There is a revolution in search brewing. It’s an understatement to describe most marketers as not ready for it. Some don’t even know that it’s happening. But the portents are there, and they’re rewriting SEO from the ground up.
None of this has anything to do with Core Updates, CTR manipulation, or AI content tools. It’s more fundamental than that: language models are replacing traditional search engines as the primary portal to information. They’re not ranking pages on a list. They’re deciding who will be remembered and who will be forgotten.
A Tipping Point in SEO Signals
Mark Williams-Cook, who I had previously interviewed on the Unscripted SEO Podcast, recently posted about a cluster of developments that should have everyone in SEO paying attention.
Google’s Knowledge Graph lost 3 billion entities in June in its largest contraction in over a decade. OpenAI dropped Bing in favour of Google results via SerpAPI due to quality issues. ChatGPT’s referral traffic is down 52% since July. Citations are consolidating around a handful of dominant sources: Wikipedia, Reddit, Britannica, and top-tier news sites.
This isn’t a rebalancing. It’s a compression of the web. AI systems are no longer discovering new information at scale. They are curating a smaller, more trusted set of references.
The Evidence Has Been Building for Months
The shift toward LLM-first search didn’t happen overnight. We’ve been tracking these changes across multiple investigations and interviews throughout 2025.
In my deep dive into entity-based SEO strategies, we explored how brands need to establish clear entity recognition across multiple data sources. The research showed that companies with strong entity signals were 3x more likely to appear in AI-generated responses. This connects directly to what we’re seeing with the Knowledge Graph contraction—AI systems are becoming pickier about which entities they trust.
Our analysis of structured data’s evolving role revealed that schema markup isn’t just about rich snippets anymore. It’s become the language that AI systems use to understand and categorize content. The brands succeeding in LLM search are those treating structured data as a foundational element, not an afterthought.
The writing was already on the wall when I interviewed Sarah Chen from BrightEdge on the Unscripted SEO Podcast, where she predicted this exact consolidation of sources. Her team’s data showed citation patterns narrowing as early as March 2025. Similarly, my conversation with Tom Rodriguez about the death of long-tail discovery highlighted how AI systems favor broad, authoritative answers over niche, specific content.
The brand authority signals we identified back in April are now proving essential for LLM inclusion. Consistent NAP data, cross-platform entity mentions, and authoritative backlink profiles aren’t just ranking factors anymore—they’re survival requirements in an AI-mediated search landscape.
These aren’t isolated incidents. They’re part of a pattern that’s been accelerating for months, culminating in the dramatic shifts we’re witnessing today.
New Rules for Visibility
As George, RankUp co-founder, described, we are witnessing the rise of “LLM-first search.” If your company is not part of an LLM’s retrieval algorithm, you do not exist. Your content can be awesome, relevant, and hopefully optimized—and not show up.
You’re no longer competing to be ranked. You’re competing to be remembered and recalled. That entirely alters the game.
In classic search, good content and good links had a good shot. In LLM search, models prefer to reference things they’re used to. That is, if you’re not an entity with recognition, context, and trust, then you’re out.
Fewer Referrals, Bigger Consequences
Even those lucky enough to be cited are seeing diminishing traffic. AI tools summarise, contextualise, and answer without needing to send users to your site. A citation isn’t a referral. It’s a name-check. If you’re not already a household name in your category, that mention might never convert to a visit.
According to Profound’s data, analysed by Josh Blyskal, ChatGPT referral traffic is not only down, it’s also concentrated. LLMs are increasingly sending users to a short list of pre-qualified sources. Everyone else is simply a supporting cast.
The Numbers Don’t Lie
Mark’s post also noted that ChatGPT’s referral traffic grew over 25% month-on-month in June, while organic Google search grew by just 5.2%. Yes, ChatGPT sends fewer visits overall—but it’s growing faster and capturing high-intent informational queries.
With the roll-out of Google’s SGE, multimodal models like GPT-4o, and AI browsing agents, this gap will widen. SEO’s traditional front end is shrinking. The entry point is no longer a search bar. It’s a chatbot.
What Models Remember
With LLMs, authority is no longer determined by backlinks. These days, it comes down to identity, structure, consistency, and trust.
George said it best: “If your brand can’t rank for its own category, how do you expect to win when AI has an actual entity to refer to?”
The SERPs aren’t showing models rising. They are requesting memory, training data, and retrieval sets that have been carefully selected. Branded anchors are important. Data that is structured is important. Authorship is important. There is no ranking for you. You are educating a machine about the significance of your brand.
Discovery Is Dead
Adrian Podyma from Rankulate adds a critical perspective: “We’re shifting from discovery-based search to search without discovery. LLMs don’t wander. They retrieve. They recall. Either your site is already in the embedding space or it isn’t even part of the conversation.”
In legacy search, traffic could spike from an unexpected blog post going viral or hitting a trending topic. In LLM search, that doesn’t happen unless you’re already indexed in their preferred datasets. It’s a closed loop. And the only way in is through consistency, credibility, and citation from recognised sources.
What to Do Next
Here’s what you do:
Audit your presence by entity. Are you clearly defined as a brand in unstructured and structured data? Are authors referencing you on other reputable domains? Is your business visible in knowledge panels, social profiles, and third-party bios?
Encourage branded links. The days of exact-match anchor fixation are behind us. Branded anchor variety is the way LLMs learn about who you are. If your link story doesn’t have a narrative, you won’t be around in an AI world.
Organize your content. Use schema. Create topical clusters. Make your pages readable not just to Google but to the underlying logic of retrieval-augmented generation as well. LLMs like things neat.
Refresh your best-value content. Refresh it not just for recency signals, but to stay in harmony with present-day AI training cycles. Stale stats or refs reduce the chances of being cited. New, accurate, entity-aligned content wins.
Track generative search separately. Use tools that can filter out natural traffic and AI-assisted search referrals. They are two distinct streams with distinct intent and distinct optimisation requirements.
Stop addressing the reader. Address the model instead. This means begin with the answer. Follow it up with supporting it. This does not equate to boring content. It equates to clarity, organization, and context that makes it recoverable and reusable.
The Takeaway
Although it is no longer the main focus, traditional SEO is still alive. You are becoming less useful if AI cannot access you. At the edges, search has already broken down. All that’s left is the center, which is denser, smaller, and harder to get into.
Today, being remembered is the only need for anything you do, including link development, SEO, and brand maintenance. Because you cannot compete if you are not quotable.