AI search isn’t replacing Google. It’s stacking on top.

Rand Fishkin flagged the new Similarweb report last week with a line that sums up the entire state of the industry: it'll infuriate both camps. The AI zealots who've been telling clients to reallocate budget away from Google. And the AI sceptics who've been quietly hoping the whole thing blows over.

Neither gets to be right. Both get a data point that undermines their position. The report — read alongside iPullRank's 13-billion-search zero-click analysis, Ahrefs' AI detection data, and Marie Haynes' read on Google's earnings — paints a picture the industry's strategy decks haven't caught up to yet.

The picture: AI search hasn't replaced anything. It's layered on top. And the businesses budgeting for a replacement are going to find themselves defunding the layer that still does 95% of the work while chasing the layer that does 5%.

I want to walk through what the data actually says, why the industry framing is off, and what it means for how you should be spending your time and money between now and Christmas.

The one number the AI-first crowd needs to sit with

Similarweb tracked audience overlap between ChatGPT and Google from March to May 2026. Of ChatGPT's 494 million users, 461 million — 95% — also used Google in the same window.

Almost nobody has left Google for ChatGPT. They've added ChatGPT to a Google habit that hasn't budged.

Zoom out further and the gap widens. Search still pulls 3.3 billion average monthly unique visitors worldwide. AI chatbots, after growing 57% year over year, sit at 655 million. Search is roughly five times the size of the entire AI chatbot category combined.

If your 2026 budget deck assumes AI search has already eclipsed traditional search, the maths says otherwise. And it's not close.

Citations tell the same story from a different angle. Only 6.8% of ChatGPT answers in the US included a link to an external source as of May 2026. That's up more than fivefold from around 1% a year earlier, which is genuinely fast growth — but it also means 93 out of every 100 ChatGPT answers still send nobody anywhere.

That's the ceiling on organic ChatGPT traffic right now. Not "small but growing." Small, growing, and structurally capped by a product that mostly doesn't cite.

The one number the AI-sceptic crowd needs to sit with

Now the other half. Because the sceptics have been just as wrong.

Both camps have been pattern-matching off the piece of the data that confirms what they already believed.

Average monthly web visits across generative AI platforms hit 9.5 billion between June 2025 and May 2026, up 70% year over year. App downloads climbed to 2.7 billion, up 134%. Half of all generative AI users are now 35 or older. Two years ago, 61% were under 34.

That last stat is the one to pay attention to. Growth concentrated in younger demographics fades with trends. Growth spreading into older generations is what durable, mainstream adoption looks like. This isn't a Gen Z fad running out its cycle. It's what a technology looks like when it stops being novel and starts being infrastructure.

Meta AI's disclosed numbers back that up from a completely different angle. Monthly active users went from 384 million in September 2024 to 1.2 billion by March 2026 — more than tripling in 18 months, entirely by riding inside Instagram, Facebook, WhatsApp, and Messenger rather than as a standalone destination. And ChatGPT ad penetration in the US jumped from 14% of desktop chats in May 2026 to 26% one month later.

Whatever you think about the maturity of AI search, advertisers don't think it's a toy.

Both camps have been pattern-matching off the piece of the data that confirms what they already believed.

What "layered on top" actually means in practice

Here's the framing shift I think matters most. The industry has been arguing about which layer wins — traditional search or AI search. That's the wrong question because the answer is neither, individually, and both, together.

Ethan Smith of Graphite made a sharp observation in the Similarweb report: users are folding their ChatGPT prompting habits back into Google itself. Average Google search length has been climbing steadily since AI Mode launched. People aren't abandoning search boxes. They're typing longer sentences into them.

That's the tell. The behavioural shift isn't "people leaving Google for ChatGPT." It's "people using both, and letting each one train how they use the other." Long conversational queries on Google. Follow-up questions in ChatGPT. Comparison shopping across both. The user isn't picking a lane. They're using the whole road.

Which means treating AI search as a separate channel with its own strategy — the way most agencies are pitching GEO right now — misunderstands what's actually happening. It's not a new channel. It's a new mode of the same channel. The distinction matters because it changes what you invest in.

The measurement problem this creates for everyone

I've written before about the measurement collapse in AI search and this data makes it sharper. If users are moving fluidly between Google and ChatGPT within a single research task — reading an AI Overview, asking ChatGPT a follow-up, going back to Google to compare specifics, then finally clicking through to buy — attribution is fundamentally broken.

Your GA4 report will show the final click. It won't show the AI Overview that answered the pre-purchase question. It won't show the ChatGPT conversation that put you on the shortlist. It won't show the Reddit thread that Perplexity summarised.

iPullRank's zero-click analysis compounds the problem. Nearly half of Google searches end without a click — but the rate varies wildly by cohort, device, country, and query type. Some searches are effectively closed loops inside the SERP. Others still reliably generate clicks. Averages hide the variance, and the variance is where strategy actually lives.

You cannot allocate budget rationally on data that averages closed-loop informational queries with high-intent commercial ones.

The businesses I see doing well right now aren't the ones with the most sophisticated GEO tracking. They're the ones who've accepted that measurement is partial, defined the outcomes that actually matter (revenue, qualified enquiries, brand mentions in relevant contexts), and stopped trying to attribute every micro-interaction.

<div style="text-align: center;">— · —</div>

Where the "AI content ranks lower" data fits in

Alongside all this, Ahrefs released new data showing AI-heavy pages appear across every position in Google's top 10 — but pages its detector reads as mostly AI-written tend to rank lower than pages with less AI-generated text.

The correlation isn't proof of causation. Ryan Law, one of the report's co-authors, was careful to say it appears the quality tends to decrease as AI usage goes up, rather than Google directly punishing AI content. John Ozuysal put it well on LinkedIn: most AI content is single-shot prompts, paraphrasing what's already there, zero original insight. Your information gain score is zero when you're just reorganising what already ranks.

Multiply 1,000 by 0 and you get 0. That's not a Google problem. That's a strategy problem.

But this connects back to the layering point. If AI search adoption is real and growing — and the Similarweb data says it is — then the content bar keeps rising in both channels simultaneously. Google is rewarding depth and original insight. ChatGPT is citing brands and sources it has strong training-set signal for, which correlates heavily with editorial reputation and earned coverage.

The businesses that thought "we'll use AI to scale content production and win both channels" are running exactly the wrong play. Both channels are getting harder to game with mediocre content at the same time. The premium on genuinely useful, differentiated writing is going up, not down.

What this actually means for how you spend the next six months

If I'm advising a business between now and end of year, here's what the stacked-not-replaced reality changes.

You're not choosing between channels. You're choosing whether to invest in the foundation both channels sit on.

Stop reallocating away from Google. If your 2026 plan involved pulling budget from SEO or Google Ads to fund GEO tools and AI-specific agencies, pause. The user base is 5x larger on Google, the conversion behaviour is mature and measurable, and the AI overlay on Google (AI Overviews, AI Mode) is itself pulling from the same signals your existing SEO investment builds. You're not choosing between channels. You're choosing whether to invest in the foundation both channels sit on.

Stop treating AI search as a separate discipline. The tactics that make you visible in ChatGPT citations are the tactics that made you visible in Google's top 10: earned coverage, brand mentions across authoritative sources, structured content, technical hygiene, actual expertise. Anyone selling "GEO" as a separate service is selling SEO with a rebrand.

Start investing in the layer that compounds across both. Brand. It's the moat in AI search — models cite what they've heard of — and it's increasingly the moat in traditional search too, as Google leans harder on quality signals to filter the flood of AI-generated content. Every pound spent building brand recognition pays out in both channels simultaneously. Almost nothing else does.

Get honest about measurement. Your dashboards will underreport AI-influenced traffic for another 12–24 months minimum, probably longer. Build a lightweight qualitative check — ask sales-qualified leads how they found you, track branded search volume as a proxy for AI-driven consideration, monitor mentions in Reddit and industry publications that AI systems train on. It won't be precise. It'll be more accurate than pretending nothing is happening.

The honest limits of this argument

A few things this analysis doesn't cover, that could change the picture.

The Similarweb data is a snapshot ending May 2026. AI search adoption is moving fast enough that a six-month gap matters. If ChatGPT citation rates keep climbing at their current pace — from 1% to 6.8% in a year — a similar jump over the next 12 months puts them at meaningful referral-traffic scale. Not comparable to Google yet, but no longer a rounding error.

The 95% overlap figure is also a US-heavy read. Other markets look different. And Meta AI's ride-along growth inside Instagram and WhatsApp is a fundamentally different distribution model than ChatGPT — the addressable behaviour there is closer to "assistant embedded in existing habit" than "new destination for search-like tasks."

And the "brand is the moat" argument holds well right now, but if AI systems get better at extracting quality signals from content that doesn't have strong brand priors, the moat gets shallower. I don't see that happening in the next 18 months. I could be wrong.

The line that matters

The practitioners who will win the next two years are the ones measuring the stack, not arguing about which layer matters more. Google isn't dying. AI search isn't a fad. They're both real, both growing, both feeding off the same underlying signals of authority, quality, and brand recognition.

If you're currently in a strategy meeting where someone is proposing to fund AI search work by cutting SEO investment, you're in the wrong meeting. The correct meeting is the one where someone asks how to invest in the things that compound across both.

That meeting is less exciting. It doesn't have a shiny new acronym. But it's the one where the money actually gets made.