How to Test AI Visibility for a Local Business

An AI visibility check asks whether a business appears in a defined set of answers. This guide provides a sampling method and results worksheet. If an answer contains a specific factual error, use the separate AI citation correction guide.

Define a small prompt set

Choose questions a prospective customer might ask. Separate discovery questions, such as who offers a service in an area, from branded questions about a named company. Start with a few questions you can repeat and document carefully.

Keep the wording fixed for comparison. A change from “pool maintenance” to “emergency pool repair” changes the task, even when the location is the same. Record the exact words rather than a summary of the prompt.

Record the conditions

For every observation, note the platform, date, visible model or product mode, account state and location supplied in the prompt. Where the product exposes a search setting, record that setting too.

Citations or a searching indicator provide evidence about the visible response, but they do not reveal every internal source. Missing citations alone do not prove that an answer came only from training data. A cited page can also be old, incomplete or incorrectly summarized.

If you compare a signed-in session with a private session, label both. Neither is a universal representation of what all prospective customers see. Features and access may differ between sessions.

Use a consistent results sheet

Keep one row per prompt and response. Preserve a screenshot or saved response when permitted by the product and relevant to your review.

Treat an unavailable product, failed search or refusal as a separate result. Do not silently count those as business absence.

Repeat and report the sample size

Run the same prompt set on more than one date. Choose a schedule that fits the decision: a baseline before source corrections and a later review can be more useful than repeatedly refreshing one answer.

For a summary, report both counts and conditions. “Named in four of twelve recorded answers” describes the sample. “One-third of customers will find us” makes a claim the test cannot support.

Presence can vary between responses. Absence across the tested prompts means absence in that sample; it does not prove the business is absent from every answer or platform. Keep ambiguous mentions separate when a company shares a name with another business.

Verify the facts behind a mention

Open the source links and compare the relevant passages with the response. Check identity, service area, contact information and current services. A mention can be accurate without linking to your website, and a citation can exist without supporting the exact claim made.

OpenAI's explanation of answer reliability notes that responses can be incorrect and can contain fabricated references. Apply the same verification habit to each platform rather than treating a confident answer as evidence.

Choose an action the evidence supports

Correct an outdated fact on a source you control. Add useful service information where the website leaves a real customer question unanswered. For third-party records, use the publisher's normal correction process. Track the request and its outcome separately.

Do not manufacture pages merely to repeat prompt wording. Clear information can help readers and discovery systems understand a business, but it cannot compel inclusion in an AI answer.

Measure referrals separately

Record attributable website visits, inquiries and customers in a separate business report. Referrer data can be missing, and a visible mention does not establish a visit. A customer's statement about how they found the business is useful evidence when recorded with its limitations.

Use this test as a dated observation log. It can show changes within a consistent sample and identify factual problems to investigate. It cannot establish a universal AI rank or guarantee future recommendations.