How to Test Your Business Visibility in ChatGPT and AI Search

People may ask an assistant for a supplier, a company comparison, or advice about choosing a service. The answer may name businesses, cite websites, or simply explain how to shop. A visibility test records how your company appears for a controlled collection of questions.

It is not a universal rank tracker. Responses can change with the exact question, the user's location, the platform's current model and tools, and sources available at that moment. A company can be cited without being recommended, and mentioned without receiving a referral.

The goal is to identify incorrect information, missing evidence, and patterns worth investigating, not to manufacture a single 'AI rank' number.

Start with the language customers use on calls, in emails, or when comparing quotes. Include direct brand research, service category discovery, a problem-based question, and a buying comparison. Avoid writing only prompts designed to force the assistant to mention your company.

For a water treatment company, realistic examples could include 'Who installs water softeners near Davis?', 'How do I compare local water softener installers?', and 'What questions should I ask before hiring someone to fix hard water?' The informational question may produce no company names at all, which can be the correct result.

• Brand prompts: 'What does [business] do?' and 'Does [business] serve [area]?'
• Discovery prompts: 'Companies offering [service] in [location]'.
• Comparison prompts: '[Service A] versus [Service B] for [problem]'.
• Trust prompts: 'What should I verify before hiring [category]?'
• Process prompts: 'How much might [service] cost and what changes the price?'

For each query, record platform, date, exact text, relevant location, whether search/browsing was enabled, and any visible sources. Capture the answer or a summary without editing it after the fact.

If a result is personalized or generated from a prior conversation, note that. A clean, comparable test should avoid changing accounts, devices, location settings, and prompts without marking the difference.

Do not call a result 'not found' because you saw no mention once. Record 'not observed in this test.' Likewise, a mention is an observation on that run, not a guaranteed future placement.

A mention means the assistant named your business. A citation means the answer visibly points to a source associated with the claim. A referral means a visitor actually came to your site and can be identified through available analytics. These are different measurements.

Check whether the assistant gave correct business details, services, and geographic coverage. Errors can be more urgent than the absence of a mention because wrong phone numbers or misrepresented services may mislead customers.

Search Console performance covers activity in Google's search ecosystem; it does not provide a universal tally of every independent assistant's mention of your brand. Do not add Google impressions to manual mentions as though they measure the same thing.

When another company appears repeatedly, check what public evidence may be supporting it: service pages, project photos, relevant listings, news, accurate business information, or specific guides.

Do not assume the assistant is rewarding one SEO tactic. The presence of a competitor may reflect the test wording, context, and available sources. Treat it as a research lead to verify against actual public pages.

Prioritize fixable gaps: missing or contradictory service details, weak contact information, broken indexing, or information that only exists in hard-to-crawl elements.

Repeat representative prompts on a consistent cadence after making a substantive website or citation improvement. Keep the old results and compare the same conditions. A change over several checks is more informative than one fortunate answer.

Also track commercial signals where possible: branded search queries, qualified site visits, incoming calls asking about an assistant, and booked work. A higher mention rate without more accurate information or qualified contact may not help your business.

End each review with one action supported by evidence, such as clarifying a service page, correcting a legitimate listing, publishing a useful process explanation, or repairing a technical indexing issue.

Write a one-page summary: the question set, platforms, dates, locations, the percentage of tests where you were mentioned, the percentage where you were cited, major accuracy errors, and three site improvements to consider.

Include a footnote that the test uses representative prompts and snapshots rather than a platform-wide ranking. Do not guarantee that a company will appear in future AI answers.

AI Deployment Partner's visibility check follows this type of evidence-first approach. We can use the findings to prioritize content, legitimate public business information, and technical fixes rather than selling a made-up ranking score.