Google remains the largest environment for AI-influenced discovery, but ChatGPT has become the dominant source of measurable referral traffic from standalone large language model assistants. That split, highlighted in a new traffic study and reinforced by fresh UK regulation, matters because it shows how AI visibility is fragmenting into two different operating realities: exposure inside search experiences and trackable traffic arriving on-site.
For technology leaders managing growth, analytics, and platform risk, the takeaway is straightforward. AI discovery can now rise even when conventional referral reporting only tells part of the story. That has implications for AI Search strategy, attribution systems, publisher governance, and broader Enterprise AI operating models.
Google's advantage is discovery scale, not referral transparency
According to Previsible's third AI Traffic Study, reported by Marketing Tech News, AI discovery inside Google, including AI Overviews and AI Mode, represents a larger volume of AI-influenced traffic than standalone LLM platforms combined. The study covered November 2024 through May 2026 and analysed 6.77 million LLM-driven sessions across 166 websites using 166 Google Analytics 4 properties.
The critical measurement caveat is that Previsible excluded Google AI Overviews from its standalone LLM referral dataset. The reason was methodological: Google's AI search features do not produce trackable referral sessions in the same way as standalone LLM platforms. In practice, that means a company can gain AI-era visibility through Google without seeing equivalent evidence in traditional referral reports.
That distinction is easy to miss. Many executive dashboards still assume that discoverability and click-through traffic move together. In AI search, they increasingly do not.
ChatGPT dominates the measurable referral layer
While Google appears to lead where users discover brands inside AI-enhanced search, ChatGPT dominates the traffic that can actually be measured once users click through from a standalone assistant. Previsible said ChatGPT accounted for 92.4% of standalone AI referral traffic across the websites analysed, up from about 84% in its December 2025 review, according to Marketing Tech News.
The same study found that total monthly standalone LLM-referred sessions grew 9.9 times between November 2024 and May 2026, rising from 65,249 to 644,478. That growth suggests standalone assistant traffic is no longer a fringe signal. It is becoming a meaningful distribution channel, at least for some sectors.
But the market is also highly concentrated. With ChatGPT contributing such a large share of measurable LLM referrals, platform dependency risk is no longer theoretical. Enterprises that rely on AI-originated traffic now face a familiar digital-platform problem in a new form: one provider can shape the majority of visible demand.
The November 2025 drop shows how fragile AI referral flows can be
Previsible reported a 50% decline in monthly standalone LLM-referred sessions in November 2025, driven mainly by a fall in ChatGPT referrals. Those ChatGPT referrals dropped from 448,412 in October 2025 to 213,345 in November 2025, then recovered to 442,609 in December 2025. Previsible did not attribute the decline to a single confirmed cause.
That one-month shock is important for decision-makers because it exposes the volatility hidden inside AI traffic narratives. Even during a period of strong overall growth, referral flows can swing abruptly. If a business has started counting on assistant-originated traffic, conversion paths, lead volume, or commerce sessions, a sudden shift in prompt design, model behavior, ranking logic, or user interface could materially affect performance.
This is one reason AI discovery should not be folded into conventional SEO reporting without additional controls. It behaves more like a hybrid of search distribution, recommendation systems, and platform dependency.
Why This Matters to Technology decision-makers
Technology leaders should treat this development as a measurement and governance problem, not just a marketing trend.
1. Analytics gaps are now strategic gaps
If Google AI Overviews and AI Mode influence discovery but do not emit referral data in the same way as standalone assistants, then standard GA4-based reporting is incomplete by design. Teams may need new instrumentation, traffic classification rules, and blended visibility metrics to understand actual AI reach.
2. Platform concentration risk is rising
ChatGPT's 92.4% share of measurable standalone AI referrals means referral exposure is concentrated in one assistant. That affects contingency planning, partner dependency analysis, and experimentation roadmaps across Models and search-adjacent product teams.
3. AI visibility now crosses legal and content policy boundaries
The people responsible for SEO or growth can no longer manage AI discovery alone. Legal, content operations, engineering, and policy teams now need shared rules on what content may be used by AI systems, under what controls, and with which measurement expectations.
4. KPI design needs revision
Session growth, CTR, and classic referral share may all become weaker proxies for influence in AI-mediated journeys. Decision-makers should expect to combine direct traffic data with brand lift, assisted conversions, citation monitoring, and content usage controls.
The UK CMA has turned AI search governance into an operational issue
The regulatory backdrop is moving quickly. On 3 June 2026, the UK Competition and Markets Authority issued a binding conduct requirement for Google under the Digital Markets, Competition and Consumers Act 2024, according to TechHQ.
The CMA ordered Google to give publishers controls to withhold content from AI Overviews, AI Mode, and broader generative AI services including Gemini and Vertex AI at both directory and page level. It also required Google to publish clear explanations of how crawled content is used, provide engagement metrics to affected publishers, and ensure attribution includes followable links.
One of the most consequential elements is the anti-retaliation provision. Google cannot penalise publishers who use these controls by down-ranking them in regular search results. Google has nine months to implement the full changes, though the CMA expects meaningful controls before that deadline. TechHQ also reported that the CMA will wait at least 12 months before deciding whether Google should be required to negotiate licensing terms with publishers.
For enterprises, the significance is broader than publishing. The CMA has effectively signaled that AI search participation, attribution, and usage transparency are becoming governable product features. Any business with proprietary content, regulated information assets, or high-value knowledge libraries should expect similar questions in other markets.
Discovery growth and zero-click behavior are related, but not the same metric
The source bundle points to a structural tension. Previsible's study suggests Google's AI search surfaces generate more AI-influenced discovery volume than standalone LLMs. Separately, TechHQ reported that zero-click searches rose by close to 30% in categories such as health and local news after the full UK rollout of AI Overviews in late 2025.
Those figures should not be treated as directly comparable. One describes AI-influenced discovery volume; the other describes on-platform answer behavior that reduces outbound clicks. But together they indicate a common direction of travel: visibility can increase while traffic monetisation weakens.
That is especially relevant for ad-supported, affiliate, or lead-generation businesses. If AI systems answer more user queries inside the platform while sending fewer followable visits, existing monetisation models come under pressure. The result is a new trade-off between reach, attribution, and commercial control.
What enterprises should do next
First, separate AI discovery reporting from standard search reporting. If your dashboards only count referral sessions, they may systematically undervalue Google's role in AI-era discovery.
Second, build a platform-risk view of AI traffic. If ChatGPT supplies most measurable assistant referrals, scenario planning should test how product changes or referral volatility affect pipeline, revenue, and customer acquisition.
Third, establish content governance rules now. The CMA's approach suggests page-level and directory-level AI permissions, attribution rules, and usage transparency will become baseline operating requirements.
Fourth, align search, content, analytics, and engineering teams. AI discovery increasingly overlaps with Developer Tools, site instrumentation, structured content, and data rights management, not just editorial optimization.
Finally, track AI visibility as a distribution layer in its own right. The organizations that adapt fastest will be those that stop asking only where clicks came from and start asking where decision journeys began.
Sources and Methodology
This article was produced in multi-source mode using de-duplicated facts from two directly relevant reports on AI discovery and AI search governance. Primary sources used were Marketing Tech News on Previsible's AI Traffic Study and TechHQ on the UK CMA's Google AI search conduct requirement. Two additional source items in the input bundle concerned pharmaceutical AI drug discovery and were treated as out of scope for this article's search and referral focus.
