How we measure AI visibility

Eligible audits. Each audit on doiexist.ai asks ChatGPT, Claude and Gemini the questions a buyer in the business’s category would ask. An audit counts when it completed, every engine answered, it was not an internal test and the business had a real category. Sectors are the categories doiexist.ai groups audits into.

Buyer questions. Only the questions a buyer asks about the category are counted. Questions that name the business are left out. The share counts only audits with at least one buyer question left.

Never named. An engine never named a business when none of its answers in the audit named it. “Not named by any AI engine” means that held for all three engines.

Two failure rates. Per question (the benchmark’s sector tables): a question fails when no engine named the business in its answer; the rate is failed questions over questions asked. Per answer (its engine table): each engine is counted on its own; the rate is that engine’s answers that did not name the business over its answers.

Why every figure is a floor. Until October 20, 2026, engine answers were stored cut at 2,000 characters. A cut answer is counted as naming the business, so every share and rate is a floor: the true figure is at least the one shown.

Audit floor and updates. No sector or question type is shown below 20 audits. Figures are whole percents. The benchmark shows a frozen snapshot; a snapshot’s figures do not change once shown, and a new one appears only when it is approved. The figures do not rank engines, give a cause or predict anything.