Build a baseline small enough to repeat
A useful first baseline does not require a monitoring platform. It requires a fixed set of real buyer questions, consistent collection fields and disciplined separation between a brand mention, visible citation and website referral.
Step 1: define the query set
Start with 15–25 questions across three intents: problem discovery, provider comparison and decision validation. Use the language customers actually use. Avoid inserting the brand name unless the purpose is to test branded accuracy.
Three intent levels
Discovery: What kind of provider handles this problem?
Comparison: Which providers serve this location and explain their process?
Validation: What should I check before choosing one?
Step 2: create the capture sheet
| Field | Why it matters |
|---|---|
| query_id and exact text | Prevents silent wording changes. |
| platform and product mode | Answers can differ by interface or feature. |
| market, language and date | Preserves collection context. |
| brand_mentioned | Records whether the answer names the brand. |
| brand_cited | Records whether a visible source supports that mention. |
| source_urls | Allows source diversity and page-level review. |
| capture_link | Connects the row to the full observation. |
Step 3: collect without interpreting
Run the frozen set in a defined time window. Save the full answer and every visible source before writing conclusions. If an answer fails or citations are unavailable, record that state instead of rerunning until a preferred result appears.
Step 4: calculate transparent metrics
- Mention rate: answers naming the brand ÷ eligible captured answers.
- Cited-mention rate: answers naming and visibly citing the brand ÷ eligible captured answers.
- Source diversity: unique cited domains and URLs, shown as counts.
- Coverage: query groups with at least one useful owned or independent source.
Always publish the count beside a percentage. A result of 3/20 is easier to interpret than “15% visibility,” especially when the sample is small.
Step 5: connect first-party outcomes
Review analytics separately for observable AI referrers, landing pages and meaningful actions. Do not infer that every direct visit came from an assistant. A query capture and a referral session are different datasets unless a specific evidence path connects them.
A 90-minute monthly routine
- Confirm the set and collection context.
- Capture the answers and sources.
- QA missing or ambiguous rows.
- Calculate counts and rates from the frozen denominator.
- Review newly cited pages and clear content gaps.
- Log any method change before the next run.
Do not rewrite the site after seeing one surprising answer. Look for repeated gaps across related questions, then choose the smallest page or evidence change that addresses them.
Where to go next
Use the Nexus AI Visibility Framework to place this topic inside a complete measurement system, or review the AI Visibility Audit scope.
13 Jul 2026 — Expanded with an independent structure, examples, implementation guidance and primary references.