Measurement guide · version 1
Understand what the numbers describe.
Use this guide to evaluate a visibility report and inspect the conventions in our worked sample. A client reporting specification must document its actual collection method and metric definitions. This page does not certify a proprietary scoring algorithm or claim universal market coverage.
Define the observation before calculating a metric.
Record the question, platform and exact surface, run identifier, date and time, geography, language and relevant account or session conditions. A consumer search interface and an API observation must not be silently treated as the same experience.
Separate unique questions from paraphrases and repeated runs. Document the competitor group, question-selection rationale, inclusion rules, failures and exclusions. A panel of 101 or 301 prompts describes coverage; it does not by itself establish statistical representativeness.
The worked sample contains three fictional questions × two fictional platforms × two runs = twelve coded records. All are valid by construction. No real collection took place.
Brand mentions
A mention records whether the business is named in an eligible response under a stated matching rule. In the sample, repeated names within one response count once. Mention coverage is mention-bearing responses divided by eligible responses: 6 / 12 = 50%.
A real specification needs alias and ambiguity rules and should show counts beside percentages. A mention is not necessarily a recommendation, a citation, a visit or a sale.
Explicit recommendations
Track a recommendation separately from a passing mention. The sample marks four of twelve responses as explicit recommendations, or 33.3%. A recommendation also has a mention flag in this particular example.
In client work, document the coding criteria, treatment of conditional or negative statements, human review and any classifier uncertainty. Do not combine different recommendation definitions without disclosure.
Cited sources
Distinguish responses containing a citation from the number of individual links and unique source domains. In the sample, three of twelve responses carry an owned-citation flag: 25% response coverage. No real source URLs are represented in those invented flags.
A real report should identify the cited URLs, ownership rules, source classification and deduplication. An observed citation does not reveal every source a system may have used, prove source authority or guarantee a future citation.
Google visibility
Label rankings, impressions, clicks and qualified actions separately. Identify the data source, queries, pages, date range and market settings before comparing them. Do not present a selected ranking panel as complete customer demand or blend it with AI-response counts without an explicit method.
For Google's AI features, Google states that the relevant SEO foundations apply and that no special schema is required. This guidance concerns Google's documented features, not every third-party AI platform. Read Google's primary guidance.
Geographic visibility
A local comparison needs the search location or grid, service or query, business location, time and competitor coverage. Keep Maps observations separate from website traffic, profile actions and confirmed business outcomes.
Document what changes between markets and do not generalize one observed location to an entire service area. The worked sample contains no real Maps measurements.
Change over time and limitations
Compare compatible questions, competitor groups, surfaces and collection conditions. Record changes in the measurement setup and consider persistence and commercial relevance before prioritizing a movement.
Repeated responses are not automatically independent samples. Do not present a confidence interval or population claim unless the study design justifies it. A change after implementation is not proof that the implementation caused it.
A composite or normalized score needs a versioned specification of inputs, weights, exclusions and scale. The service pages use descriptive measurements and a labeled teaching sample instead of unexplained proprietary scores.