Measure Claude brand visibility by running a fixed set of buyer-relevant prompts on a defined schedule and recording structured fields: brand mention, position, recommendation context, linked citations, cited domains, sentiment, competitors, and answer uncertainty. Preserve the exact prompt, date, mode, and verification evidence for every run.
Build a prompt set by intent
Use multiple prompt groups:
- Problem discovery: How do teams solve a specific problem?
- Category discovery: What tools or approaches fit that problem?
- Comparison: Which options differ on a defined feature or constraint?
- Brand-specific: What does the brand do, cost, support, or integrate with?
- Risk and trust: What limitations, complaints, security controls, or alternatives appear?
Do not rewrite prompts between runs unless the change is recorded as a new test series. A small wording change can alter the answer.
Record structured metrics
For each prompt, capture:
- Whether the brand appears.
- Its first position among named options.
- The exact claim made about the brand.
- Whether a clickable source is present.
- The cited URL and domain.
- Whether the source supports the claim.
- Competitors mentioned in the same answer.
- Positive, neutral, negative, or mixed context.
- Material factual errors or omissions.
Separate mentions from citations. A brand can appear without a linked source, and a brand-owned page can be cited without the brand being recommended.
Calculate sampled metrics transparently
Useful formulas include:
Mention rate = prompts with a brand mention / eligible prompt runsCitation rate = runs with a linked brand source / eligible prompt runsFirst-position rate = runs where the brand appears first / runs naming any vendorSupported-claim rate = verified supported claims / reviewed factual claimsError rate = runs with a material brand error / reviewed runs
Publish the denominator beside every percentage. Segment results by prompt intent so a large group of brand-specific questions does not inflate general category visibility.
Use a written coding guide for sentiment and recommendation labels. Mixed answers should not be forced into positive or negative when the system gives both strengths and limitations.
Anthropic’s developer documentation explains that citations can be attached to supported source material in API workflows, while Claude’s web-search documentation describes current search behavior. Review the official citations guide and web-search tool guide when designing an API-based test.
Control the rerun conditions
Record the Claude product or API path, model, search setting, location or language where relevant, account state, and date. Run enough repeats to distinguish a persistent pattern from one response variation.
Do not present the resulting percentage as a universal market-share metric. It describes visibility within the sampled prompts and conditions.
Repeat a subset of prompts within the same collection window to measure response variability. Report both the aggregate result and the stability of individual prompts. A trend that depends on one unstable answer deserves less confidence than a change repeated across several related prompts.
Verify sources manually
Open every consequential citation. Check whether it supports the adjacent claim, whether the page is current, and whether the canonical URL belongs to the correct entity. Flag unsupported or stale claims separately from low visibility.
Use a second reviewer for disputed sentiment, recommendation, or source-support labels. Record disagreements and resolve them against the written coding guide instead of silently choosing the more favorable classification.
Retain both the original and corrected labels for auditability.
Use BrandJet’s full Claude visibility monitoring protocol for the broader analytics and crawler-log workflow.
Frequently asked questions
What is a Claude visibility score?
It is usually a custom summary of performance across a defined prompt set. The formula, weighting, prompt sample, and limitations must be disclosed.
How many prompts should I track?
Use enough prompts to cover important intents without creating an unreviewable dataset. Start small, validate the taxonomy, and expand deliberately.
How often should I rerun the prompts?
Weekly or monthly can suit strategic monitoring, while high-change topics may need more frequent checks. Keep the cadence consistent.
Are Claude referrals the same as Claude citations?
No. A citation is a source shown in an answer. A referral is a website visit attributed to a Claude surface. Measure them separately.