How to Measure Your Brand's Visibility in Claude

Measuring Claude visibility uses the standard answer-engine KPI set — mention rate, citation rate, share of voice, answer position, referral traffic — plus one metric unique to selectively-searching assistants: grounding rate, the share of your prompts where Claude actually searched the web. Since Claude's web search arrived in March 2025, brand visibility here is a blend of what the model learned in training and what it retrieves live, and good measurement separates the two.

Why does Claude need a different measurement lens?

Perplexity searches on essentially every query; Claude searches when it judges the prompt to need current information. That means two prompts about your brand can take entirely different paths — one answered from parametric memory, one grounded in retrieved pages with citations. The distinction is diagnostic gold. Strong mentions in ungrounded answers but absence from grounded ones means your live content underperforms your reputation; the reverse means the model's stored picture of you is stale or thin while your pages compete well. Each failure mode has a different fix, and a measurement system that doesn't record grounding can't tell them apart.

Which KPIs belong on the scorecard?

The split metrics (rows 3-4) are where Claude-specific insight lives. Compute them separately rather than blending, and expect them to move on different timescales — grounded citation rate responds to content changes within weeks, while the ungrounded footprint shifts only as models retrain on a refreshed web corpus.

How do you build the baseline?

Assemble 50-150 prompts across discovery, evaluation, comparison, and factual intents, phrased conversationally. Run each 3-5 times per weekly cycle, recording answer text, brands mentioned with order, whether search fired, and cited URLs. Four weeks of this produces a baseline with honest confidence bands; resist reporting anything sooner, because week-one numbers on a new prompt set are dominated by sampling variance.

Then set targets against the baseline rather than absolutes. "Grounded citation rate from 12% to 25% on comparison prompts this quarter" is actionable; "be more visible in Claude" is not.

What should trend reports actually show?

Report the KPI grid split by prompt intent, month over month, with model-update dates annotated — Anthropic's releases can step-change answer patterns overnight, and unannotated charts get misattributed to content work. Include the descriptor language Claude uses for your brand as a qualitative track; its careful, hedged register makes wording shifts ("a leading option" gaining "though pricing is higher than alternatives") an early-warning channel for reputation drift in your review corpus.

Sustaining this manually — hundreds of runs weekly, grounding detection, citation parsing — is exactly the toil AI visibility tracking in Menra exists to absorb, running the same prompt set across Claude and its peer engines so share-of-voice comparisons stay apples-to-apples. For prompt-set construction and scoring rubrics in depth, start with the mention tracking methodology guide; the framework there ports directly, with grounding rate added as Claude's extra column.