How Our AI Audit Works
What we ask, what we count, and what the numbers in your AI visibility report can and cannot tell you. Written so you can check our work.
We scan your site the way AI crawlers see it
The scan fetches your homepage like GPTBot does and checks six things: whether robots.txt lets 8 AI crawlers in (OpenAI, Anthropic, Perplexity, Google, Bing and others), your structured data (JSON-LD such as Product, Organization and FAQ), whether your content gives direct answers an assistant can quote, trust signals (reviews, about page, policies), llms.txt and your sitemap, and mobile speed through Google PageSpeed. Each check is scored and weighted into a 0 to 100 score with a letter grade. The fixes are ranked by how many points each one is worth to your score.
We write buyer questions without your brand name
From your site's own description we write 3 questions a shopper would type when they are still choosing, for example where to buy a product online or which brand is best for a use case. Your brand and domain are removed from the wording, because a shopper who does not know you cannot ask for you by name. We keep the same wording if we run the check again, so results can be compared.
We ask about you by name, separately
One more question names your brand directly and asks what you offer. This shows whether the assistant knows you exist and describes you correctly. We report it apart from the buyer questions, because a brand almost always looks better when asked about by name. The buyer-question number is the one that reflects what shoppers actually see.
Each question runs on its own, with live web search
Every question is sent as its own request with no chat history, so one answer cannot influence the next. We use Gemini with Google Search grounding, which returns the answer and the web pages it read. When live search is not available for a run, the report says so and the answers come from the model alone.
What we count
- Named: an answer counts as naming you when your brand or domain appears in its text or in one of its cited sources.
- Share of voice: every brand, store or maker the answers recommend, and how many of the answers named each one. Your brand is always in the list, even at zero.
- Sources cited: the websites the assistant read to write its answers, grouped by site, and whether your own site is among them. Being cited is not the same as being recommended.
What a free report is not
It is a snapshot from one assistant on one day. Answers change with model updates, location, phrasing and time; the same question can name different brands tomorrow. Three buyer questions are enough to show whether you are in the conversation, not enough to measure a trend. A score of 100 does not make an assistant recommend you, and nobody can promise that it will. What the report gives you is a clear picture of where you stand today and the fixes most likely to move it.
What we do with your request
Your email is used to send the report and for a reply from our team if you ask for one. We do not sell or share it. Scans are cached for 24 hours per site, and the AI answers for 7 days, so a repeat request within that window returns the same results.