Choose the decision before the metric

An AI visibility report should answer a practical question: are relevant buyers encountering accurate information about your business? Start by naming the product, buyer, geography, and buying situation. A broad total can disguise weak performance in the one segment that actually purchases. Treat this framework as a measurement design, not a complete audience census. Give each metric a job. Mentions describe recognition, citations identify linked sources, referrals measure attributed visits, and conversions describe subsequent actions. Keeping these stages separate makes the report useful to both content teams and people responsible for revenue.

Create a stable observation set

Collect questions from sales calls, support tickets, internal search, and existing organic queries. Label them by informational, evaluation, implementation, and purchasing intent. Freeze a baseline before adding exploratory questions. Otherwise, changing the questions can move a visibility score even when the underlying answers have not improved. For every observation save the prompt, date, engine, product surface, search setting, language, and exact response. Record unsuccessful requests separately. A failed collection is missing data; it is not evidence that your business was absent. Assign stable identifiers so future reports can reproduce the selection.

Calculate transparent rates

Calculate mention rate as valid answers mentioning the intended business divided by valid answers collected. Calculate domain citation rate separately. Count each answer once for these binary rates, even if it mentions your name repeatedly. Also keep raw counts so a percentage cannot conceal a tiny sample. Hypothetical example: an audit collects 40 valid answers and finds the business in eight. The observed mention rate is 20%. Four answers link to its domain, giving a 10% citation rate. These results describe the audit sample, not all users or market share.

Connect platform reports carefully

Google documents AI-feature traffic within Search Console's Web reporting. Bing documents citation activity in its AI Performance report. These measurements answer different questions. Preserve their original labels rather than merging them into one supposedly universal rank. Alongside the observation set, review landing-page visits and meaningful actions such as completed trial setup. A page may supply a cited definition without driving many clicks. That can still be useful, but it should not be reported as customer acquisition without corresponding evidence. Explain which data comes from your tests and which comes from platform reports.

Decide what happens next

Assign actions to findings: correct inaccurate product information, improve a genuinely missing explanation, investigate inaccessible pages, or gather more observations. Give each action an owner and review date. Avoid rewriting a page solely because one answer omitted it. Keep a methodology note attached to every export. Include the prompt version, valid sample count, excluded failures, time window, and known limitations. A defensible small report is more valuable than an impressive score whose inputs nobody can inspect. Reuse the same baseline at the next review and annotate any changes in collection.

PUT IT TO WORK

Build a measurement plan that separates brand mentions, citations, referrals, and business outcomes.

Sources & further reading

Official documentation informs this guide. Examples are illustrative, and product interfaces or limits can change.

THE WORKSPACE WE'RE BUILDING

From search signals to a useful first draft.

SERP.so is in development. Join the waitlist for a workspace that connects research, content review, and publishing.

Join early access ↗