Search with passage-level citations
Since March 2025 Claude searches the web and cites sources at the level of specific passages, not whole pages. Pages win when a precise, verifiable sentence can be lifted out of them.
There is no ranking inside Claude. There is an answer your company is in or absent from. EntityRise measures this on 20 approved buyer questions, separately for each of five providers, and shows what to fix so the answer includes you.
How Claude builds an answer
Claude answers carefully and flags uncertainty, so brands with an unambiguous description of what they are and who they serve win. Vague positioning ends in omission, not in a bad recommendation.
Sources and weights
Since March 2025 Claude searches the web and cites sources at the level of specific passages, not whole pages. Pages win when a precise, verifiable sentence can be lifted out of them.
Anthropic does not officially confirm its search provider, but documentation and analyses point to the Brave Search index. The practical conclusion: a site strong in Google can be weakly visible in the index Claude draws on.
Claude cites more rarely and more carefully than Perplexity. Generalities and pages without specifics lose faster here; documentation, dated numbers and clear definitions win.
What this changes about your sources: check how your company looks in Brave Search results, and keep fact pages a model can cite one sentence from.
Sources: TechCrunch (2025) and the Profound analysis. Checked on September 16, 2026.
Measurement
In the report Claude has its own matrix column and index row; description accuracy is especially instructive here, because the model says openly what it does not know about you.
Each of the 20 questions goes to Claude independently; the result is never averaged with other providers.
Mention, active recommendation, alternative, position, citation support and description accuracy.
The exact model version used in your audit is recorded in Audit Integrity.
Common questions
Ask the questions your buyer would ask, without your company name, and record the answers. The audit does the same in a controlled way: 20 approved questions, a sealed set and a repeatable result.
No. No provider sells placements in answers. The result is earned openly: a clear company description, content for buyer questions and evidence a model can cite.
Partly. The same work raises the odds everywhere, but each provider weighs sources differently, which is why the audit measures each one separately instead of showing one average.
The $49 audit shows Claude answers to 20 of your buyer questions, next to the four other providers.