New data from Profound reveals Claude and Claude Code search the web differently, creating a fragmented visibility landscape for global brands and SEOs.

For months, the prevailing assumption among digital marketers and business owners has been that an AI ecosystem operates as a monolithic entity. If a brand is visible in a specific LLM, it is assumed to be visible across all its iterations. However, new data from Profound, a firm specializing in AI brand visibility, suggests that the reality is far more fragmented. The gap between Anthropic's standard Claude and its developer-centric Claude Code is not just a matter of interface, but a fundamental divergence in how they perceive and navigate the internet.

A structural split in web exploration

The scale of the discrepancy is stark. In a study involving 1,724 prompts and over 24,000 responses, Profound discovered that Claude behaves as an exhaustive web explorer, triggering a search in 93% of its interactions. In contrast, Claude Code is remarkably selective, utilizing web search in only 13% of cases. This suggests that Claude Code relies far more heavily on its internal training data or the specific local context provided by the developer, rather than seeking real-time external validation.

This is not merely a difference in frequency but a difference in intent. When the two tools are asked the same question, they rarely arrive at the same set of references. The overlap of brands mentioned for identical prompts is only about 20%. For a global business, this means that being the top recommendation in a Claude chat does not translate to being the recommended tool within Claude Code. The two products function as separate answer engines with distinct retrieval logics.

Where the agents actually go

The divergence extends to the specific types of pages these agents visit. Analysis of the top 1,000 pages visited by each agent reveals a clear split in behavioral patterns. Claude's agent spends a significant portion of its time on the periphery of a website, with 60% of its visits directed toward robots.txt files, sitemaps, and homepages. It acts as a scout, attempting to map the overall structure and content of a domain.

Claude Code ignores the map and goes straight for the manual. Nearly 75% of its visits are targeted at documentation, informational pages, and pricing tables. While Claude asks what a site contains, Claude Code asks how a product works and how much it costs. This behavioral split is further evidenced by the output format; over half of Claude Code's responses include a table, compared to just 11% for the standard Claude, reflecting a preference for structured, technical data over conversational prose.

The impact on brand visibility and SEO

This fragmentation creates a new challenge for SEO professionals and brand managers. The traditional goal of AI optimization can no longer be a one-size-fits-all strategy. If a company focuses solely on high-level brand storytelling on its homepage, it may capture the attention of Claude but remain invisible to Claude Code.

To capture the agentic workflow, companies must prioritize the technical hygiene of their documentation. Since Claude Code targets pricing and info pages, the accuracy and machine-readability of these specific assets become the primary drivers of visibility. The data shows that for coding-related prompts, Claude tends to surface general code editors and IDEs, whereas Claude Code identifies specific workflow and code-quality tools. This indicates that the more specialized the agent, the more it values granular, utility-driven content over general authority.

Beyond the chatbot: Real-world business adoption

While the technical divergence is a concern for marketers, the broader Claude ecosystem is already seeing deep integration into professional services. In the US real estate sector, for instance, brokers are moving beyond simple chat interfaces. According to reports from the National Association of REALTORS, some brokerage leaders are using Claude to automate complex workflows and build proprietary technology.

The adoption is spreading across the product suite: Claude.ai for general assistance, Claude Cowork for executing processes across desktops, and Claude Code for developers building custom tools. This shift from chatbot to autonomous agent means that the AI is no longer just summarizing information for a human; it is actively navigating the web to execute tasks. When an agent like Claude Code decides not to search the web (which it does 87% of the time), it relies on the quality of the data it has already ingested or the documentation it has previously indexed.

Security risks in the agentic era

As these agents become more autonomous and integrated into business workflows, the security surface area expands. Anthropic's own threat intelligence indicates a sophisticated landscape of misuse. Between December 2025 and August 2026, the company disrupted operations involving state-sponsored groups and financially motivated criminals attempting to use Claude for cyber operations and surveillance.

The risk is particularly acute as agents move from being assistants to orchestrators. The ability of an agent to connect with external tools and data sources—a core feature of Claude Code—creates potential vectors for attack. Anthropic has identified threats ranging from Russian espionage to the use of AI in autonomous attack frameworks. For the enterprise, this means that deploying agentic AI requires a dual strategy: optimizing for visibility while implementing rigorous safeguards to prevent the tool from being leveraged in illicit distillation or cyber-attacks.

The divergence between Claude and Claude Code proves that we are no longer optimizing for a single AI, but for a variety of specialized agents with different browsing habits and goals.

Global implications for US and UK enterprises

For businesses operating in the USA and UK, this divergence signals the end of the generalist AI strategy. In these markets, where the competition for AI-driven lead generation is intensifying, the focus must shift toward Agentic SEO. This involves creating distinct content paths: one for the human-centric, exploratory search (Claude) and one for the task-oriented, technical search (Claude Code).

From a regulatory perspective, the UK's pro-innovation approach and the US's evolving framework on AI safety mean that companies have significant freedom to deploy these agents, but they bear the full responsibility for the output. The fact that Claude Code rarely searches the web means it may rely on outdated internal data unless the company's technical documentation is frequently updated and easily discoverable. For global firms, the priority is now clear: ensure that pricing, API docs, and technical specifications are structured for machine consumption, or risk being omitted from the workflows of the world's most powerful coding agents.

FAQ

---

Sources: Searchenginejournal, Letsdatascience · , anthropic.com , nar.realtor