Rand Fishkin, the co-founder and CEO of SparkToro, posted on LinkedIn on July 24, 2026, to say that Similarweb’s latest report would “most likely infuriate” two teams directly. The AI zealots who’re sure each advertising and marketing greenback ought to already be chasing chatbot visibility. And the AI skeptics who’ve suspected the hype was overblown however by no means had the receipts to show it to a boss affected by what Rand calls AI derangement syndrome.
Most trade studies consolation one camp and annoy the opposite. This one, Similarweb’s “2026 Generative AI Panorama: The Evolution of AI Search,” manages handy either side an information level that undercuts their certainty. I learn all 38 pages as a result of I’m into ground-truthing, and I wish to stroll by means of why Rand is true, what the report truly exhibits, and what it is best to do in a different way tomorrow morning due to it.
The Quantity That Ought to Fear The AI Zealots
Similarweb tracked viewers overlap between ChatGPT and Google between March and Could 2026, and 461 million of ChatGPT’s 494 million customers (95%) additionally use Google in the identical window. Virtually no person has left Google for ChatGPT. They’ve added ChatGPT to a Google behavior that hasn’t budged.
Zoom out additional, and the hole will get starker. Search nonetheless pulls 3.3 billion common month-to-month distinctive guests worldwide. AI chatbots, even after rising 57% 12 months over 12 months, sit at 655 million. So, search remains to be roughly 5 instances the scale of all the AI chatbot class mixed. In case your 2026 finances deck assumes AI search has already eclipsed conventional search, the mathematics on this report says in any other case.
Citations inform the identical story from a unique angle. Solely 6.8% of ChatGPT solutions within the U.S. included a hyperlink to an exterior supply as of Could 2026. That’s up greater than fivefold from round 1% a 12 months earlier, which is genuinely quick progress, nevertheless it additionally means 93 out of each 100 ChatGPT solutions nonetheless ship no person wherever. Ethan Smith of Graphite makes a sharper level within the report about how customers are folding those self same prompting habits again into Google itself, with common search size climbing steadily since AI Mode launched. Individuals aren’t abandoning search packing containers. They’re simply typing longer sentences into them.
The Quantity That Ought to Fear The AI Skeptics
Now for the half of the report that punctures the opposite aspect’s confidence. Common month-to-month internet visits throughout generative AI platforms hit 9.5 billion between June 2025 and Could 2026, up 70% 12 months over 12 months. App downloads worldwide climbed to 2.7 billion, up 134%. Half of all generative AI customers are actually 35 or older, in comparison with 61% beneath 34 simply two years in the past, which is the clearest sign I’ve seen that this isn’t a Gen Z fad operating out its pattern cycle. Michael Horrocks of Miro places it plainly within the report: “Development concentrated in youthful demographics can fade with traits; progress spreading into older generations is commonly what sturdy, mainstream adoption appears like.”
Meta AI’s personal disclosed numbers again that up from a very totally different angle. Publicly reported month-to-month lively customers went from 384 million in September 2024 to 1.2 billion by March 2026, greater than tripling in 18 months, completely by driving inside Instagram, Fb, WhatsApp, and Messenger quite than as a standalone vacation spot anybody needed to hunt down. And ChatGPT advert penetration within the U.S. jumped from 14% of desktop chats in Could 2026 to 26% only one month later. No matter you consider the maturity of AI search, advertisers clearly don’t assume it’s a toy.
My learn is that each camps are pattern-matching off the piece of the info that confirms what they already believed, and each are ignoring the half that complicates it. AI search hasn’t changed something. It has stacked a brand new, fast-growing, inconsistently distributed layer on high of a search ecosystem that was already there, and the practitioners who will win the subsequent two years are those measuring the stack as a substitute of arguing about which layer issues extra.
Why The Disconnect Between Citations And Clicks Issues Extra Than Both Camp Realizes
Probably the most helpful chart in the entire deck, and the one I believe will get underappreciated within the LinkedIn debate, comes from Aleyda Solis of Orainti. She factors out that 65% of the URLs ChatGPT cites sit two or three folders deep, the pages doing the precise evidentiary work behind an AI reply. However 58.8% of the referral visitors that AI sends again to websites lands on the homepage, not the cited web page in any respect. Cited pages and clicked pages are virtually completely totally different populations of URLs.
That single knowledge level ought to reorganize how companies report AI efficiency to purchasers. In the event you’re solely monitoring whether or not your deep product or weblog content material will get cited, you’re lacking the truth that the people who truly click on by means of are touchdown someplace else completely and wish their very own conversion path. Rand makes a associated level within the report itself, evaluating this to how Twentieth-century advertisers proved billboard and radio spend labored by measuring carry in retailer visits quite than counting who glanced on the signal. The mechanism has modified. The self-discipline of measuring downstream conduct as a substitute of floor impressions hasn’t.
3 Issues To Change In Your Technique This Week
1. Cut up your reporting into two separate metrics. Monitor quotation fee and quotation folder depth as one key efficiency indicator that measures whether or not AI trusts your content material sufficient to make use of it as proof. Monitor referral touchdown pages and downstream conversion as a very separate KPI that measures what truly occurs as soon as a human clicks by means of. Conflating the 2 in a single dashboard is how manufacturers miss each issues directly.
2. Cease treating “AI visibility” as a single class. Similarweb’s model visibility index exhibits how category-specific this already is. CeraVe leads magnificence at an index of 100 whereas NYX Cosmetics sits at 19 in the identical class. Kevin Indig of Development Memo argues within the report that share of voice is the metric that issues right here as a result of it’s a relative comparability in a stochastic system, not an absolute rating. Pull your individual class’s leaderboard earlier than you assume you’re profitable or shedding.
3. Match your content material to the platform’s precise viewers, not the platform’s general measurement. The affinity knowledge exhibits ChatGPT skews towards on a regular basis shoppers researching eating places, well being, and trend; Claude customers are 25 instances extra possible than the common searcher to go to college websites and skew closely towards professionals and college students, and Gemini customers over-index on graphics, safety, and {hardware} content material. A single piece of “AI-optimized” content material aimed toward all three is aimed toward none of them.
The Backside Line
I don’t assume the AI zealots are unsuitable that one thing structural is shifting. 9 and a half billion month-to-month visits and a fivefold leap in quotation charges in beneath a 12 months are usually not a rounding error. However I additionally don’t assume the skeptics are unsuitable that a lot of the trade’s AI panic is operating nicely forward of the particular visitors numbers, provided that Google nonetheless instructions 5 instances the viewers of each AI chatbot mixed and 95 out of each 100 ChatGPT customers by no means left Google within the first place. Rand’s submit nailed the discomfort as a result of the info refuses to let both aspect maintain its story easy. The manufacturers that can truly profit from this report aren’t those choosing a aspect. They’re those pulling the quotation and referral numbers for their very own class this quarter and constructing a method round what the info says quite than what the argument on LinkedIn says.
Extra Sources:
Featured Picture: Rawpixel.com/Shutterstock
