It’s the end of search as we know it, and marketers feel fine. Sort of.
我们所知的搜索结束了,市场营销人员感觉还不错。算是吧。
For over two decades, SEO was the default playbook for visibility online. It spawned an entire industry of keyword stuffers, backlink brokers, content optimizers, and auditing tools, along with the professionals and agencies to operate them. But in 2025, search has been shifting away from traditional browsers toward LLM platforms. With Apple’s announcement that AI-native search engines like Perplexity and Claude will be built into Safari, Google’s distribution chokehold is in question. The foundation of the $80 billion+ SEO market just cracked.
在过去的二十多年里,SEO 一直是在线可见性的默认操作手册。它催生了一个完整的行业,包括关键词填充者、反向链接经纪人、内容优化师和审计工具,以及运营这些工具的专业人士和机构。但在 2025 年,搜索正逐渐从传统浏览器转向 LLM 平台。随着苹果宣布将像 Perplexity 和 Claude 这样的 AI 原生搜索引擎集成到 Safari 中,谷歌的分发控制地位受到质疑。价值超过 800 亿美元的 SEO 市场基础刚刚出现裂痕。
A new paradigm is emerging, one driven not by page rank, but by language models. We’re entering Act II of search: Generative Engine Optimization (GEO).
一个新的范式正在出现,它不是由页面排名驱动,而是由语言模型驱动。我们正进入搜索的第二幕:生成引擎优化(GEO)。
Traditional search was built on links. GEO is built on language.
传统搜索是建立在链接之上的。GEO 是建立在语言之上的。
In the SEO era, visibility meant ranking high on a results page. Page ranks were determined by indexing sites based on keyword matching, content depth and breadth, backlinks, user experience engagement, and more. Today, with LLMs like GPT-4o, Gemini, and Claude acting as the interface for how people find information, visibility means showing up directly in the answer itself, rather than ranking high on the results page.
在 SEO 时代,能见度意味着在结果页面上排名靠前。页面排名是通过基于关键词匹配、内容深度和广度、反向链接、用户体验参与度等对网站进行索引来确定的。如今,随着 LLMs(如 GPT-4o、Gemini 和 Claude)作为人们获取信息的接口,能见度意味着直接出现在答案中,而不是在结果页面上排名靠前。
As the format of the answers changes, so does the way we search. AI-native search is becoming fragmented across platforms like Instagram, Amazon, and Siri, each powered by different models and user intents. Queries are longer (23 words, on average, vs. 4), sessions are deeper (averaging 6 minutes), and responses vary by context and source. Unlike traditional search, LLMs remember, reason, and respond with personalized, multi-source synthesis. This fundamentally changes how content is discovered and how it needs to be optimized.
随着答案格式的变化,我们的搜索方式也在变化。AI 原生搜索正变得在 Instagram、Amazon 和 Siri 等平台上碎片化,每个平台都由不同的模型和用户意图驱动。查询变得更长(平均 23 个单词,而不是 4 个),会话更深入(平均 6 分钟),响应因上下文和来源而异。与传统搜索不同,LLMs 记忆、推理并以个性化的多源综合方式作出响应。这从根本上改变了内容的发现方式以及内容需要如何优化。
Traditional SEO rewards precision and repetition; generative engines prioritize content that is well-organized, easy to parse, and dense with meaning (not just keywords). Phrases like “in summary” or bullet-point formatting help LLMs extract and reproduce content effectively
传统的 SEO 奖励精确性和重复性;生成引擎则优先考虑结构良好、易于解析且富有意义的内容(不仅仅是关键词)。像“总之”这样的短语或项目符号格式有助于 LLMs 有效提取和再现内容。.
It’s also worth noting that the LLM market is also fundamentally different from the traditional search market in terms of business model and incentives. Classic search engines like Google monetized user traffic through ads; users paid with their data and attention. In contrast, most LLMs are paywalled, subscription-driven services. This structural shift affects how content is referenced: there’s less of an incentive by model providers to surface third-party content, unless it’s additive to the user experience or reinforces product value. While it’s possible that an ad market may eventually emerge on top of LLM interfaces, the rules, incentives, and participants would likely look very different than traditional search.
值得注意的是,LLM 市场在商业模式和激励方面与传统搜索市场根本不同。像 Google 这样的经典搜索引擎通过广告来货币化用户流量;用户用他们的数据和注意力进行支付。相比之下,大多数 LLMs 是收费墙、基于订阅的服务。这种结构性变化影响了内容的引用方式:模型提供者对展示第三方内容的激励较少,除非它对用户体验有增益或增强产品价值。虽然在 LLM 界面上最终可能会出现广告市场,但规则、激励和参与者可能与传统搜索大相径庭。
In the meantime, one emerging signal of the value in LLM interfaces is the volume of outbound clicks. ChatGPT, for instance, is already driving referral traffic to tens of thousands of distinct domains.
与此同时,LLM 接口价值的一个新兴信号是外部点击量。例如,ChatGPT 已经为数以万计的不同域名带来了推荐流量。
It’s no longer just about click-through rates, it’s about reference rates: how often your brand or content is cited or used as a source in model-generated answers. In a world of AI-generated outputs, GEO means optimizing for what the model chooses to reference, not just whether or where you appear in traditional search. That shift is revamping how we define and measure brand visibility and performance.
Already, new platforms like Profound, Goodie, and Daydream enable brands to analyze how they appear in AI-generated responses, track sentiment across model outputs, and understand which publishers are shaping model behavior. These platforms work by fine-tuning models to mirror brand-relevant prompt language, strategically injecting top SEO keywords, and running synthetic queries at scale. The outputs are then organized into actionable dashboards that help marketing teams monitor visibility, messaging consistency, and competitive share of voice.
Canada Goose used one such tool to gain insight into how LLMs referenced the brand — not just in terms of product features like warmth or waterproofing, but brand recognition itself. The takeaways were less about how users discovered Canada Goose, but whether the model spontaneously mentioned the brand at all, an indicator of unaided awareness in the AI era.
This kind of monitoring is becoming as important as traditional SEO dashboards. Tools like Ahrefs’ Brand Radar now track brand mentions in AI Overviews, helping companies understand how they’re framed and remembered by generative engines. Semrush also has a dedicated AI toolkit designed to help brands track perception across generative platforms, optimize content for AI visibility, and respond quickly to emerging mentions in LLM outputs, a sign that legacy SEO players are adapting to the GEO era.
We’re seeing the emergence of a new kind of brand strategy: one that accounts not just for perception in the public, but perception in the model. How you’re encoded into the AI layer is the new competitive advantage.
Of course, GEO is still in its experimental phase, much like the early days of SEO. With every major model update, we risk relearning (or unlearning) how to best interact with these systems. Just as Google’s search algorithm updates once caused companies to scramble to counter fluctuating rankings, LLM providers are still tuning the rules behind what their models cite. Multiple schools of thought are emerging: some GEO tactics are fairly well understood (e.g., being mentioned in source documents LLMs cite), while other assumptions are more speculative, such as whether models prioritize journalistic content over social media, or how preferences shift with different training sets.
Despite its scale, SEO never produced a monopolistic winner. Tools that helped companies with SEO and keyword research, like Semrush, Ahrefs, Moz, and Similarweb, were successful in their own right, but none captured the full stack (or grew via acquisition, like Similarweb). Each carved out a niche: backlink analysis, traffic monitoring, keyword intelligence, or technical audits.
SEO was always fragmented. The work was distributed across agencies, internal teams, and freelance operators. The data was messy and rankings were inferred, not verified. Google held the algorithmic keys, but no vendor ever controlled the interface. Even at its peak, the biggest SEO players were tooling providers. They didn’t have the user engagement, data control, or network effects to become hubs where SEO activity is concentrated. Clickstream data — records of the links users click as they navigate websites — is arguably the clearest window into real user behavior. Historically, though, this data has been prohibitively hard to access, locked behind ISPs, SDKS, browser extensions, and data brokers. This made building accurate, scalable insights nearly impossible without deep infrastructure or privileged access.
GEO changes that.
This isn’t just a tooling shift, it’s a platform opportunity. The most compelling GEO companies won’t stop at measurement. They’ll fine-tune their own models, learning from billions of implicit prompts across verticals. They’ll own the loop — insight, creative input, feedback, iteration — with differentiated technology that doesn’t just observe LLM behavior, but shapes it. They’ll also figure out a way to capture clickstream data and combine first- and third-party data sources.
Platforms that win in GEO will go beyond brand analysis and provide the infrastructure to act: generating campaigns in real time, optimizing for model memory, and iterating daily, as LLM behavior shifts. These systems will be operational.
That unlocks a much broader opportunity than visibility. If GEO is how a brand ensures it’s referenced in AI responses, it’s also how it manages its ongoing relationship with the AI layer itself. GEO becomes the system of record for interacting with LLMs, allowing brands to track presence, performance, and outcomes across generative platforms. Own that layer, and you own the budget behind it.
That’s the monopolistic potential: not just serving insights, but becoming the channel. If SEO was a decentralized, data-adjacent market, GEO can be the inverse — centralized, API-driven, and embedded directly into brand workflows. Ultimately, GEO by itself is perhaps the most obvious wedge, especially as we see a shift in search behavior, but ultimately, it’s really a wedge into performance marketing, more broadly. The same brand guidelines and understanding of user data that power GEO can power growth marketing. This is how a big business gets built, as a software product is able to test multiple channels, iterate, and optimize across them. AI enables an autonomous marketer.
Timing matters. Search is just beginning to shift, but ad dollars move fast, especially when there’s arbitrage. In the 2000s, that was Google’s Adwords. In the 2010s, it was Facebook’s targeting engine. Now, in 2025, it’s LLMs and the platforms that help brands navigate how their content is ingested and referenced by those models. Put another way, GEO is the competition to get into the model’s mind.
In a world where AI is the front door to commerce and discovery, the question for marketers is: Will the model remember you?
Zach Cohen is a partner on the consumer tech team at Andreessen Horowitz, where he focuses on companies building at the application layer in generative AI.
Seema Amble is a partner at Andreessen Horowitz, where she focuses on SaaS and fintech investments in B2B fintech, payments, CFO tools, and vertical software.