AI Search: 3 Similarweb Figures That Change Your Traffic Strategy

What Similarweb just proved—and why it changes everything

A client calls me on a Tuesday morning. He invested $8,000 in an « AI-first » program. Articles reformatted for chatbots. Citations in third-party databases. Schema markup everywhere.

Result after 6 months? 47 sessions from ChatGPT. 47. For $8,000.

His mistake wasn’t the budget. It was the angle.

Similarweb published a 38-page report in July 2026 that debunks two beliefs. And if you chase AI traffic without these figures in mind, you’ll make the exact same mistake my client did.

I’ll give you the three numbers. And more importantly, how to turn them into real strategy.

95% of outbound clicks: the scarcity that creates value

95%. That’s the percentage of clicks leaving ChatGPT toward a tiny group of sites. According to Similarweb, AI responses direct traffic to a handful of ultra-authoritative domains. Wikipedia, a few U.S. media outlets, GitHub, Amazon.

In other words, AI traffic doesn’t scatter. It concentrates.

On the surface, that’s discouraging. If you’re not in that club, you assume the game is rigged.

I see it the opposite way. The higher the concentration, the higher the value of each seat. That’s the scarcity bias from the DOSE framework that Guillaume Attias teaches at BMO Academy. The brain perceives what’s rare as valuable. And in AI search, viable citation positions are extremely rare.

My client didn’t grasp that. He thought a « spray and pray » approach would work. Instead, you need to target one referring domain and one query type. Become the AI’s preferred source for one very specific topic.

« Perceived scarcity triggers an automatic engagement response. Fewer seats mean more value for whoever occupies them. » — Guillaume Attias, DOSE training

AI traffic is 10 seats in a concert hall. Not 100,000 festival tickets. Ask yourself instead: « For which query do I have a legitimate shot at one of those 10 seats? »

6.8% of responses include a link: the invisible funnel

Second figure. In May 2026, only 6.8% of ChatGPT responses in the U.S. contained a clickable link to an external source. Plain and simple: 93 times out of 100, the user reads the answer and stops. No click.

A year ago, that rate was around 1%. Growth is real. But the vast majority of responses lead nowhere for websites.

I see too many e-commerce teams treat AI traffic as a natural extension of SEO. As if every generative response were a Google results page by another name. It’s not.

AI traffic is a two-stage funnel:

• The AI must choose to cite you (scarcity).
• The user must choose to click (intent).

This double filter completely changes content logic. Content Google rewards for a commercial keyword won’t necessarily be what ChatGPT cites for an advice or comparison question.

I see a pattern among my clients who win on this channel: they create formats the AI summarizes but doesn’t replace. Dense comparison tables. Guides with structured lists. Scored reviews. Content the AI can’t compress into two sentences without losing utility. Where the click becomes essential to access the full data.

If your content fits entirely in the AI response without a click, you won’t get traffic. Period.

Google weighs 5 times more: AI is a layer, not a competitor

3.3 billion monthly unique visitors for classic search. 655 million for all AI chatbots combined, growing 57% year-over-year but still five times smaller. That Similarweb figure puts things in perspective.

And crucially, 461 million of ChatGPT’s 494 million users also use Google in the same period. 95% of ChatGPT’s audience hasn’t left Google. They added ChatGPT on top.

AI doesn’t cannibalize search. It layers on it.

For an e-commerce business, it’s simple: your Google traffic stays your backbone. AI traffic is an additional, highly qualified but low-volume layer. You don’t replace SEO with an AI strategy. You add a stratum that uses the same signals: domain authority, brand mentions, presence in primary sources the models use.

One of my clients, an appliance parts specialist, worried the AI would steal his « how to repair » visitors. After 8 months of tracking, his Google traffic stayed flat. But the 1,200 additional sessions from Perplexity and ChatGPT Search represented an average basket 3.1 times higher than his site average. Expensive traffic, but very profitable.

The lesson: don’t cut your SEO budget to follow AI. Add to it intelligently.

3 actions to turn these figures into real traffic

Here’s what I do for clients who want to capture this channel without burning through their budget.

1. Identify the 5 queries where AI already cites a competitor.
Use an appearance tracker (I like Marie Haynes’s tool or manual alerts on ChatGPT, Perplexity, Google AI Overviews). Spot responses that contain an outbound link. If a competitor is cited, study their format: headline, structure, data. Often you can beat them with fresher, richer content.

2. Create anti-summary content.
Long-form with structured data the AI can’t absorb whole: tables, numbered lists with figures, detailed comparisons. The goal: the AI response says « Here are the three main criteria. For the full comparison, see this source. » And adds a link. That’s the « mandatory click » mechanism. I always apply it to my technical guide pages.

3. Build hyper-focused topical authority.
Generative AIs aggregate sources. They trust the most specialized sites. Don’t aim for 40 topics. Become the reference on one precise subject. One of my clients doubled his ChatGPT citations in 14 weeks by concentrating effort on a single product category, with a dedicated semantic cluster.

AI traffic is like a press mention in a publication that only prints 10 copies but gives them to the 10 best buyers in the world. Concentrated. Rare.

The real signal isn’t volume—it’s concentration

Many marketing leaders look at global volumes: 655 million visitors, that sounds huge. But in reality, almost all that traffic never clicks to any site.

The only thing that matters is the slice of that traffic that lands on a product page, an advice sheet, a buying guide.

The figure that hit me hardest in the Similarweb report is the vast white space between the 95% of concentrated clicks and the 6.8% of responses with links.

AI generates billions of responses. Only a tiny fraction generates external traffic. And that fraction is locked by a handful of players.

Looks closed. But it’s the best news for an agile e-commerce team.

Why? The barriers are visible. You know exactly who the 10 dominant domains are. You can analyze their formats, identify queries where AI cites a source without leaving a link, and become the one who provides the linked answer. You see the gaps.

With my clients, I don’t aim for 10,000 AI sessions per month. I aim for 120 ultra-qualified sessions on a high-intent query. With a conversion rate 3 to 5 times higher than standard Google traffic. It’s a margin channel, not a volume one.

If you have mature SEO flow, AI traffic is a profitable bonus. If your SEO is weak, focus on Google first. AI will follow after, as a finishing layer.

You’re closer to AI traffic than you think

The promise of AI search isn’t dead. It’s just more selective than advertised.

Six months after adjusting strategy, my client who spent $8,000 for 47 sessions got 1,430 AI sessions the next quarter. Not by blanketing everywhere. By occupying one single citation position on a query he already dominated in Google. The AI layer added naturally on top of his existing authority.

94% of his responses still had no link. But the 6% that did—he was in them. And that 6% was worth 24% of his quarter’s additional revenue.

How many pages on your site have a real chance to become the AI’s unique reference on their topic?

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn