How to Run Your Own AI Visibility Audit: A Free 7-Step Method for 2026
More buyers now start with an AI assistant instead of a search box. They ask ChatGPT, Claude, Gemini, or Perplexity to recommend a vendor, compare options, or explain a category, 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 simply 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. This is a method you can run yourself in an afternoon with nothing but the free versions of the major assistants and a spreadsheet. Below is the exact process, the prompts to use, and a simple way to score the results so you end with a prioritized fix list rather than a vague worry.
What "AI visibility" actually measures
AI visibility is how often, how accurately, and how favorably AI answer engines mention your business when someone asks a question your business could answer. It breaks into three things worth measuring separately: presence (does the assistant name you at all), accuracy (is what it says about you correct and current), and positioning (does it frame you as a strong option or an afterthought). A brand can score well on one and badly on another - being mentioned with out-of-date pricing is a different problem from not being mentioned at all, and each has a different fix.
Before you start: set up a simple scoring sheet
Open a spreadsheet and make one row per test question and one column per assistant (ChatGPT, Claude, Gemini, Perplexity). In each cell you will record a 0-2 score: 0 if you are not mentioned, 1 if you are mentioned but with an error or only in passing, and 2 if you are named clearly and described accurately. This turns a fuzzy impression into a number you can compare across engines and re-check next quarter.
The 7 steps
Step 1 - List the questions your buyers actually ask
Write 8 to 12 real buyer questions, not brand searches. Think category and problem language: "best [your service] in [your city]", "alternatives to [a well-known competitor]", "how do I [the problem you solve]", "who offers [specific capability]". These are the prompts where you either show up or lose the buyer silently. Brand-name questions ("what is [your company]") come later - they test accuracy, not discovery.
Step 2 - Ask each engine your category questions
Run every category question through all four assistants, one at a time, in a fresh chat so earlier answers do not bias the next. Paste each answer into your sheet and score it 0-2. Do not argue with the model or lead it toward your name - you want the unprompted answer a real buyer would get. Perplexity and Gemini will often cite sources; note which sites they pull from, because those are the pages influencing your category.
Step 3 - Test brand accuracy directly
Now ask each engine about your business by name: "What does [your company] do?", "What are [your company]'s prices?", "Is [your company] reputable?" Score these too. This is where you catch the expensive errors - wrong pricing, an old address, a service you no longer offer, or a confident "I don't have information about that company," which tells you the assistants have nothing reliable to draw on.
Step 4 - Check who shows up instead of you
For every category question where you scored 0 or 1, write down which competitors the assistant named. A pattern will emerge fast: the same two or three names keep appearing. Those competitors are winning the AI recommendation, and looking at what they have that you do not - clear service pages, third-party reviews, directory listings, comparison content - is the shortest path to your fix list.
Step 5 - Find the sources the engines trust
Look at the citations Perplexity and Gemini gave, and ask ChatGPT and Claude "what sources would you use to answer that?" You are mapping the pages that feed AI answers about your category: review platforms, directories, industry roundups, Wikipedia-style references, and a handful of authoritative blogs. If your business is absent or thin on those specific sources, that absence is why the assistants cannot name you confidently.
Step 6 - Score, total, and spot the pattern
Add up your sheet. A low total on category questions but a decent brand-accuracy score means people who already know you get good answers, but you are invisible to new buyers - a discovery problem. High presence but frequent errors means an accuracy problem: the data feeding the models is stale. Zero across the board usually means a foundational gap - little structured, extractable information about your business exists on the open web. Each pattern points to a different first move.
Step 7 - Turn gaps into a prioritized fix list
Rank fixes by impact and effort. The usual high-impact, low-effort wins: publish or clean up clear service and pricing pages with plain, extractable language; claim and complete the directory and review profiles the engines cite; add FAQ and comparison content answering the exact category questions you tested; and make sure your site has basic structured data so machines can read it. Re-run the same sheet in 60 to 90 days to confirm the scores moved - AI answers change as the underlying sources change.
What a DIY audit does and does not give you
Running this yourself is genuinely valuable: it costs nothing, it builds intuition for how the engines see your category, and it produces a real action list. Its limits are time and repeatability. Answers vary run to run, covering four engines across a dozen questions 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 hand to a team or repeat on a schedule - a structured audit does the same work systematically across all four engines and returns a scored report with a fix plan. Either way, the goal is the same: stop guessing whether AI recommends you, and start measuring it.
See where you stand in minutes
Run a free AI visibility snapshot to check whether the major assistants name your business - then decide whether to work the fix list yourself or get the full, sourced AI Visibility Audit with a prioritized plan.
Get your free AI score See the AI Visibility AuditFrequently Asked Questions
How long does a do-it-yourself AI visibility audit take?
Plan for two to three hours to test a dozen questions across ChatGPT, Claude, Gemini, and Perplexity and score the results. The scoring sheet is what keeps it from sprawling - once it is set up, each question is a quick paste-and-rate, and the totals point you straight at the biggest gaps.
Which AI engines should I check?
Start with the four that carry most consumer and business queries in 2026: ChatGPT, Claude, Google's Gemini and AI answers, and Perplexity. The free tiers are enough for an audit. Perplexity and Gemini are especially useful because they show their sources, which tells you which pages are shaping answers about your category.
Why do the answers change every time I ask?
AI assistants are probabilistic, so wording shifts between runs, and their answers also change as the web sources behind them change. That is why you score presence and accuracy rather than memorizing exact phrasing, and why re-running the same question set on a schedule is more useful than a single check.
Is a free DIY audit enough, or do I need a paid one?
A DIY audit is a real, useful first step and often all a small business needs to find its top few fixes. A paid audit earns its cost when you want consistency, full coverage across engines and competitors, and a repeatable sourced report you do not have to assemble by hand. A sensible sequence is to run the free method first, act on the obvious wins, and use a structured audit when you want to measure progress rigorously.