Engines

Claude

Anthropic Claude · Claude.ai

Definition

Claude is Anthropic’s conversational model family and the chat product people use to ask category questions. For earned media measurement it is a third retriever, not a ChatGPT skin. It often names organizations and papers with fewer clickable footnotes than Perplexity, and it refuses more aggressively on medical, legal, and personal-data prompts.

Treat each Claude surface separately: claude.ai chat, API, and any product that embeds the model with its own retrieval. A citation log from one is not a log from all.

How it works

Run a fixed prompt set logged-out if the product allows it. Record whether the brand is named, whether a URL appears, which URL, and whether the sentence is true on that page. Store the model name and date. Claude’s retrieval and refusal settings change; a May sheet is not a September sheet.

Earned pages that state a sourced number still help, because when Claude does retrieve, it prefers documents with methods over slogans. Wikipedia, docs, and journalism show up often. Paid native roundups less so.

Worked example

Forty prompts on “best SOC 2 auditors for a 40-person SaaS firm.” Claude names the firm in 8 answers, shows a source list in 3, and links a trade write-up twice. ChatGPT named the firm in 6 with more URLs. Perplexity listed the methods page in 22 of 50. Report three rows. Do not average them into one “AI score.”

How it differs

ChatGPT is a sparser citer in several public crawls but a larger consumer audience. Perplexity is built to show sources. Google AI Overviews live in Search. GEO is the practice across them. Claude is one engine.

Common errors

Using a staff Claude Project with uploaded PDFs as “how the public model sees us.” Counting a refusal as a negative citation. Ignoring trade coverage because “Claude doesn’t browse” — some products do retrieve.

Sources