AI search is not replacing Google. That is the central finding of Similarweb’s “2026 Generative AI Landscape: The Evolution of AI Search,” a 38-page report covering audience and traffic data through May 2026. Rand Fishkin, co-founder and CEO of SparkToro, wrote on LinkedIn on July 24, 2026 that the data would “probably infuriate” two camps simultaneously: AI zealots and AI skeptics. He was right. Each side gets a data point that complicates the other’s argument.
Why haven’t AI chatbots pulled users away from Google?
The simplest answer: most people never chose between them. Similarweb tracked audience overlap between ChatGPT and Google from March to May 2026 and found that 461 million of ChatGPT’s 494 million users (95%) also used Google in the same window. Almost nobody replaced one with the other. They added ChatGPT on top of a Google habit that had not changed.
The scale gap confirms this picture. Traditional search averages 3.3 billion monthly unique visitors worldwide. All AI chatbot platforms combined, even after growing 57% year over year, reach 655 million. Search is still roughly five times larger than the entire AI chatbot category. As of May 2026, only 6.8% of ChatGPT answers in the U.S. included a link to an external source, up from approximately 1% a year earlier. That fivefold jump in citation rate (the share of AI answers that link out to a webpage) is genuine growth. But it also means 93 out of every 100 ChatGPT answers still send no traffic anywhere.
Ethan Smith of Graphite, quoted in the Similarweb report, notes that users are folding AI prompting behavior back into Google itself: average search query length has climbed steadily since Google AI Mode launched. People are not abandoning search boxes. They are typing longer sentences into them.
If AI search isn’t replacing Google, what does the real growth story look like?
Significant, and increasingly mainstream. Average monthly web visits across generative AI platforms hit 9.5 billion between June 2025 and May 2026, a 70% year-over-year increase, according to Similarweb. Worldwide app downloads reached 2.7 billion, up 134%. Half of all generative AI users are now 35 or older, compared to 61% under 34 just two years earlier.
Michael Horrocks of Miro puts the demographic shift plainly in the report: “Growth concentrated in younger demographics can fade with trends; growth spreading into older generations is often what durable, mainstream adoption looks like.”
Meta AI illustrates the scale of platform-driven adoption. The company’s publicly reported monthly active users rose from 384 million in September 2024 to 1.2 billion by March 2026, more than tripling in 18 months, without requiring users to seek out a dedicated AI destination. It grew entirely inside Instagram, Facebook, WhatsApp, and Messenger. ChatGPT ad penetration in the U.S. also jumped from 14% of desktop sessions in May 2026 to 26% just one month later, a signal that advertisers are treating the platform as a real channel rather than an experiment.
Why does the gap between cited pages and clicked pages matter?
This is the finding most likely to get underweighted in strategy discussions. Aleyda Solis of Orainti highlights in the Similarweb report that 65% of the URLs ChatGPT cites sit two or three folder levels deep in a site’s architecture, typically detailed product pages, blog posts, or evidence-heavy guides doing the actual evidentiary work behind an AI answer.
But 58.8% of the referral traffic that AI sends back to websites lands on the homepage, not on any cited page. The pages AI trusts as sources and the pages humans actually click through to are almost entirely different populations of URLs.
Rand Fishkin draws a parallel in the report to how 20th-century advertisers measured billboard and radio spend: not by counting who glanced at the sign, but by measuring lift in store visits afterward. The mechanism has changed. The discipline of measuring downstream behavior rather than surface impressions has not.
What this means for AI-search visibility
Here is the practical implication the Similarweb data points toward but does not spell out directly: most brands are currently optimizing for one signal when they need to manage two entirely separate ones. The cited-page-vs-clicked-page split is the clearest evidence yet that AI-search performance requires two measurement tracks, not one dashboard.
Think of it this way. Imagine a kitchen appliance brand whose product pages get cited in ChatGPT answers about blenders. That is a citation win. But if 58.8% of the humans who then click through land on the homepage, and the homepage has no clear path to the blender category, the brand earned the credibility signal without capturing the conversion opportunity. Conversely, a brand with a well-converting homepage but no deep content worth citing will receive some AI-referred traffic without ever being the source AI trusts when forming its answer.
The audience segmentation data adds another layer that complicates any single “AI optimization” approach. According to Similarweb’s affinity data, ChatGPT users skew toward everyday consumers researching restaurants, health, and fashion. Claude users are 25 times more likely than the average searcher to visit university sites, skewing heavily toward professionals and students. Gemini users over-index on graphics, security, and hardware content. A single piece of content aimed at all three platforms is effectively aimed at none of them.
Kevin Indig of Growth Memo argues in the report that share of voice (a brand’s relative share of AI citations compared to competitors in its category) is the metric that matters most here, because it is a relative comparison in a system where AI answers vary with every query. Similarweb’s brand visibility index shows how category-specific this already is: CeraVe leads beauty at an index of 100, while NYX Cosmetics sits at 19 in the same category. That gap likely reflects differences in citation depth and content authority, not just overall brand size. Pulling your own category’s leaderboard before assuming you are winning or losing is the analytical starting point the data suggests, not an absolute score in isolation.
Before you adjust your AI search strategy: 5 things to check first
- Pull citation data and referral landing page data separately, and confirm they are not the same URLs before drawing any conclusions about AI-search performance.
- Check which folder depth your most-cited pages sit at. Similarweb found 65% of ChatGPT citations go two or three levels deep, not to homepage or top-level category content.
- Map the conversion path from your homepage specifically, since that is where 58.8% of AI-referred visitors actually arrive, regardless of which page was cited.
- Identify which AI platform is most relevant to your category’s actual audience before producing platform-specific content. ChatGPT, Claude, and Gemini attract meaningfully different user profiles according to the Similarweb affinity data.
- Check your brand’s share of voice in your category rather than your absolute citation count. A relative comparison against competitors reflects competitive position more accurately in a system where AI answers vary by query.
The next indicator worth tracking: whether ChatGPT’s citation rate continues growing past its current 6.8% baseline, and whether the homepage-vs-cited-page referral split narrows as AI platforms experiment with deeper link attribution. Both trends were in early stages as of May 2026 and could shift the calculus on which signal matters more.
Lam Nguyen · Hingewise
Sources:
Similarweb, “2026 Generative AI Landscape: The Evolution of AI Search,” July 2026, via Search Engine Journal (https://www.searchenginejournal.com/ai-search-isnt-replacing-google-its-layering-on-top-similarweb-data/583378/)