More buyers now start with an AI assistant instead of a search box. They ask ChatGPT, Claude, Gemini, or Perplexity to recommend a vendor or compare options, and the assistant answers with a short list of names. If your business is not on that list, you never find out - there is no impression count and no "page 2" to climb.
An AI visibility audit is the disciplined act of checking what those assistants say about your category, your brand, and your competitors, so you can see the gap and close it. You do not need a tool or a budget to start. Here is a method you can run yourself in an afternoon with the free versions of the major assistants and a spreadsheet.
What "AI visibility" actually measures
It breaks into three things worth measuring separately: presence (does the assistant name you at all), accuracy (is what it says correct and current), and positioning (does it frame you as a strong option or an afterthought). Being mentioned with out-of-date pricing is a different problem from not being mentioned at all - and each has a different fix.
Set up a simple scoring sheet
One row per test question, one column per assistant (ChatGPT, Claude, Gemini, Perplexity). In each cell record a 0-2 score: 0 not mentioned, 1 mentioned but with an error or only in passing, 2 named clearly and described accurately. This turns a fuzzy impression into a number you can compare and re-check next quarter.
The 7 steps
1. List the questions your buyers actually ask. 8-12 real buyer questions, not brand searches: "best [service] in [city]", "alternatives to [competitor]", "who offers [capability]". These are where you show up or lose the buyer silently.
2. Ask each engine your category questions. Run every question through all four assistants in a fresh chat so earlier answers do not bias the next. Score each 0-2. Do not lead the model toward your name - you want the unprompted answer a real buyer gets. Note the sources Perplexity and Gemini cite.
3. Test brand accuracy directly. Ask each engine about your business by name: what you do, your prices, whether you are reputable. This catches the expensive errors - wrong pricing, an old address, a dropped service, or a confident "I don't have information about that company."
4. Check who shows up instead of you. For every question you scored 0 or 1, record which competitors the assistant named. The same two or three names keep appearing - those are winning the AI recommendation, and what they have that you do not is the shortest path to your fix list.
5. Find the sources the engines trust. Map the pages feeding AI answers about your category: review platforms, directories, industry roundups, authoritative blogs. If your business is absent or thin there, that absence is why the assistants cannot name you confidently.
6. Score, total, and spot the pattern. Low category scores but decent brand accuracy = a discovery problem (invisible to new buyers). High presence but frequent errors = an accuracy problem (stale data). Zero across the board = a foundational gap (little structured, extractable info about you on the open web). Each points to a different first move.
7. Turn gaps into a prioritized fix list. Usual high-impact, low-effort wins: clear, extractable service and pricing pages; claimed and completed directory and review profiles the engines cite; FAQ and comparison content answering the exact questions you tested; basic structured data. Re-run the same sheet in 60-90 days to confirm the scores moved.
What a DIY audit does and does not give you
Running this yourself costs nothing, builds real intuition, and produces an action list. Its limits are time and repeatability: answers vary run to run, covering four engines by hand takes hours, and it is easy to miss patterns across dozens of cells. If you would rather not spend the afternoon - or you want a consistent, sourced snapshot you can repeat on a schedule - a structured audit does the same work systematically and returns a scored report with a fix plan.
This method and a free AI visibility snapshot are at a3eecosystem.com/audit. Full walkthrough: the original article.
Top comments (0)