How to measure AI visibility without getting distracted by a score

Your business appeared in an AI answer. Good.

The next questions are more useful: was it the right question, was the information accurate, and did the appearance help anyone take the next step?

Measure AI visibility at four levels: relevant appearances, accuracy, identifiable website activity and qualified business opportunities. No single score captures all four.

A mention is a useful signal. It is not a purchase or a booked project.

1. Start with the questions that matter to your business

Build a manageable question set around your products or services, availability and customer decisions. Our suggested starting point is 15 to 25 questions, adjusted to the business. That is a practical testing recommendation, not an industry benchmark.

Include three groups:

• Discovery: Which businesses could handle this type of project in this area?
• Evaluation: What should someone compare before requesting quotes?
• Brand accuracy: What does the assistant say about your business, products or services and availability?

For an online store, discovery questions can ask which products suit a particular need. Evaluation questions can compare product specifications, compatibility, delivery or returns. Keep service-provider questions where they reflect the business.

Keep branded and unbranded questions separate. Being mentioned after someone supplies your name is different from being suggested when they do not know you exist.

Record the exact wording. Changing the questions every month makes the comparison less useful.

2. Separate a mention, a citation and a recommendation

Use a simple reporting table:

A citation to your educational article might be valuable even when your business is not recommended. Equally, a recommendation for a product or service you do not provide needs correction, not celebration.

For repeat tests, record the platform, date, location setting and whether web search was used. Save the response and supporting links. Treat the results as a sample under those conditions, not everybody's experience.

3. Understand what each tool can tell you

Semrush can help discover topics and compare monitored appearances. Its topic volume is estimated, and its visibility metrics reflect the dataset and method being used. They are directional research, not a count of all real customer conversations. Semrush methodology.

Google Search Console now documents a Generative AI performance report for organic impressions in AI Overviews and AI Mode. It supports views such as page, country and device. Check your property's access and available data. This is Google-specific information, not a report on ChatGPT, Claude or Grok. Google's report documentation.

Bing Webmaster Tools offers AI visibility reporting, including citation-related insights for its covered experiences. Read the report's definitions before comparing its figures with another tool's score. Bing's AI visibility reporting update.

Website analytics, order data and your CRM answer different questions: who arrived, what they did and what happened to the purchase or enquiry. For paid campaigns, consistent campaign tags help distinguish the source and campaign. Google Analytics campaign-tag guidance.

4. Follow the journey through to a customer outcome

Our recommended monthly scorecard includes:

• Relevant unbranded appearances across the fixed question set.
• Incorrect descriptions found and corrected at the source.
• Identifiable AI-referred website visits and purchase or enquiry actions.
• For service businesses: enquiries that match the service area and project requirements, quotes issued, booked projects and recorded value.
• For online stores: verified purchases and attributed revenue. Where the data supports it, include customer acquisition cost and return on ad spend for paid activity.

Keep revenue distinct from profit, and label attribution limits rather than assigning every sale to AI.

Ask customers how they found you, but allow multiple answers. Someone may see an ad, discuss options with ChatGPT, search your name and then call. The last website visit cannot always explain the whole decision.

Use “unknown” where attribution is genuinely unknown. Reclassifying every unexplained sale or enquiry as AI success makes the report look stronger and the decisions worse.

A useful way to calculate your own sample rate

Suppose your business appears in 8 of 20 selected test responses. Your observed appearance rate for that test is 40%.

That is an illustrative calculation. It does not mean you reached 40% of the market or that 40% of Australians will see you. Repeat the test under consistent conditions and examine whether changes are meaningful over time.

What should an agency report show?

Expect the question set, platform coverage, evidence, limitations, work completed and next priorities. The report should explain which pages need improvement and how purchase outcomes or enquiry quality are changing.

Elite's AI Visibility service includes ongoing monitoring alongside website and content improvements. Where paid AI advertising is included, we keep paid results and organic visibility distinguishable while looking at the overall path to a purchase or enquiry.

Book your Elite Consult to discuss a measurable Elite AI Package built around your business.

Related reading: AI Ads or AI Visibility: what should you prioritise?.