If your content strategy treats all Anthropic products as a single traffic source, Profound’s monitoring suggests you may be missing half the picture. Tests run in July found Anthropic’s general-purpose Claude and developer-oriented Claude Code visit different page types, mention different brands, and produce differently structured answers — outcomes that change what pages you should optimize for AI visibility.

What Profound measured

Between July 13 and July 23, Profound ran 1,724 prompts against both Claude and Claude Code (web search enabled), generating 24,135 responses. The company reports consistent differences in search usage, output shape and brand mentions:

  • Search usage: Claude invoked web search in about 93% of responses; Claude Code did so in roughly 13%.
  • Brand overlap: For the same prompt, the two agents mentioned the same brands only around 20% of the time on average.
  • Response shape: Claude Code’s answers were shorter and more structured (avg. ≈322 words) versus Claude’s longer replies (avg. ≈459 words). Over half of Claude Code responses included a table, compared with 11% for Claude.
  • Brands per response: Claude Code mentioned about 6.6 brands per response versus 5.2 for Claude.

How the agents browse sites

Profound also tracked the top 1,000 pages each agent visited (observed between July 18 and August 18) across domains it monitors. The browsing patterns diverged sharply:

  • Claude Code: Nearly three-quarters of observed visits went to documentation, informational pages and pricing pages.
  • Claude: About 60% of observed visits targeted robots.txt files, sitemaps and homepages; only about 4–5% of Claude Code’s visits went to those page types.

Profound interprets these patterns as Claude exploring site structure and entry points, while Claude Code seeks pages that explicitly contain technical and pricing details.

Why marketers and SEOs should care

If your users include developers or teams using developer-focused agents, a single “Anthropic” metric can hide meaningful differences in discovery and attribution. Practical implications:

  • Don’t treat products as identical traffic sources. Different agent behavior means different pages are likely to be surfaced and quoted in AI answers.
  • Documentation and pricing matter for developer agents. Claude Code’s preference for docs and pricing pages means those pages are first-class assets for visibility with developer audiences.
  • Structural files still matter for discovery. Claude appears to consult canonical site files and homepages, so accurate sitemaps and robots directives remain important for broader AI discovery.

Limits and caveats

  • Sample and labeling: The browsing analysis covers a single month and the top 1,000 observed pages across domains Profound monitors. Profound used an AI model to label page types and does not report human verification or how domains were selected.
  • Correlation vs. causation: The report recommends structuring docs and pricing pages for Claude Code, but it does not test whether those changes actually increase brand mentions or referral traffic from agents.
  • Commercial interest: Profound sells AI visibility tools and promotes its platform in the report; that context matters when evaluating tactical recommendations.

Practical next steps for teams

You don’t need definitive validation to act; use the findings to prioritize experiments that are low-cost and measurable:

  • Audit agent traffic. Where possible, segment logs or referral sources by agent or product to see whether developer-focused agents are already driving queries to documentation and pricing pages.
  • Make facts machine-friendly on docs and pricing pages. Lead with concise answers, expose exact compatibility and pricing details, and use question-shaped headings so content is easily extractable by agents.
  • Provide clear structural signals. Keep accurate sitemaps, robots directives and a sensible homepage hierarchy so agents that crawl those files find the pages you want surfaced.
  • Measure changes. After updating pages, track brand mentions in AI outputs where feasible and monitor referral, engagement and conversion metrics tied to developer-focused channels.

These steps prioritize low-friction wins: make the machine-readable facts explicit, then measure whether AI mentions or referrals change.

What to watch next: look for third-party replications of agent browsing patterns, any disclosures from Anthropic about agent behavior, and controlled tests that link specific page changes to improved AI mentions or traffic. Those confirmations will determine whether the recommended content changes should move from experimental to standard practice.