For the past two years, the conversation around generative search optimisation has focused almost entirely on model quality. Which LLM is most accurate? Which one hallucinates less? Which one cites the best sources?
Those are the wrong questions.
The real battle isn't being fought at the model level. It's being fought at the distribution layer. And the browser is where it's all going to be decided.
Jan 2025 to early 2026 [1]
Over the same period [1]
Gemini's surge is primarily due to Google embedding AI Mode directly into Search, a product used by billions. Interactions with AI overviews in Search are counted as Gemini sessions, often without users consciously choosing it over alternatives.
True adoption is better reflected in deliberate, intentional usage and enterprise deployments. Despite only 2% consumer traffic, Claude is reportedly winning in enterprise deployments, indicating where informed buyers are consciously making choices [2].
Beyond the distribution story, there's a meaningful technical difference between how different LLMs retrieve and ground their responses. And it has direct implications for GSO strategy.
Gemini grounds its responses using Google Search. That sounds like an advantage — Google has the largest, most current web index in the world. And for time-sensitive queries, it helps.
But traditional ranking signals remain heavily weighted toward authority and optimisation patterns rather than direct semantic accuracy. Domain authority. Backlinks. On-page SEO. Freshness.
That means Gemini's grounding inherits all of Google's ranking biases. Well-optimised but shallow content can surface over deeper, more authoritative material. High-authority but outdated pages can outrank more accurate newer sources. For niche technical topics — exactly the content enterprise technology brands need to win on — the most SEO-savvy page isn't always the most accurate one.
Contrast this with retrieval systems built on semantic search. Exa (formerly Metaphor) surfaces content based on meaning and relevance rather than PageRank signals [3]. Models or tools that use semantic retrieval find the genuinely best answer to a query, not the best-ranked page for related keywords. It's one reason Perplexity's citations often feel more directly on-point than Gemini's.
The implication for GSO is significant. Optimising for Gemini and optimising for Google Search are largely the same task. But optimising for Perplexity, Claude, or ChatGPT requires a different emphasis — semantic clarity, citation-worthiness, and genuine authority over keyword density and link profile.
While the browser wars play out at the distribution layer, something equally significant is happening within Google Search itself.
Total search volume is holding roughly steady. But the behaviour underneath that headline number has changed dramatically.
Drop in US Desktop Searches
Year over year in the US (Datos/SparkToro) [4]
Decline in US Organic Referrals
From Google Search year on year [5]
Google search traffic to publisher websites globally was down by a third in the year to November 2025, according to Chartbeat data [5]. Users aren't abandoning Google — they're just searching less often. Because AI Overviews are resolving their questions before a second or third search becomes necessary.
This is Google cannibalising its own referral model to defend against external AI competition. By answering queries within the SERP, AI Overviews keep users inside Google's ecosystem. It's a rational defensive move. But it fundamentally breaks the implicit contract that has underpinned content marketing and SEO for two decades.
A brand can rank in position one, earn an AI Overview citation, and still see clicks collapse — because the answer was delivered without requiring a visit to the site. Visibility no longer reliably translates to traffic.
This is precisely why GSO has moved from a nice-to-have to a strategic necessity. The traffic model that most enterprise content strategies were built on is structurally broken. The question is no longer whether to adapt. It's how fast.
Here's where the distribution story gets genuinely interesting.
If Google's moat is default access through Chrome, Android, and Search, the logical counter-move for every other AI lab is to own the browser layer themselves. And that's exactly what's happening.
Perplexity launched Comet — a full AI-native browser that replaces the URL bar with a prompt interface. Rather than competing for placement within Google's ecosystem, Comet intercepts queries at the point of intent. The user never touches a search engine.
OpenAI has been moving in the same direction, pushing harder into mobile and developing agent capabilities designed to live inside the browsing session.
Anthropic took a different route. Rather than building a standalone browser and asking people to switch, they launched Claude for Chrome — an extension that sits in a sidebar within Chrome itself [6]. It can see what you're looking at, take actions on your behalf, fill forms, click buttons, and maintain context across tabs and sessions. A smarter enterprise play: getting a company to deploy a Chrome extension is a far easier sell than migrating an entire workforce to a new browser.
Most brands think about AI visibility as a single problem. It isn't. It operates across four distinct layers — and each one requires a different strategic response.
Which LLM is processing and generating the response. GPT-4o. Claude 3.5. Gemini 3. Each reasons differently, weights sources differently, and has different training data biases.
How users access the model in the first place. Standalone apps, native Search integration, API-powered products. This is where Gemini's inflated market share numbers live — and where Google's structural advantage is most pronounced.
How the model finds information to ground its response. Google Search, Exa, Bing, internal RAG pipelines. The retrieval architecture determines what information the model has access to — and therefore sets a ceiling on response quality regardless of model sophistication.
The emerging frontier. Whoever owns the browser owns the session context, the query intent, and ultimately the citation opportunity. Comet, Claude for Chrome, and OpenAI's browser ambitions are all plays at this layer.
At Evolv, we track performance across all four layers separately. Because what's happening at the distribution layer — Gemini's AI Mode dominance — is completely different from what's happening at the retrieval layer, where semantic search systems increasingly determine who gets cited. Conflating the two leads to misallocated budgets and strategies optimised for the wrong signal.
Traditional SEO performance and generative visibility performance overlap — but they operate on fundamentally different mechanics. Treating them as the same measurement is where most enterprise strategies are currently getting it wrong.
The browser wars don't make GSO simpler. They make it more complex — and more important to get right.
Each AI platform already has meaningfully different retrieval behaviours, citation patterns, and ranking signals. Add proprietary browser layers with their own AI integrations on top and the surface area for optimisation expands significantly.
Structured, semantically clear, well-cited content travels well across retrieval mechanisms regardless of the underlying browser or model. Chasing the quirks of any single platform is a short-term play in a fast-moving landscape.
If Claude is winning enterprise with 2% consumer traffic, the people making deliberate, high-stakes decisions are disproportionately using Claude. For B2B technology brands, that's exactly the audience that matters.
Being cited in an AI overview in Google Search is closer to earning a featured snippet than being recommended by a chatbot. The intent profile of that user is different — earlier in the funnel, less actively researching, more passively consuming.
Most enterprise technology content is still optimised for traditional SEO signals. The brands that transition earliest to genuinely semantically strong content will have a structural advantage in retrieval systems that care about meaning over rank.
The browser is becoming the next battleground for AI dominance for the same reason Chrome mattered in 2010. Whoever controls the default interface controls the query.
Every major AI lab understands this. The race to own that layer is accelerating.
For GSO practitioners, the takeaway isn't to pick a platform and optimise for it. The fragmentation is structural and ongoing. The landscape will look materially different again in twelve months.
The labs that win the browser wars will likely win the distribution game. But the brands that win at content will get cited regardless of which browser their audience is using.
That's the bet worth making.
Most enterprise technology brands have no idea how they're performing across the four layers of the visibility stack — or where their biggest gaps are relative to competitors.
Evolv's Generative Visibility Audit maps your current citation performance across ChatGPT, Perplexity, Claude, and Gemini, benchmarks you against direct competitors, and identifies exactly where retrieval architecture changes or content strategy shifts will have the highest impact.
Similarweb Global AI Tracker, January 2026. Reported by Trending Topics EU — Google's Gemini eats into ChatGPT's market share, Grok overtakes Perplexity
xpert.digital — ChatGPT's Market Share Falls to 68 Percent as Gemini Closes the Gap
Evolv Agency — What is Exa: The Hidden Search Infrastructure Revolutionizing Generative Search Optimization
Datos / SparkToro Q4 State of Search Report (January 2026). Reported by Search Engine Land — Google searches per U.S. user fell nearly 20% YoY
Chartbeat data via Reuters Institute for the Study of Journalism. Covered by Press Gazette — Global publisher Google traffic dropped by a third in 2025
Anthropic — Piloting Claude in Chrome