CitadexDocs

Scores and Metrics — Visibility, Recommendation Share, Recommendation Levels, Ranges, Citations and Sentiment

How each metric is calculated: the Dashboard visibility score, mentions, citations and share of voice, the mention-to-recommendation rates with 95% ranges, weighted recommendation share, the recommendation level of each answer, coverage and position, sentiment, context health, and the AEO readiness score on generated articles.

AEO Fundamentals5 min read

Visibility score (Dashboard gauge, 0 to 100)

The share of tracked prompts mentioned by at least one AI engine within the current filter (market, platform, date range): prompts mentioned ÷ prompts checked × 100. It follows the filter bar so you can look at one market or engine on its own.

Mentions, citations, share of voice

  • Mention: one per prompt × engine × scan where the brand appears in the answer.
  • Citation: one per answer that links to a URL on your domain. "Cited pages" counts distinct URLs.
  • Share of voice: your mentions ÷ (your mentions + competitor mentions in the same answers) × 100.
  • Change compares the average of the second half of the period with the first half, which is steadier on small samples than first-versus-last.

Recommendation level (per answer)

Being mentioned is not being recommended. Each answer gives your brand a recommendation level, shown with its reason in the evidence drawer:

LevelRule
Top pickThe answer names it as the single best or first choice, with no condition. Being first in a list is not enough
RecommendedThe answer actively recommends it (alongside other brands counts too), with no condition
ConditionalRecommended only for some buyers or under a condition, e.g. "if your budget allows", "for teams that need X"
Listed optionListed among options with a neutral description, not recommended
Mentioned onlyNamed in passing, used as a comparison, criticised or advised against

A model judges the level from the answer's own words, and every level comes with the quote it rests on. Languages that haven't passed calibration yet use rule-based judging for now; see Methodology.

If a level is wrong, open Earlier answers in the evidence drawer, click "Judged wrong?", choose the level it should be and submit. The correction is stored next to the answer; the answer itself is never changed.

From mention to recommendation

The Dashboard card From mention to recommendation counts answers as a funnel: AI answers → mention you → cite your site → recommend you → first pick, expandable per AI engine. It also shows:

  • Of answers citing you, share that recommend you: whether being cited turns into a recommendation.
  • Of answers mentioning you, share that recommend you.
  • Weighted recommendation share: each recommendation is weighted by the tracked competitors in the same answer: recommended alone counts 1, alongside 3 competitors counts 1/4; divided by all answers. The Competitors page uses the same method for every brand.
  • Sole recommendations: answers that recommend you and name no tracked competitor.

See From mention to recommendation.

Ranges and data source

Rates show a 95% range: measured again under the same conditions, the value would very likely fall inside it. Under 10 answers the card shows x/n instead. A narrow range means many answers; when it is wide, wait for more scans or turn on repeated sampling.

Every metric also states its data source: Real product UI (the same as what users see in the AI app) or Official API, and the models used. Coverage and candidates in the Prompt Library carries a 95% range as well; see Prompt coverage and Measurement confidence.

Coverage and position per prompt (Prompt Tracking)

  • Coverage: engines that mention you ÷ engines that answered × 100, counting only the engines the market scans. Engines that errored or hit a quota are excluded so outages don't drag the number down.
  • Position: the index of the first numbered or bulleted line containing your brand, shown as #n on the engine's tab in the evidence. Mentioned outside a list counts as "mentioned"; position 1 is "top result", 1 to 3 is "listed".
  • Repeated sampling: with "Ask 3× per scan" on, each engine answers three times and the result shows as mentioned x/3, so you can see how often the brand appears across answers.
  • Cited: the host of a link in the answer is your domain or a subdomain.
  • Competitors in the answer: the first column of a comparison table if there is one, otherwise the list items, filtered to real names and capped at eight.
  • Best rank, average rank, change are stored per scan and feed trends and achievements.

Sentiment

Every answer gets a sentiment score. Above +10 is positive, below −10 negative, in between neutral. The Dashboard "positive sentiment" figure is positive ÷ (positive + negative); neutral answers are left out of the denominator.

Context health

Each answer is judged against three failure modes: the brand described as the wrong kind of company, an off-topic answer, or an answer in the wrong language. The verdict is match, mismatch or unrelated; mismatch and unrelated show a context card in that prompt's Evidence, and Prompt Tracking can be filtered by "Context issue". The judgement uses the Business info in Content Rules, so keep it accurate. Wrong phone numbers, addresses and prices in answers to brand searches are checked separately under Automations → Brand fact check.

AEO readiness of generated articles (GEO score)

A rule-based score out of 100 (the same article always gets the same score) across five dimensions:

DimensionWhat it looks at
StructureWhether the heading hierarchy, an FAQ section and structured data are in place
Entity clarityWhether the brand and topic are clear and appear at the start
Data supportWhether there are concrete numbers and comparisons
LengthWhether the article is long enough to answer the question fully
Direct answerWhether the first paragraph answers the question directly

The article page shows how many dimensions pass (N / 5): 5 = fully optimized, 4 = good, 3 = ready to publish, 2 or fewer = needs work.

Frequently asked questions

Why does the Dashboard number differ from Prompt Tracking? The Dashboard follows the filter bar and counts a prompt as covered if any engine mentions you; Prompt Tracking shows each prompt's own coverage across engines.

All metrics fell on the same day? Check the Dashboard for Platform-wide changes. When an engine changes for every project at once (fewer cited links per answer, a new model), the cause is usually the platform.

Does the score predict traffic? No. It measures presence in AI answers. Visits from AI are on AI Traffic, and the leads and revenue they bring are on Revenue Attribution.

Was this article helpful?