AI visibility is best measured as a set of repeatable observations, not a single rank. Track what assistants say for relevant customer questions, whether the facts are right, and whether people can take the next step.
Get a baseline: Run Loop's free AI Visibility Checker for your business and location. Save the questions and results so the next check is comparable.
Choose a fixed group of questions
Pick six to ten prompts that represent real customers: broad category searches, specific needs, “open now” questions and booking intent. Include your neighbourhood and the wider city if both matter. Keep the exact wording and location constant in the core test. You can add new exploratory questions separately without corrupting the baseline.
For example, a hotel could track “quiet hotel near the station with parking” and “family hotel near the convention centre.” A restaurant could track “book vegetarian dinner near the theatre.” The testing guide shows how to build that set.
Record three different outcomes
- Recommendation: Was your business named for a suitable question?
- Accuracy: Were its location, hours, services and booking options described correctly?
- Action: Did the answer lead to a working site visit, enquiry or reservation path?
Keep results by assistant and date. A simple sheet with the prompt, location, answer, named businesses, cited links and errors is more useful than a composite grade alone. Note that results may vary between runs. No third-party tool has access to Google's internal ranking systems, as Google's AI Search guidance explains.
Separate visibility from customer outcomes
Count website visits, enquiries and confirmed bookings in your own analytics or booking system. Do not call a recommendation a sale, or treat a booking click as a completed reservation. Where attribution is unavailable, say so. This is particularly important when assistants send customers through other booking services.
If your site is hard to use after an assistant mentions you, see the AI-ready website checklist and Loop's free website assistant. The booking audit covers a full reservation test.
Review changes on a sensible schedule
Run the core test monthly or after a meaningful change to your business information. First correct factual errors; then improve pages for recurring customer questions. Give search and AI systems time to discover changed information. Look for a pattern across prompts and months rather than a one-off win.
Google offers a generative AI performance report in Search Console for its own search features. That data is different from a cross-assistant prompt test. Use each source for what it actually measures.
Read the full local business guide or diagnose a missing recommendation.
Design a baseline that you can repeat
A baseline is not the highest number of prompts you can run. It is a small, stable set of questions that represents real demand. Choose several categories: general discovery, specific need, time-sensitive request and booking intent. For each, define the location and any date or time. A request such as “hotel near the station” is different from “hotel near the station tonight with accessible parking.”
Keep the core set unchanged for several review cycles. Add a separate exploratory set when new services, seasons or customer questions appear. This prevents the report from looking better simply because you replaced difficult prompts with easier ones. The testing guide includes examples by business type.
Build a simple observation sheet
| Field | Why it matters |
|---|---|
| Date and assistant | Answers and products change over time |
| Exact prompt and location | Allows a fair repeat test |
| Was the business named? | Measures inclusion for that question |
| Was the answer accurate? | Separates helpful visibility from misinformation |
| Sources or links shown | Points to information you may need to correct |
| Available next action | Shows whether a customer can enquire or book |
| Notes and change date | Connects observed differences to real work |
Save the complete answer or a screenshot alongside the sheet when practical. A short label such as “included” can hide a serious error: the assistant may name you but describe an unavailable service. If two businesses share a similar name, note whether the answer identified the correct location.
Measure inclusion, accuracy and actionability separately
For a set of ten suitable prompts, you could report the number that named the business, the number with correct key facts, and the number with a working route to the next action. Give the denominator and the exact prompt set. Do not call these percentages “market share” or “rank.” They describe only the questions tested.
Accuracy is often the first improvement to pursue. A customer may make a wasted journey because the assistant gives old hours, or miss a booking because it points to a closed provider. Actionability is the next test: can a person confirm the right table, room or appointment? The website checklist and AI-ready website tool address that part of the experience.
Use Search Console and analytics for different questions
Loop's checker samples answers from assistants for local prompts. Google Search Console measures your site's performance within Google Search. Google announced a dedicated generative AI performance report for impressions in AI features, now available worldwide according to its 2026 update. That report is valuable for Google visibility, but it does not measure all mentions across ChatGPT, Gemini or other products.
Your website analytics and booking system answer still another question: did someone visit, enquire or reserve? Referral data may be incomplete when a customer changes devices or uses an assistant without a direct click. Use confirmed booking records for reservation totals, and say when the source of a booking is unknown. The AI-assisted bookings guide explains this attribution limit.
How to interpret changes after a fix
Suppose you add a page explaining accessible rooms and then appear in two more relevant answers. That is encouraging, but it does not prove the page alone caused the change. The assistant may have changed, another source may have been updated or the prompt may have been interpreted differently. Record the change date and inspect the cited or visible sources where possible.
Look for improvement across several prompts and review periods, particularly in accuracy and useful next steps. A temporary increase in mentions is less meaningful than a sustained reduction in wrong information. Do not test a change once and declare an SEO win. Google's AI Search guidance says even eligible, well-optimized pages are not guaranteed to be served.
A reporting format for a small business
Keep the monthly report to one page: questions tested; assistants and locations; inclusion count; significant factual errors; working booking routes; fixes completed; enquiries and confirmed bookings where known. Attach a few representative answers rather than a giant export. Make the next action explicit: correct hours, add a service page, repair a booking link or ask a provider about availability data.
For a DMO or district with many member businesses, aggregate only what operators agree to share. A member's visibility check is not automatically a regional performance measure. The member playbook explains a careful approach.
What if the prompt set needs to change?
Keep the original prompts in the historical report and start a new version of the test set. Explain why you added or removed a question, such as a new service, a moved location or a change in the season. Do not compare the raw inclusion rate from two different sets without noting the change. A business that starts offering dinner should add dinner prompts, but its old breakfast baseline can still be tracked separately. Versioning preserves a fair trend while allowing the test to reflect the business as it evolves.
Separate neighborhood and city-wide results
Distance and local context can change what a customer reasonably sees. Keep neighborhood, city-wide and visitor-origin prompts in separate groups. A vineyard may be relevant to someone planning a regional day trip but not to a person looking for a wine bar within walking distance downtown. Combining those prompts into one percentage makes the report hard to interpret. State the location context beside every result and compare like with like over time. That gives you a better basis for deciding whether to change your content, profile or booking journey.
Frequently asked questions
Is a third-party AI visibility score an official metric?
No. It summarizes the provider's sampled tests. Google's official guidance says third-party products do not have access to its internal ranking or AI systems. Keep the prompts and raw answers so you know what a score represents.
Should I count every AI mention as a lead?
No. A mention is exposure in a sampled answer. A lead requires a meaningful enquiry or another defined action. A booking is confirmed only when the reservation system says so.
How long should I wait after updating my site?
There is no universal interval. Search and assistant systems discover changes at different rates. Record the update date, repeat checks on a regular schedule and avoid overreacting to the first result.
What if I have no analytics setup?
Start with the prompt log and your existing enquiry or booking records. Add website analytics when you can use them reliably. Clear definitions matter more than a complicated dashboard.
Next step: Run and save your first AI visibility check.