Your analytics are lying: the downstream impact of AI visibility

Here is an uncomfortable possibility: AI is already sending you customers, and your analytics are quietly filing them under something else. A new behavioural study followed real people from an AI recommendation to the website they visited days later — and found that most of that traffic comes back through branded search, not a tidy “chatgpt.com” referral. If you are judging AI by the referral line in Google Analytics, you are judging it by the one number designed to miss it.

The study that connected the dots

SimilarWeb built the first study to trace an AI recommendation to a subsequent, real site visit using opted-in clickstream data rather than surveys or modelled estimates.1 Over six months of US desktop browsing (July–December 2025), across three industries — finance, travel and beauty — they watched what genuinely new users did in the seven days after an AI tool recommended a brand (excluding anyone who had already named that brand in their prompt).

Being the name the AI says is not a vanity metric. It changes what a real person does next. The brand that gets recommended gets the visit; the competitor mostly does not.

Why your dashboard can’t see it

Here is the twist that matters for anyone reading their own reports. When SimilarWeb broke down how the AI-influenced visits actually arrived, only a small slice came directly from an AI platform. The majority arrived through search — overwhelmingly branded search, where someone types the company’s name because an AI just mentioned it.

So the causal chain looks nothing like a normal referral. It looks like this:

The AI did the persuading. Search got the credit. Your dashboard tells you AI “isn’t driving traffic” — when really it can’t be seen.

And the visitors are worth more

The traffic AI sends is not just under-counted — it is, on these numbers, more engaged. AI-influenced visits ran to roughly twice the pageviews and twice the time on site of everything else.

That fits the behaviour: someone who arrives because a trusted assistant told them you are the answer is further along than someone idly clicking a blue link. The commerce data agrees — Adobe found that over the 2025 holiday season, AI-referred visitors converted more, not less, than other channels.4

This is why GEO needs SEO — and vice versa

If AI now shapes discovery, the obvious question is: where does AI get the names it recommends? The answer is reassuringly familiar. It reads the same web your SEO already targets.

ChatGPT’s search results lean heavily on Bing’s index: one analysis found 87% of SearchGPT citations matched Bing’s top organic results for the same query.5 Google’s AI Overviews, in the original large-scale study, drew 76% of their citations from pages already ranking in Google’s top ten.6 In other words, classic search visibility is the floor for AI visibility.

But the floor is not the ceiling. A follow-up found that overlap shrinking — only 38% of AI Overview citations now come from top-ten pages,7 which means ranking is becoming necessary but no longer sufficient. What closes the rest of the gap is off-site authority and on-page structure. Across 75,000 brands, Ahrefs found branded mentions correlated with AI visibility far more strongly than backlinks did8 — and a peer-reviewed study showed that GEO-specific tactics (clear, structured, statistic-bearing, answer-shaped content) lifted citation rates by up to 40%.9

So SEO and GEO are not competitors. They are one funnel. Get the fundamentals right — crawlable, indexed in Bing as well as Google, structured, and cited on the third-party sources AI trusts — and you become eligible. Layer GEO on top, and you turn “rankable” into “recommended”.

This is also why it isn’t a choice between an SEO tool and a GEO tool. The work overlaps: the same fixes — faster Core Web Vitals, clean titles and meta descriptions, valid structured data, a repaired sitemap, HTTPS and crawlable content — lift your classic Google ranking and your eligibility to be cited by AI. A single LoopCite scan grades both in one pass — together with your security and trust signals, your multilingual reach and your agent-readiness — so one report shows where your SEO, your trust and your AI visibility each stand, and every fix you ship pays off on both sides at once.

Discovery has already moved

None of this would matter if people still started every search in a search box. They don’t. When SimilarWeb asked where AI versus search was most useful across the buying journey, AI led at every stage — and led decisively at the very start, where new customers form their shortlist.

What this means for a South African business

A caveat worth stating plainly: the figures above are US panel data. We have not found equivalent published South African numbers — so treat them as directional, not local. But the mechanism is not country-specific. If your customers ask ChatGPT, Gemini or Copilot “who’s the best [your service] in [your city]”, the business that gets named gets the visit — and you will struggle to see it happening in your own analytics.

That is exactly the gap LoopCite measures. You can’t manage AI visibility from a referral report that can’t detect it. You measure it directly — by checking whether AI engines name and cite you, across every engine your customers use, and then fixing the readiness that makes you eligible.