A Profound study finds Claude's chat and coding apps cite different brands

Profound, a company that tracks how brands appear in AI answers, published a study comparing Claude and Claude Code on the same prompts. The two products share a model family but behave like different answer engines. Profound sells AI visibility tracking and promotes its platform in the report, so treat the framing as vendor material even when the raw numbers are its own.

Between 13 July and 23 July 2026, the company ran 1,724 prompts through both products with web search enabled on each side, producing 24,135 responses in total. Claude used web search on 93 percent of responses. Claude Code did so on 13 percent. On average, the brands mentioned for the same prompt overlapped by about 20 percent between the two products.

Two products, two kinds of answer

Repeat runs from a single product were more consistent than the two products were with each other. Two responses from Claude to the same prompt shared roughly half of their brands. Two responses from Claude Code shared about 40 percent. Variation within one product falls inside those ranges, so the gap between products reflects a real difference.

Despite searching less often, Claude Code mentioned a higher number of brands per response, 6.6 on average, compared with 5.2 for Claude. On coding prompts, Claude favoured code editors and IDEs. Claude Code more often mentioned code-quality and workflow tools. Claude Code answers were also shorter and more structured: 322 words on average versus Claude's 459, and more than half of Claude Code responses included a table, compared with 11 percent from Claude.

What the agents actually fetch

The second dataset in the report looks at where each product's agent goes when it does browse. Between 18 July and 18 August, Profound recorded the top 1,000 pages visited by each agent across domains it tracks internally. Roughly three-quarters of Claude Code's observed visits went to documentation, informational and pricing pages, against 5 percent for Claude. In the other direction, 60 percent of Claude's visits went to robots.txt files, sitemaps and homepages, against 4 percent for Claude Code.

Profound reads that split as Claude's agent exploring what a site contains and Claude Code's agent looking for specific details on known page types. The report says an AI model labelled the page types, without stating how the two agents' traffic was distinguished or whether a human reviewed the labels, so a reader should treat the specific ratios with caution.

One tracker label hiding two behaviours

The practical point is that the products a team already measures under a single "Claude" label are doing different things. If Claude Code is the product your customers use to pick between developer tools, the pages Claude Code's agent visits, docs, technical detail, pricing, are where the brand mentions come from, and generic homepage or category content will not produce those mentions. Profound's own advice, again framed to sell its platform, is to state technical facts plainly and put answers under question-shaped headings, but it does not test whether that change increases brand mentions.

The main finding is simpler: for a product that offers an assistant built on a base model, its agent's search behaviour can differ from the base product's, and measurement that treats the pair as one thing will hide that. If your team already tracks AI visibility, checking whether the tracker separates the two products at all is a five-minute job with a real answer.

Source

• Profound: Claude and Claude Code are distinct answer engines
• Search Engine Journal: Claude Searches Web Differently From Claude Code, Data Shows