How to Run an AI Visibility Audit

Before you can improve how AI engines talk about your brand, you need to know what they currently say. An AI visibility audit answers three questions: do the major assistants mention you when it matters, is what they say accurate, and who are they recommending instead? Here's the full process — no tools required beyond the assistants themselves and a spreadsheet.

Step 1: Build your question set

The audit is only as good as its questions. Build a list of 50–100 questions across four types:

• Problem questions — what buyers ask before they know solutions exist ("how do I keep leads from going cold").
• Category questions — "best [category] tools for [segment]", "how to choose a [category] platform".
• Comparison questions — "[you] vs [competitor]", "alternatives to [competitor]".
• Brand questions — "what does [your brand] do", "is [your brand] worth it", "[your brand] pricing".

Pull these from sales call notes, support tickets, search query reports, and your own team's knowledge of how customers actually talk. Write them the way a person types, not the way a marketer would.

Step 2: Run the questions

Run every question through at least four surfaces: ChatGPT, Perplexity, Gemini, and Google (checking whether an AI Overview appears and what it says). Use clean sessions — logged out or fresh chats — so personalization doesn't skew results.

For each answer, record:

1. Mentioned? — yes/no, and position (first recommendation, listed among others, absent).
2. Accurate? — flag anything wrong about your pricing, features, positioning, or availability.
3. Sentiment — recommended, neutral, or caveated ("however, users report...").
4. Cited sources — which URLs the answer drew from, when shown.
5. Competitors named — who appears and how often.

Expect variance: the same question asked twice can produce different answers. Run important questions two or three times and score the pattern.

Step 3: Score what you found

Compute three numbers:

• AI answer share — the percentage of category and problem questions where you're mentioned. See AI answer share for the full method.
• Accuracy rate — the percentage of brand-question answers with no factual errors.
• Citation map — which of your pages (or which third-party pages) the engines cite when they do mention you.

The citation map is the strategic gold: it tells you which content is doing your GEO work. Teams are routinely surprised to find an old comparison post or a third-party review carrying most of their AI visibility.

Step 4: Turn findings into actions

Sort findings into four buckets:

• Factual errors → fix the source. Trace where the wrong fact lives (stale pricing page, outdated third-party listing, old press mention) and correct or outrank it. Treat errors machines repeat as P1 content bugs.
• Absent from category questions → content gaps. You need citable pages answering those exact questions — see the GEO content checklist for what makes a page citable.
• Competitor-dominated comparisons → build honest comparison content. Engines cite pages that compare fairly and specifically; vendor puffery gets skipped.
• Weak citations → strengthen what's working. Pages already being cited deserve updates, better structure, and fresher data before you build anything new.

Step 5: Make it a loop

A one-off audit decays fast. Re-run the question set monthly, track the three scores over time, and add new questions as your category shifts. If the manual run becomes a burden, this is the point to evaluate GEO tracking tools — you'll now know exactly what you need them to do.

The first audit usually takes a day. The recurring one takes an hour. Few hours of marketing work this year will tell you more about where your next customers are actually forming their first impression of you.