How SOV Tracker measures AI visibility

SOV = valid mentions of the target brand ÷ (valid mentions of the target brand + valid mentions of every configured competitor) × 100, rounded to a whole percent.

Illustrative example: in one weekly measurement the answers name the target brand 6 times, competitor A 9 times and competitor B 5 times. SOV = 6 ÷ (6 + 9 + 5) = 6 ÷ 20 = 30%. Names outside the configured competitor set are not added to the denominator.

• Valid answers came back, a competitor set exists and competitors were named but the target brand never was: SOV is a measured 0%. If no brand in the set was named (0/0), SOV is not measured.
• No competitor set is defined, or no valid answer came back: SOV is not measured and no number is shown.
• SOV is a share inside one scope. 30% in one country and language cannot be added to or averaged with a result from another scope.

Sampling: in the weekly measurement each tracked question is asked once on each engine, so a week holds one answer per question and engine. One answer is one observation; AI answers can change from run to run, which is why a week is compared only with earlier compatible weeks.

Aggregation: SOV is pooled. Within a single measurement scope and on healthy engines only, on non-branded questions, valid mentions of the brand and of each configured competitor are first added up across every answered response in the window (all questions and engines), then divided once. It is not an average of per-answer or per-question percentages. Discovery Visibility and the other rates are likewise counted over all eligible answers in the window.

The free scan asks ChatGPT only, with 5 automatically generated questions that do not contain the brand name. It can run once every 30 days for the same business.

Its headline number is Discovery Visibility: in how many of the answered questions the brand appeared. It is not the SOV Score, which needs a configured competitor set and the weekly five-engine measurement.

Because the free scan's ChatGPT measurement does not return citation evidence, its citation and owned-site citation rates are not measured.

Each of the 5 questions is asked once. If fewer than 3 usable answers come back, the scorecard shows a low-sample warning and the result should be read as directional.

Some providers expose structured citation or grounding fields; when they do, those URLs are used as evidence. URLs written inside the answer text may also be kept for reference, but they are not counted as engine citations in Owned-Site Citation Coverage. That metric's denominator excludes only engines that do not support citation evidence and answers whose evidence could not be captured; an answer with captured evidence that does not cite the brand's site is counted as a miss, not dropped.

A citation means a URL was observed in the evidence available to the scan. It does not automatically mean the brand was recommended or reveal the provider's hidden retrieval pool.

No citation evidence ≠ no source

• No citation evidence: the engine does not return structured source data (ChatGPT, Claude) or no answer was received. The answer stays out of the denominator; it is not counted as zero.
• No source: an engine that returns source data (Gemini, Perplexity, Google AI Overviews) answered but cited no URL. The answer stays in the denominator and counts as a miss.

2026-10-15: the owned-site citation denominator is the answered responses from engines that return citation evidence; an answer that shows no source counts as a miss. Scorecards created before this date are shown recalculated under this rule.