Visibility Audit
What It Does
AI Visibility Checker tests whether your brand is cited by AI search engines (ChatGPT, Perplexity, Gemini). It also audits robots.txt, llms.txt, schema readiness, and crawlability - returning a scored grade.
Running a Check
1. Open Tools → AI Visibility Checker. 2. Enter your brand name and website URL. 3. Click Check - the tool queries multiple AI engines and returns citation scores.
Improving Your Score
- Low entity score → run Entity Stack - Missing llms.txt → create one (the tool provides a template) - Poor schema → add Organisation schema from Entity Stack output - No AI citations → launch a CloudStack or DocStack campaign
How we measure it (and what the numbers mean)
AI answers are not deterministic. Ask the same question twice and you can get different businesses named — that is a property of the engines, not a fault in the measurement. Everything below exists so the numbers stay honest about that. EVERY PROMPT IS ASKED OF EVERY ENGINE. Your prompt set is put to each engine you have configured, so per-engine results are directly comparable — same questions, same run. A blended score would hide that you can be strong in one engine and invisible in another. THE 95% RANGE, NOT JUST THE PERCENTAGE. Each rate is shown with a confidence range and the number of prompts behind it. "Named in 60%" from 10 prompts and from 100 prompts are not the same claim, and the range is what tells them apart. A wider range means fewer samples, not a worse result. ASK EACH PROMPT MORE THAN ONCE (optional). On the run form you can ask each prompt once, 3 times, or 5 times. This measures something the confidence range does not: • COVERAGE — how sure we are of a rate across your prompt set. That is what the range shows. • STABILITY — whether one prompt gives the same answer when asked again. They answer different questions and are reported separately. A run that asked once shows no stability figure at all, because nothing about repeatability was measured — we would rather show nothing than a reassuring "100%". Repeats multiply the API calls on your own key, so the cost is stated on the form and it is never raised for you automatically. Scheduled runs stay at one ask per prompt for the same reason. WHERE THE ANSWER CHANGED. When you use repeats, the report lists the exact prompts whose answers disagreed between attempts, with counts rather than percentages. Those are usually the most useful lines in the report: they are the questions where the engine has not made its mind up about you. WHY YOUR HISTORY DOES NOT ALWAYS JOIN UP. Every run records which engines it measured. Runs on different engine sets are deliberately not plotted on one line — changing engines is a change of method, not a change in your visibility. Adding an engine starts a fresh trend line and your existing history stays under the old set. Runs that stopped early (a provider out of credits, for example) are excluded from trends rather than scored, because a partial sample is skewed towards whichever engines did answer.