a chatbot than typing a keyword. Whether that becomes the default experience or settles at a plateau is still an open question, but the direction is clear enough to plan around.

Most AI answers still don't credit anyone

Only 6.8% of US ChatGPT answers carried a source citation as of May 2026, against roughly 1% a year earlier. Citation behaviour is moving fast, in other words, it just started from a very low base: in raw terms, the large majority of ChatGPT answers still send nobody anywhere.

Where citations do appear, they're often not where the traffic lands. Similarweb's data shows most cited URLs sitting two or three folders deep on a site, while most referral clicks from ChatGPT land on the homepage instead. Citations and clicks are effectively answering two different questions: which page proves you're a credible source, and which page a visitor actually arrives on. Both need to hold up on their own terms.

This is the reasoning behind the "core model optimisation matters more than citation-chasing" argument raised on the panel. With citation rates this low, trying to build visibility purely by getting individually cited in answers is a slower, shakier lever than making sure your brand and its content are genuinely well represented in what the models already know, not just what they retrieve live.

Trust still beats rank, but rank wins first

Kevin Indig's own H1 2026 research found that roughly three out of four people go with whichever option an AI recommends first. Position does most of the work. The one thing that reliably overrides it: if a brand the person already recognises and trusts appears anywhere on the shortlist, even second or third, they'll pick that brand over the AI's top pick.

That's a genuinely useful distinction for prioritising effort. Being first still matters more than anything else you can influence directly. But existing brand trust, built the slow way, is the one lever that can beat position when you're not.

What this means in practice

Pulling the threads together, here's what's worth acting on:

• Treat core model presence as the priority, not just citations. If AI answers mostly draw on what a model already knows rather than what it retrieves live, being genuinely present in training data matters as much as being crawlable.
• Run your own experiments rather than copying a published tactic wholesale. Optimisation mechanisms look broadly similar across platforms, but the weighting differs enough that a tactic proven on one model won't necessarily transfer to another.
• Keep talking to customers directly. Referral and citation data is getting noisier by the month, so qualitative signal from real conversations is filling gaps that platform-level metrics increasingly can't.
• Publish something genuinely original , and let real, named experts carry some of that authority, rather than relying on commodity content alone.
• Track broader, intent-led queries alongside exact-match keywords. That's closer to how AI Mode's multi-step retrieval actually works.

None of this replaces solid SEO fundamentals. Google's index still feeds AI Overviews, AI Mode and Gemini simultaneously, so a penalty in one place costs you everywhere else too. AEO is an extension of that work, not a separate discipline, something we went into in more detail in AEO for B2B: what it is and why it works differently.

If you'd like a second pair of eyes on where your own brand actually stands in AI search right now, get in touch