Opinions expressed by Entrepreneur contributors are their own.
Key Takeaways
- Model swaps are the new algorithm updates — unannounced and arriving on multiple platforms at once — and the only durable defense is content built on verifiable claims and named expertise.
- “AI search” is no longer one thing: only 11% of domains are cited by both ChatGPT and Perplexity, so each engine has to be treated as its own ecosystem.
On January 27, 2026, the visibility of thousands of businesses changed overnight — and almost nobody noticed why. That day, Google quietly swapped a new model, Gemini 3, into AI Overviews and AI Mode. There was no “core update” announcement and no warning to site owners.
Yet according to an analysis of the aftermath, the number of cited sources per AI answer jumped by roughly a third, freshness suddenly carried more weight and entity-rich websites gained share at the expense of thinner ones. ChatGPT, running on an entirely separate pipeline, was completely unaffected.
I have come to think of this event as the template for the second half of 2026. The first half made AI search official. Google published its first optimization documentation. Marketers began allocating more budget to AI search than to traditional SEO. The discipline of generative engine optimization moved from experiment to expectation.
The second half will be defined by something less visible but more consequential: the engines themselves are changing underneath us.
Model upgrades are the new algorithm updates
For two decades, marketers learned to brace for Google’s algorithm updates. The AI-era equivalent is the model swap — and unlike Google’s updates, these arrive unannounced, undocumented and on multiple platforms at once.
The mechanics explain why each one matters so much. Modern AI search systems do not process your question as a single query. They fan it out into many parallel sub-queries — often eight to twelve, and in ChatGPT’s case up to twenty — retrieve sources for each, verify claims and synthesize an answer.
As models become more capable, this process grows more thorough and, crucially, less gameable. Research already shows that only a quarter to a third of AI citations come from pages ranking in the traditional top ten. The newest models reason more, check more and trust selectively.
In our own client work, we observed this firsthand around the release of GPT-5.4: noticeably increased volatility in AI recommendations across accounts. Not necessarily steep drops, but a constant reshuffling that would have been unthinkable in the comparatively stable world of classic search rankings.
With Gemini 3.5 Pro arriving and further flagship releases expected from every major lab before year’s end, businesses should plan for several more of these invisible resets in the second half. The only durable hedge is content that survives machine scrutiny: verifiable claims, named expertise and consistent factual signals about who you are and what you do.
Three platforms, three different games
The second thing to understand about the coming months is that “AI search” is no longer one thing. The three dominant assistants are diverging into fundamentally different strategies.
ChatGPT is doubling down on personalization and monetization: its memory features are maturing rapidly, and its advertising pilot is expanding internationally to the UK, Mexico, Brazil, Japan and South Korea. Gemini is fusing Google’s retrieval and trust infrastructure with in-chat commerce, letting users complete purchases without ever leaving the conversation. Claude, by contrast, has positioned itself as the ad-free option focused on professional and agent-driven work.
The consequence is measurable: one large-scale citation study found that only 11% of domains are cited by both ChatGPT and Perplexity, and brand recommendations can differ by 40 to 60% across platforms for identical queries. Each engine is its own ecosystem, with its own biases and blind spots.
What makes this genuinely workable, however, is that the models will often tell you about those blind spots — if you ask. When we run AI visibility audits, we routinely ask the models directly why a client was not included in a recommendation.
These explanations should be taken with a grain of salt, since we cannot rule out that they are post-hoc rationalizations. But they frequently surface actionable insights.
One example: a dropshipping platform we work with was being recommended heavily by Gemini as a top option, yet had vanished entirely from certain ChatGPT recommendations — despite having been ChatGPT’s number-one pick for the same prompts just two months earlier. When we asked why, ChatGPT explained that the platform was not known for working well with Shopify, even though our prompt had never mentioned Shopify at all.
The model had silently made ecosystem compatibility part of its decision. After the client published substantial content addressing Shopify integration specifically, they reappeared in those recommendations.
That is the texture of GEO in late 2026: less about rankings, more about understanding — and correcting — what each model believes about you.
The personalization endgame
There is a deeper shift hiding inside the memory race. As ChatGPT, Gemini and Claude all build systems that remember individual users — their preferences, their history, their context — two people asking the identical question will increasingly receive different recommendations.
I argued recently that AI visibility is a winner-takes-all game, because most users simply accept an assistant’s initial recommendation rather than browsing alternatives. Personalization extends that logic to its conclusion: the contest becomes winner-takes-all per user.
A brand that wins the early interactions with a customer’s assistant gets reinforced within that relationship, query after query, while competitors become progressively harder to surface. It also means third-party visibility tools, which track generic prompts from anonymous accounts, will capture an ever-smaller slice of reality.
Expect measurement to get harder in the second half, not easier — and expect the premium on being a customer’s first AI-recommended choice to keep rising.
The web starts charging admission
The final trend on the horizon concerns the infrastructure beneath all of this. Publishers and infrastructure providers are erecting toll booths. Cloudflare now blocks declared AI crawlers by default and offers a pay-per-crawl model, millions of sites have opted out of AI training, and licensing intermediaries are signing up mid-sized publishers.
Every business now faces a strategic question that did not exist two years ago: open your content to AI systems and compete for citations, or block them and protect your work at the cost of invisibility.
My view — informed by having sat on the publisher side of the table as well as the marketer’s — is that history overwhelmingly favors staying open.
When Spotify effectively killed CD revenues, the music industry did not die. It restructured. Artists today earn far more from live events than their predecessors did, and smaller acts can build an audience and income through self-publishing at a speed that was impossible in the label-gatekeeper era.
AI will impose a similar restructuring on many industries, and not all of it will be comfortable — but businesses that withdraw from the new distribution layer to protect old revenue lines have rarely ended up on the winning side of such transitions. Adaptation, not retreat, is the historical pattern.
One practical aside: despite the hype, the llms.txt file — often sold as a quick AI visibility fix — is still used by no major AI provider in production, and Google has said on record it does not support it. For now at least, you can spend your energy elsewhere.
Key Takeaways
- Model swaps are the new algorithm updates — unannounced and arriving on multiple platforms at once — and the only durable defense is content built on verifiable claims and named expertise.
- “AI search” is no longer one thing: only 11% of domains are cited by both ChatGPT and Perplexity, so each engine has to be treated as its own ecosystem.
On January 27, 2026, the visibility of thousands of businesses changed overnight — and almost nobody noticed why. That day, Google quietly swapped a new model, Gemini 3, into AI Overviews and AI Mode. There was no “core update” announcement and no warning to site owners.
Yet according to an analysis of the aftermath, the number of cited sources per AI answer jumped by roughly a third, freshness suddenly carried more weight and entity-rich websites gained share at the expense of thinner ones. ChatGPT, running on an entirely separate pipeline, was completely unaffected.
I have come to think of this event as the template for the second half of 2026. The first half made AI search official. Google published its first optimization documentation. Marketers began allocating more budget to AI search than to traditional SEO. The discipline of generative engine optimization moved from experiment to expectation.
