How to run an AI visibility audit that leads to action

Step 1: build a buyer-question set

The audit should begin with the questions a buyer would ask before choosing a company. Include category questions, alternatives, use cases, problems, comparisons, price considerations and location-specific needs when location matters.

Do not rely on one phrasing. Muck Rack's research shows that the way a question is framed can change the sources an AI platform cites. Group several natural variations under the same buyer intent so the audit measures a pattern instead of a single screenshot.

Label each prompt by intent and decision stage. A broad discovery prompt should not be averaged with a high-intent comparison question without preserving the difference.

• Best company or product for a specific need
• Alternatives to a known competitor
• Category comparison and buying criteria
• Problem and solution questions
• Trust, safety, price and proof questions

Step 2: record more than a yes or no

A mention count is useful, but it is not enough. Capture whether the brand is recommended, neutrally listed, described inaccurately or omitted. Record the competitors that appear, the order of appearance and the sources attached to the answer.

Run the same question set across the platforms your customers actually use. ChatGPT, Gemini, Google AI Mode, Claude and Perplexity should not be treated as one engine because their retrieval and citation patterns differ.

Save the date, location, account state and exact prompt. Answers may vary, so the observation needs enough context to be repeated and interpreted later.

• Mentioned: yes or no
• Recommendation position
• Description and sentiment
• Linked or cited source
• Competitors present
• Material factual errors

Step 3: map the source gap

Now compare the pages supporting competitors with the pages supporting your brand. Some gaps belong on the company website. If the product page is vague, outdated or blocked from crawling, fix the owned source first.

Other gaps live on third-party pages. A competitor may appear in a relevant buyer guide, directory or review profile while your brand is absent. Those pages become placement opportunities only after they are checked for relevance, credibility, accuracy and contactability.

Mark each opportunity as owned, earned or uncertain. That simple field stops technical fixes, content work and publisher outreach from being mixed into one vague recommendation.

Step 4: turn findings into a 90-day plan

Prioritize actions by expected usefulness and level of control. Technical fixes and fact-sheet updates are usually within the brand's control. Publisher outreach is controllable as an activity, but acceptance is not.

Create a baseline, repeat the same core prompts monthly and record changes. Keep exploratory prompts separate so the baseline remains comparable. The report should explain what moved, what did not and what the team will do next.

A 90-day plan should name the owner, the evidence needed, the target page and the next review date. If the plan ends with 'improve visibility,' the audit has not done its job.

Sample enough to see a pattern

One answer is an anecdote. A useful audit groups prompt variations, platforms and repeated runs so the team can see which outcomes persist. The sample does not need to be enormous, but it needs to be consistent.

Start with 25 to 40 prompts across five to eight buyer intents. Run them on the platforms that matter to the audience. Repeat the baseline monthly, then use exploratory prompts to investigate new questions without changing the core set.

When an answer changes, inspect the supporting pages before writing a conclusion. A new citation, a page update or a model change can all alter the output.

Report decisions, not screenshots

A screenshot can prove what appeared at one moment. It cannot explain whether the result is common, important or actionable. The report should connect each finding to a decision.

Use a simple matrix: keep, fix, pursue or watch. Keep accurate owned pages. Fix missing or outdated facts. Pursue relevant outside pages. Watch volatile prompts until a pattern appears.

The final page should separate observations from recommendations. That distinction makes the audit honest and helps the client understand what the agency can control.