# 91% of extra visits after an AI recommendation carry no AI referral, Scrunch finds

Published: 2026-11-14T04:00:47.275Z · Source: Sitecore (Scrunch) (https://www.sitecore.com/company/newsroom/press-releases/2026/11/40-of-ai-assistant-sessions-never-touch-the-open-web-new-scrunch-study-finds)
Modified: 2026-11-14T16:48:26.000Z
Source date: 2026-11-12
Flags: Scrunch study — vendor-published; panel size and countries not disclosed, full report gated
Entities: chatgpt, perplexity, gemini

Scrunch, an AI-visibility company owned by Sitecore, published a study on Nov. 12 that linked an opt-in panel's sessions in ChatGPT, Perplexity, Gemini and Claude to the same people's web searches and page visits, mostly from July to September 2026. When an assistant recommended a brand to someone who had not engaged with it recently, that person became more likely to search for the brand and visit its site within a week, Scrunch says, and about 91% of that rise in visits arrived without an identifiable AI referral source, so it would appear as direct or organic traffic.

Scrunch also reports that 40.8% of AI-assistant sessions involved no web search before the conversation and no click after it, against 21.5% of traditional search sessions; that 77% of AI replies naming specific brands did so without the user asking for recommendations; and that words from a page someone had just visited were nine times more likely to appear in their next prompt. The release says the panel covers two countries but names neither and gives no panel size, and the full report is gated. Scrunch sells AI-visibility software.

How this fits: It points the same way as an earlier measurement taken from a different direction. When ChatGPT stopped passing a referrer (https://anythingengineoptimization.com/item/2026-10-14-chatgpt-removes-utm-tracking-from-its-outbound-links/) in October, Adobe saw ChatGPT-attributed visits on about 500 pages fall by roughly 31,000 a day while direct visits rose by about 26,000. Scrunch counts visits that come later, after a recommendation, which a referrer wouldn’t catch either way.

Why it matters: If most of the visits an AI recommendation causes land as direct or organic traffic, referral reports understate AI's effect on a site, and a channel judged only on its referrals will look smaller than it is; that is a vendor panel's estimate, not a measured attribution.

Editor's note (signed by the editor): I’d believe the direction before the number. A vendor panel of unstated size, in two unnamed countries, with the full report gated, can’t carry 91% on its own. But it points the same way as Adobe’s October data (https://anythingengineoptimization.com/item/2026-10-14-chatgpt-removes-utm-tracking-from-its-outbound-links/), and that’s the part worth acting on: don’t judge AI by its referral line alone.

## What this answers

**Do AI assistants send traceable traffic to websites?**

Mostly not, in Scrunch's panel: about 91% of the rise in site visits after an AI assistant recommended a brand arrived with no AI referral source, so it would show up as direct or organic traffic.

**How often do AI chatbot sessions end without a click?**

Scrunch found 40.8% of AI-assistant sessions had no web search before and no click after, against 21.5% of traditional search sessions.


Canonical: https://anythingengineoptimization.com/item/2026-11-14-91-of-extra-visits-after-an-ai-recommendation-carry-no-ai-referral/
From Anything Engine Optimization (AEO Wire) — https://anythingengineoptimization.com/ · Standards: https://anythingengineoptimization.com/standards/
