Nobody Clicked, But They Stayed 4.5 Minutes: What Zero-Click Search Looks Like From Our Own Analytics

A company answers the same customer question for years, gathers real examples, publishes the piece. It ranks first on Google. The user arrives, finds the answer in the AI summary, and leaves without ever visiting the site.

This is no longer the exception — it's the average behavior. Ahrefs' December 2025 study, covering 300,000 search queries, found that when an AI Overview appears, the top-ranking result's click-through rate lands 58% below what would be expected without the summary. A separate Pew Research Center study of 900 US users confirms a similar pattern: users click a traditional result in 8% of AI-summary searches, versus 15% when no summary appears.

We're a GEO agency, and we didn't treat this as theoretical. This week we built our own measurement and looked at the numbers.

From our own data: AI traffic is small but behaves differently

GA4 session duration by channel, last 28 days:

Channel	Sessions	Avg. duration	Bounce rate
Direct	210	34s	87%
Organic Search	43	140s	58%
Paid Search	35	108s	63%
AI Assistant	33	267s (4.5 min)	52%

33 sessions from the AI Assistant channel (referrals via ChatGPT) average 267 seconds — eight times Direct traffic, twice organic search. Sessions landing on the homepage from ChatGPT specifically average 427 seconds (7 minutes).

That's a small sample and shouldn't be generalized on its own. But the direction is clear: the small number of visitors AI assistants send us aren't skimming.

But "AI visibility" isn't a single number

Here's where we need to split things apart. The same week, we measured our own AI visibility with Rankie — three engines, commercial-intent queries, margins of error included.

Engine	Queries	Mention rate	Margin of error
ChatGPT	249	11.2%	±3.9 pts
Gemini	379	0%	±0.5 pts
Google AI Overview	372	0%	±0.5 pts

We're mentioned in roughly one of nine ChatGPT answers. In Gemini and Google AI Overview — 751 queries combined — we're mentioned in zero.

Set this next to the GA4 session quality and the picture becomes clear: the small volume of quality traffic from ChatGPT exists because ChatGPT is already introducing us inside the answer — the visitor arrives already knowing the brand, curious rather than cold. In Gemini and Google AI Overview, that channel isn't open at all right now; there's no "small but quality" traffic there, because there's no traffic.

A single blended "AI visibility score" hides this split. Without breaking it down by engine, you can't know what's actually working on which platform.

Organic ranking and AI mentions are not the same thing

The second split is here. Per our Search Console data, we sit on page 3 organically for "geo consulting" (avg. position 29.8; 39 impressions, 0 clicks over 28 days) — the same term we pay roughly $1.30 per click for on Google Ads.

What's more striking: on some queries we rank on page one and still get zero clicks. A query shaped like "I heard about something called AI optimization, is there an agency that does that" — phrased the way you'd ask an AI assistant, not type into a search box — averages position 2.5, with 30 impressions and 0 clicks. We're surfacing inside Google's own AI Mode layer and not getting clicked. That's the article's thesis, reproduced in our own numbers.

So there are three separate things to measure, and they can move independently:

1. Organic ranking — where we show up in traditional Google
2. AI engine mentions — whether ChatGPT, Gemini, and Google AI Overview cite us when answering
3. Quality of AI-referred traffic — what that small number of visitors actually does once they arrive

Improving one doesn't automatically improve the others. In our case, organic ranking isn't bad, we're partially mentioned on ChatGPT, but we're entirely invisible in Google's own AI layer — three metrics, three different stories.

The same logic applies to LinkedIn

If the article's question holds for content marketing, it holds for LinkedIn posts too. Instead of linking out to the full piece, writing the core idea directly into the post — and leaving the link for whoever wants to go deeper — signals both "the idea got read" and "we have something to say on this." Decision-makers can read the post and get their question answered; people wanting detail click through.

What we're claiming, and what we're not

What we're claiming in this piece: the two numbers we measured ourselves (GA4 session quality, Rankie mention rate) are accurate and reported with their margins of error.

What we're not claiming: that this sample generalizes. 33 AI Assistant sessions and a single 28-day window is a starting point, not a settled trend. Ahrefs' and Pew's numbers come from large samples; ours doesn't — which is why we're not conflating the two.

Can someone who never visits your site tell what you're actually good at? The answer depends on the platform. You can't know without measuring.