Understand how Claude synthesizes your brand evidence.
Measure recommendations, source-backed claims and long-form comparison context across prompts where detailed reasoning matters.
How Claude actually answers buying questions
From our own benchmark: 240 answers this engine produced for brand-free buyer questions across 12 categories, alongside the other engines on the same questions.
| Answers analysed | 240 | across 12 categories |
|---|---|---|
| Answers naming at least one provider | 48% | the rest gave guidance or pointed to a directory |
| Providers named per answer | 2.41 | fewer than the 2.70 average across all engines measured |
| Answers carrying citations | 85% | engine average 96% |
| Citations per answer | 5.77 | counted only from markers inside the answer text |
| Answers ordered as a ranked list | 25% | a ranked list is harder to enter than an unordered one |
| Answers pointing to a directory or review site | 16% | where this is high, a directory listing is a precondition |
| Model queried | anthropic/claude-haiku-4-5 | disclosed so the figures can be reproduced |
Where Claude names brands, and where it stays generic
Sorted by how many providers this engine names per answer. Categories at the bottom are open ground: the engine currently answers them without naming anyone.
| Category | Answers | Name ≥1 provider | Providers / answer | Most-named by this engine |
|---|---|---|---|---|
| Project management software | 20 | 65% | 3.35 | monday.com |
| Furniture brands | 20 | 55% | 3.20 | IKEA |
| Restaurants | 20 | 45% | 3.20 | Bounty Kitchen |
| Skincare brands | 20 | 45% | 2.80 | La Roche-Posay |
| Law firms | 20 | 55% | 2.35 | Regas & Dallas P.C |
| Gyms & fitness studios | 20 | 45% | 2.30 | Planet Fitness |
| Pet supplies | 20 | 50% | 2.30 | PetSafe |
| Dentists | 20 | 40% | 2.25 | Austin Dental Spa |
| Hotels | 20 | 35% | 2.00 | Faena Hotel Miami Beach |
| HVAC & plumbing services | 20 | 45% | 1.90 | Lennox |
| Electric bikes | 20 | 50% | 1.75 | Aventon |
| Real estate agents | 20 | 40% | 1.55 | Compass |
What Claude cites
The domains this engine cited most across the benchmark. Engines differ here more than they differ on which brands they name, which is why source work has to be planned per engine.
| Domain | Type | Citations | Categories |
|---|---|---|---|
| reddit.com | Communities (Reddit, Quora…) | 34 | 9 |
| alibaba.com | Other sites | 15 | 3 |
| rickjanson.com | Named providers' own sites | 9 | 1 |
| nytimes.com | News & media | 6 | 3 |
| forbes.com | News & media | 6 | 3 |
| ebikeoracle.com | Named providers' own sites | 6 | 1 |
| povison.com | Named providers' own sites | 6 | 1 |
| dermapproved.com | Other sites | 6 | 1 |
| linkedin.com | Other sites | 5 | 3 |
| wrike.com | Named providers' own sites | 5 | 1 |
Letting Claude read your site
Anthropic separates content gathering from user-triggered fetches in the same way. Both are named agents, so both can be allowed or blocked deliberately.
ClaudeBot · Claude-User
Test your site against these agents · Read the crawler access guide
Anthropic documents server-side web search as a tool available in supported Claude API workflows.
Anthropic tool use ↗What to inspect in Claude.
Keep a shared core framework, then add the engine-specific evidence needed to explain differences.
A repeatable Claude visibility workflow.
The goal is evidence-led improvement, not guaranteed placement.
Keep exploring
Questions about GEO for Claude
Can Claude use web search?
Claude supports server-side web search in supported product and API experiences. Availability and citation behavior can vary by configuration.
What evidence works well for long-form synthesis?
Clear methodology, stable primary documents, precise definitions and internally consistent product facts make synthesis easier to verify.
Should Claude be measured with the same prompts as ChatGPT?
Keep a shared core prompt set for comparison, then add engine-specific prompts for tasks where long-form analysis or document synthesis matters.
How often should Claude be re-measured?
Often enough to separate a real change from sampling variance. A single run is a baseline, not a trend.
Why does Claude name a competitor instead of us?
Usually because the evidence it can read about them is clearer or more corroborated. That gap is what the knowledge base and content work are for.
Do results from Claude transfer to other engines?
Partly. The underlying evidence is shared, but retrieval behaviour and citation formats differ, so each engine is still measured on its own.
Build a measurable Claude visibility program.
Create a project, confirm your market and start measuring the questions that matter.