How Synthic measures AI visibility
Synthic measures brand mentions and native source citations in a recorded set of AI answers. Every result belongs to its exact question, collection method, model and time; it does not estimate visibility across all AI users.
What belongs in the collection record?
Preserve the exact question, provider and model identifier, date, original answer, retrieval mode, repetitions and available native source references. Imported answers retain their supplied provenance. API collection and consumer application observations must remain distinguishable.
Keep a stable research set when comparing periods. A changed question mix, engine set or retrieval mode can produce a changed result without a changed website. Prompt wording about a market does not establish a location-specific result.
How is the mention rate calculated?
Mention rate = eligible scored answers mentioning the brand ÷ all eligible scored answers × 100. A completed, relevant, scored answer without a mention counts as a zero. Failed, pending, off-topic and uncertain observations are kept separate from the eligible denominator.
The analyzer validates the scored output and requires a verifiable named quote for a claimed brand mention. A name appearing in an answer is not necessarily an endorsement. Inspect the answer when interpreting a mention.
How are citations different from mentions?
A mention names the brand. A native citation is a source reference supplied with the provider response; it may point to the brand’s website or a third-party source. A mention can exist without a citation, and a cited page does not establish that every nearby statement is supported.
The displayed owned-domain citation rate uses collected answers as its denominator: collected answers citing an owned domain ÷ collected answers × 100. The citation component of the composite score uses eligible scored answers instead. Inspect counts and denominators before comparing the two figures.
What is the visibility score formula?
Visibility score = 55 × mention fraction + 0.30 × mean prominence among mentions + 15 × owned-domain citation fraction among eligible answers. Prominence uses the analyzer’s 0, 25, 50, 75 or 100 scale. With no eligible scored answers, the score is unavailable rather than zero.
The score is a Synthic product heuristic on a 0–100 scale, not a validated measure of revenue, market share or all-user AI visibility. The separate mention, citation, sample-size and coverage figures explain more than a composite score alone.
A worked example
Illustrative data: 10 eligible answers, four brand mentions, mean prominence of 50 among those mentions, and two eligible answers citing an owned domain. The composite is 55 × 0.4 + 0.30 × 50 + 15 × 0.2 = 40 out of 100.
If two more requests failed, report 12 requested, 10 eligible and two failures. Keep the failed requests visible; do not turn them into two additional zero-mention answers. This example is arithmetic, not a Synthic customer result.
How should trends and uncertainty be interpreted?
Repeated answers can disagree. Compare results with the same question wording, model, retrieval method and repetition count. The displayed Wilson interval describes the observed mention proportion under statistical assumptions; correlated prompts and repeated model responses limit how broadly it can be interpreted.
After changing a page, retain the original version, change date and later comparable observations. A later rise in mentions is an observation, not proof the edit caused it. Check provider changes, question mix and unrelated source changes before drawing a causal conclusion.