In recent months, what until recently was considered a hypothetical conversation has now become a tangible shift.
Click by click, the impact of artificial intelligence platforms on web traffic is no longer a matter of “if it will happen,” but of how much it is already occurring without many noticing. David Bell, from the SEO consulting firm Previsible, has documented this with data in a recent analysis published in Search Engine Land: between January and May 2025, sessions originating from assistants such as ChatGPT, Perplexity, Claude, Gemini, and Copilot rose from 17,076 to 107,100 across 19 Google Analytics 4 properties, an increase of 527% in just five months.
The phenomenon is not distributed evenly. Bell demonstrates that the sectors in which trust and accuracy are essential account for the majority of this new traffic. Legal, Finance, Health, Small Businesses, and Insurance now comprise 55% of the visits originating from language models. In some cases, such as in particular SaaS sites, more than 1% of all sessions (still a small number, but growing) are already arriving from these platforms.
This is no coincidence: users tend to ask AI complex questions and highly contextualized queries—the kind they previously reserved for a human specialist, such as contract preparation, medication compatibility, or payroll structuring.
In this new distribution, ChatGPT clearly leads, generating between 40% and 60% of AI-driven traffic across nearly all sectors. However, the landscape is becoming more diversified.
For example, Perplexity has established a strong presence in Finance, Legal, and Small Business; Copilot stands out especially in Finance and Legal; Gemini is beginning to gain traction in Insurance and Small Businesses; and Claude, though still with modest figures, is present in all verticals. For Bell, this diversity is a clear warning: optimizing content for a single platform is a risky strategy.
However, the fundamental change goes beyond the origin of the visits. The way language models find and display content does not follow Google’s rules. According to the report, there is no gradual crawling process, no fierce competition to secure a blue link on the first page, nor are there waiting periods for canonical tags to propagate. The criterion is immediate: if the content answers the question in a clear, reliable, and structured manner, the model can cite it instantly.
Bell describes this change as the beginning of an “era of instant surfacing”, in which an article, a guide, or even a product page can be discovered and utilized by AI before it rises in organic rankings:
“Content does not need to appear at the top of Google’s SERPs to be discovered. It must be clear, structured, and cited by the model, whether it is on a blog, a help document, a case study, or a knowledge base. And this implies that the old mindset—publish, wait, and hope that Google discovers it—is dangerously outdated.”
This term, “surfacing,” is not particularly common (at least it is not frequently heard) and refers to the moment when content is found and displayed by a platform (in this case, an LLM such as ChatGPT, Perplexity, or Gemini) in response to a user query. That is, in traditional SEO, the process for your content to “surface” (apologies for the term) involves several phases:
This flow can take days or weeks, and visibility depends on competing with other pages for a spot in the results. In contrast, in the era of instant surfacing (again, apologies) described by Bell, LLMs may display your content without waiting for it to be well ranked in Google. The concept is that if an LLM such as ChatGPT with web browsing capabilities, or Google’s Search Generative Experience (SGE), is trained or has access to your content, it can instantly “select” the relevant information. It does not need to wait for your page to “climb up” in the rankings.
This shift requires a fundamental reconsideration of visibility strategies. For Bell, the first step is to measure, even imperfectly, the traffic coming from these platforms: use tracking parameters, observe unusual spikes in direct traffic, and look for correlations with mentions in AI chats.
From there, content must be restructured to perform not only on human screens, but also within responses generated by a model: clarity, conciseness, well-defined sections, and formats that facilitate information extraction. The key is no longer to rank in a specific position, but to become the source a model chooses and displays. This principle applies across the entire funnel, from articles and case studies to product pages or support documentation.
The underlying change extends beyond the origin of visits. As discussed, the way language models find and present content does not adhere to Google’s traditional rules. This is where Bell’s picture intersects with the reality already visible in the ecosystem dominated by Google. The company, through its Vice President and Head of Search, Liz Reid, recently argued that its new AI-powered search experience is maintaining stable web traffic and even improving the “quality” of clicks. Nevertheless, the experience of creators and marketers does not always align: impressions may remain steady, but organic clicks are declining.
This tension produces an interesting contrast: while Google assures that its AI “highlights the web” and Previsible shows that LLMs can open new avenues of discovery, the fact remains that in both scenarios, the model of information access is moving away from traditional organic clicks. In Google, visibility no longer guarantees traffic; in LLMs, selection depends on factors such as clarity, structure, or authority as perceived by the model, not the usual SEO ranking.
It is clear that we are still adapting to the new era of generative AI search, and in the coming months, countless reports like this will be published. However, what Previsible’s study makes evident is that SEO is not disappearing, but rather splitting. One part will remain focused on positioning in traditional search engines; the other, increasingly significant, will depend on how AI models find and choose to display information.
And, as has happened with every prior rule change, those who adapt first will not only maintain their visibility, but may also conquer new ground before the rest of the competitors react.
Image: ChatGPT
Your email address will not be published. Required fields are marked *
Δ