Before you spend on content, tools or an agency, you need one fact: what do the engines say about you today? Without it, every plan is a guess, and every later result has nothing to be compared with.
This page gives the full method we use, including the scoring rubric most guides leave out, a log you can copy, and a worked example. You can run it yourself. If you would rather we do it, the last section says what you get.
01What an AI visibility audit is
An AI visibility audit is a structured test of how generative AI engines present your brand when buyers ask real questions. It is different from a classic SEO audit, which checks whether pages rank in a list. This one checks what is written in the answer.
It has three outputs: a baseline you can compare later results against, a diagnosis of why each gap exists, and a ranked list of fixes. A report with a score and no diagnosis is a measurement, not an audit.
02The method, step by step
Write the question set
Ten to thirty questions in six types: branded, category, best-for, comparison, pricing and use-case. Pull them from sales calls, support emails and Google's People Also Ask boxes. Include at least two that name a competitor.
Run them on several engines
Use ChatGPT, Perplexity and Gemini at minimum, each in a fresh session. Run each question more than once, because answers vary between sessions.
Log every answer
For each run, record whether you are named, whether a page on your site is cited, how the brand is described, and every source listed. Use the template further down.
Sort the sources
Put each cited URL into one of five groups: your own site, review sites, editorial coverage, community forums and competitors. Sources that recur across questions are the pages that shape your category.
Find the cause of each gap
Missing page, blocked crawler, conflicting facts or weak outside sources. We explain these under the causes heading below.
Rank the fixes
Order them by how many of your questions each fix touches.
Several guides we reviewed suggest ten to twenty prompts and a weekly rerun. We think more runs per question matters more than more questions, because the main weakness in any audit is that one run proves little.
03A log you can copy
| Question | Engine | Run | Named? | Cited? | Description | Sources listed |
|---|---|---|---|---|---|---|
| Best scheduling software for a small dental clinic | ChatGPT | 1 | No | No | n/a | Review site A, review site B |
| Best scheduling software for a small dental clinic | ChatGPT | 2 | Yes | No | Correct | Review site A, our site |
| What does [brand] cost? | Perplexity | 1 | Yes | Yes | Wrong price | Our old pricing page |
The rows above use an invented brand. What matters is the structure: one row per run, never an averaged row, so you can see the variation.
04How to score it
Most published audits tell you to check sentiment and accuracy and then give no rubric. These definitions are ours, and you should write yours down before you start so the scoring does not drift between rows.
| Measure | How we score a run | How we summarise |
|---|---|---|
| Named | Yes if the brand appears in the answer text, even without a link | Share of runs where Named is yes |
| Cited | Yes if a page on your domain is listed as a source | Share of runs where Cited is yes |
| Accuracy | Correct if price, product and location match your site; partly if one is off; wrong if two or more are off | Share of named runs marked correct |
| Tone | Positive if the answer recommends you, neutral if it lists you, negative if it warns against you | Count of each, shown as it is |
| Share of voice | Your mentions divided by all brand mentions across the runs | One percentage per question type |
| Consistency | Same result in at least four of five repeated runs | Count of questions that are stable |
A word on tools. Free checkers and paid trackers are useful for repeating runs quickly, but each defines a mention, a citation and a score its own way, and some do not publish how. If you use one, keep the log method alongside it for the first round, so that you know what the tool is counting before you rely on its numbers. When a tool and your own log disagree, trust the log, because you can see every answer in it.
05From gap to cause
| Gap | Cause to check first | How to check |
|---|---|---|
| Never named | A crawler is blocked, or no page answers the question | Read robots.txt and search your site for the question |
| Named with wrong facts | Conflicting facts across sources | List every place price and address appear |
| Cited on one engine only | Different sources or crawler rules | Compare the sources each engine listed |
| Cited but not named | Page used as a reference, brand not recommended | Compare how pages describe the brand |
| Unstable results | Thin corroboration across sources | Count independent sources that name you |
06A one-afternoon version you can run now
Do this before you buy anything
0 of 8 doneWhen you finish, do one more thing: read the five or six sources that appear most often and ask why each is there. A comparison page, a forum thread and a directory listing all say something different about how the engines learned about your category. The pattern tells you where your brand needs to be described, and it is usually a shorter list than people expect. It also shows which sources you can influence, such as your own pages and listings, and which you cannot, such as a competitor's blog.
What free audit tools and quick checks give you
- A score from a method you cannot see
- One run per prompt
- A dashboard, with no stated cause for gaps
- A prompt to buy more software
What our audit gives you
- Named questions and a log you can rerun
- Repeat runs and a range
- A cause for each gap
- A ranked fix list tied to your questions
07If you want us to run it
We run the same method on a larger question set, with screenshots of every answer and the sources they listed. You get the log, the scoring, the cause of each gap and a ranked fix list. It is also the first stage of our monthly work, which is described on the pricing page. We do not promise results from it. We promise that you will know what the engines say about you and why.
This suits you if
- You do not know what ChatGPT or Perplexity say about you
- You are about to spend on content or an agency
- Competitors keep appearing in answers and you do not
- You want a baseline to judge later results against
Skip it if
- You already have a recent, repeated test on a fixed question set
- You only need a quick look, for which the free checker is enough
- You want a guaranteed outcome
- Your site is not yet indexed
For a quick first look, use the AI visibility checker. To understand each measure in more depth, read how to measure AI visibility. If your gap is that ChatGPT never mentions you, see brand not showing in ChatGPT; if it states wrong facts, see wrong information in ChatGPT. Our case studies page shows how results are reported, generative engine optimization services covers the fixes, and the AI visibility hub gives the wider view.

