How are mention rate, citation rate and visibility score calculated, and what is the denominator?

Mention rate is the share of unbranded-prompt observations (the 44-prompt group) that produced a normal answer, counted as unique prompt × platform pairs, in which the answer text names the brand, including aliases and the brand's website domain. Citation rate is the share of observations whose answer carries sources. Visibility score is mention rate × 0.7 plus ranking quality × 0.3. All three follow one rule: the denominator holds only reviewed, unbranded observations that produced a normal answer; every other status is excluded and recorded separately.

Mention rate: denominator and numerator

• Denominator: of the 50 prompts per cycle, the 44 unbranded discovery and comparison prompts that produced a normal answer on each platform, counted as unique prompt × platform; each prompt runs once per platform by default.
• Numerator: those observations whose answer text mentions the brand. Aliases and the website domain count as mentions so spelling variants are not missed.
• Excluded: the 6 branded fact checks, refusals, not-triggered, failed, blocked and unknown observations.
• Only reviewed observations enter the denominator; anything that cannot be verified is flagged as pending, kept out of the denominator and never stated as fact.

Citation rate, ranking and sentiment

Citation rate is the share of observations whose answer includes sources. Source URLs are stored one by one and are not deduplicated across prompts, so a page cited in several prompts is counted for each, which is exactly what a content team needs for remediation. Ranking or position records the order in which the brand appears in the answer; sentiment is a positive, neutral or negative label applied by a human reviewer, not an automatic model score. The visibility score weights mention rate at 0.7 and ranking quality at 0.3 and is meant for comparing cycles within one project, not for comparing brands.

Status and the denominator

Branded fact checks stay out because a branded prompt names the brand and the AI will almost always mention it; counting those 6 prompts would inflate the numerator. Their job is to check whether the AI describes the brand's facts accurately, and their findings appear in the fact-check section. Real exposure is being mentioned when the buyer has not asked for you by name, which is why only unbranded prompts form the denominator.

Runtime-exception rate and evidence completeness

Runtime-exception rate per cycle is runtime exceptions (failed plus blocked) divided by planned tasks; evidence completeness is the share of tasks with complete evidence. Both are quality indicators rather than business metrics: a high exception rate means the cycle's denominator shrank and the trend needs care, and low completeness means some observations lack one of the nine evidence items. The monitoring log shows both figures for every cycle.

Does the denominator change from cycle to cycle?

Yes. It is the number of unbranded observations that actually produced a normal answer in that cycle, so cycles with more failures or not-triggered results have a smaller denominator. Always read the trend together with the denominator.

Can I compare visibility scores across brands?

Not meaningfully. The score depends on each project's own frozen prompt set, and different brands ask different questions. Use it to compare cycles within one project.

Is a recommendation slot a separate metric?

Ranking or position records where the brand appears in the answer and feeds the ranking-quality part of the visibility score. Miaowa GEO does not promise any uplift in mention rate or ranking.

Last updated 2026-09-12. The capabilities page and the pricing page are authoritative where they differ.