You can run a useful first AI visibility audit without a complicated score or a hundred prompts. Set aside an hour, choose a handful of questions that could lead to business, save the answers properly, and look for repeated gaps. The checklist below is designed to end with a decision, not a thicker report.

Before you open an AI tool

Write down the decision you want to understand. ‘Does ChatGPT know us?’ is too loose. ‘Would a homeowner looking for emergency drain cleaning in Dieppe see us as a suitable option?’ gives you a service, customer, place, and reason to care.

Choose one or two AI products your customers are likely to use. You can widen the audit later. Starting everywhere at once creates more screenshots than insight.

  • Confirm the services and locations you genuinely want to test
  • List the important facts you can verify, including hours, phone number, credentials, and service limits
  • Choose a small set of real competitors, not every business in the category
  • Create one place to save prompts, answers, dates, settings, and source links

Build six useful questions

Use questions a customer could reasonably ask while choosing whom to contact. Keep the language natural. A useful set usually covers category discovery, a specific problem, comparison, trust, fit, and location.

For a plumbing company, one question might be: ‘Which plumbers serve Riverview after 8 p.m. and handle sewer backups?’ That is only an example. Replace it with the work, geography, and constraint that matter to your business.

  • Who provides this service in the area?
  • Who handles this particular problem or type of customer?
  • How do the credible options differ?
  • Which provider has the qualification or proof this job requires?
  • Which option fits the customer's timing, budget, or service constraint?
  • What should a buyer check before hiring a provider in this category?

Record the test so you can repeat it

Run the question as written. Note the product, date, location setting, and whether the answer shows signs of a web search. Keep the complete response, not just the sentence that names your company. Open and record every source the interface shows.

Do not regenerate repeatedly and keep the friendliest version. If you choose to run each question more than once, decide the number before you begin and preserve every valid result. The point is to see variation, not edit it out.

  • Exact question
  • AI product and mode
  • Date and location conditions
  • Full answer or permanent capture
  • Businesses named and the role each one receives
  • Visible citations and source pages

Score the outcomes separately

A single visibility percentage hides too much. Read each answer and record a few plain outcomes. Was the business named? Was it presented as a genuine option? Were the important details accurate? Did the answer link to the business, to an outside source, or to neither?

Keep a note for context. ‘Mentioned only as the expensive option’ is different from a neutral mention. ‘Recommended for Moncton, although we only serve Saint John’ is visible but wrong. A simple yes-or-no field cannot carry those distinctions by itself.

  • Mentioned: the business name appears
  • Recommended: it is offered as a suitable choice for the request
  • Cited: a visible source link supports the business or claim
  • Accurate: important facts match verified information
  • Competitive fit: the other businesses actually serve the same need

Read the evidence behind repeated gaps

One odd answer may not justify any work. Look first for problems that recur across useful questions or repeated runs. Then inspect the pages behind them.

If competitors appear, compare the information a customer can verify. Are their services and locations clearer? Do professional registers, association pages, case studies, or reviews support claims that your site leaves unproven? Is your own website accurate but an important profile is out of date?

Separate a missing fact from a missing reputation. You can add a forgotten service area to the right page today. You cannot manufacture years of credible work, and you should not pretend to.

Choose the smallest action that resolves the gap

Give priority to an issue when it affects a real customer, appears more than once, and can be corrected honestly. Write the proposed action beside the evidence that supports it.

  • Wrong hours in a visible directory: correct that listing
  • A service is offered but hard to find: improve the relevant service page
  • A qualification is claimed but not verifiable: add the real credential and link to an authoritative register where appropriate
  • The business is missing outside its service area: take no action
  • The result varies without a stable pattern: collect more evidence before changing anything

Set the follow-up before you finish

Record what changed and when. Allow a reasonable discovery window, then repeat the original question set under similar conditions. Compare the full set, including answers that stayed flat or became worse.

A later improvement does not prove that one edit caused it unless the test isolated that edit. It does show whether the business is moving in a useful direction. If nothing changes, revisit the diagnosis before adding more work.

Sources

  1. OpenAI: ChatGPT Search inclusion guidance
  2. Google Search Central: AI features and your website