How to measure AI search visibility without invented scores

Understand the decision

Choose the questions and platforms relevant to the business. Record the date, context and method of any manual observation, since answers can vary. Use available webmaster reports to understand which pages are cited and why. Do not claim full attribution when referral data or reporting is incomplete. Evaluate whether the cited page still provides a useful next action for a customer.

A practical approach

Write a direct answer to a real customer question, then provide the context that prevents the answer being misleading when quoted alone. Define terms, identify the business or product clearly and distinguish your own method from external evidence. Keep essential facts in accessible page text. Google’s generative search guidance continues to emphasise SEO fundamentals and useful content; special files or phrasing do not guarantee selection by an AI answer system.

• Use documented platform reporting where available.
• Record manual-check context and limits.
• Separate citations, visits and qualified enquiries.
• Inspect the usefulness of cited pages.

Illustrative example

A business could compare its service-guide citations in a webmaster report over time and inspect the cited pages. It could also review relevant enquiries for the same period. Neither observation alone would prove that an AI answer caused a particular sale or that every assistant sees the business similarly.

A mistake to avoid

Avoid combining a few prompt tests into a percentage labelled as total AI authority. The denominator and sampling method matter.

What to review

Report observed citations, relevant page visits and enquiry evidence with their known gaps rather than promising a universal AI performance measure.

Turn the guide into a working brief

Keep a record of the question, the page that answers it, the evidence supporting factual claims and the date the facts changed. Where a platform provides citation reporting, use it to inspect which pages appear and the surrounding query context. Review quoted passages for accuracy and usefulness rather than chasing a single visibility score. Avoid hidden prompts, invented credentials and unsupported claims about guaranteed AI citations.

Common questions