An AI visibility audit should separate website readiness from observed answers. A readiness assessment checks the website foundation. An observation records what a named product returned to an exact question at a particular time.
Define the question panel
Select discovery, comparison and due-diligence questions relevant to a buyer and market. Keep branded questions separate from unbranded provider queries. Freeze the wording before collecting a baseline and retain a versioned copy.
Do not choose only questions that already mention your business. That tests recognition under a prompt that supplies the answer candidate, not independent discovery.
Record controlled conditions
For each attempt, record the product surface, model when exposed, search setting, account or personalization controls, tester location, intended market, prompt and timestamp. Use fresh conversations and repeat each question three times.
A consumer ChatGPT observation is not interchangeable with an API test. An absent Google AI Overview is not the same event as a network failure; record what was actually displayed and use a predefined classification.
Classify each successful response
Record mentions, explicit recommendations and citations separately. Preserve the original answer and supporting links. If an answer contains an explicit ordered list, you may record position. Do not assign a rank to unordered prose or citation order.
A successful answer without your name is a valid no-mention observation. Failed, blocked and timed-out attempts belong in a separate error count, not a zero visibility score.
Calculate interpretable rates
Mention rate is successful observations naming the business divided by all successful observations in the same condition and segment. Use the same denominator discipline for recommendations and citations. Report counts alongside percentages.
If no successful observations exist, report Not measured. Do not pool different markets, settings and products into a headline rank. Our measurement protocol describes the internal benchmark and client-package distinction.
Diagnose cautiously
An absent mention does not prove a missing schema field or lack of training data. Inspect actual technical failures, inaccurate company facts and gaps in buyer information. Treat a model’s explanation of its choice as a lead to investigate, not an authoritative diagnostic log.
Google’s AI search documentation distinguishes eligibility from inclusion and explains that AI-feature traffic is part of Search Console’s Web reporting.
Decide the next implementation step
Prioritize verified crawl or page errors, unclear service information and unsupported claims. Record the change, deployment date and subsequent observations. Before/after association is not proof of causation.
The free audit provides a readiness summary only. Requesting a comprehensive audit confirms receipt of an inquiry; it does not automatically generate or email a report or fixes.
For implementation scope, see AI Visibility Engine, Dubai delivery and Engine pricing.