Always identify the unit and eligible population

A rate needs a numerator, denominator, scope, period, and aggregation rule. An answer-level brand recommendation rate, a retailer win rate among decidable comparisons, and a share of purchase answers won by your shop are different quantities. Matching labels alone do not establish that AEO and Commerce Intelligence use the same underlying observations. Compare definitions and evidence before comparing percentages.

Core rates

The plain unweighted forms below explain the units. Some headline AEO rates apply configured question or model weights, so the displayed weighted percentage need not equal the raw supporting fraction. A missing assessment is excluded from an assessable denominator rather than being silently converted to absence.

MetricPositive outcomeEligible base
Mention rateApplicable identity is mentionedAnswers where that mention can be assessed
Recommendation rateApplicable identity is positively recommendedAnswers where that recommendation can be assessed
Retailer first-recommendation rateOwn shop holds the first supported retailer positionPurchase answers with an assessable retailer order
Retailer win rateOwn shop strictly outranks other recommended retailersDecidable wins, losses, and ties
Question coverageA question has at least one qualifying positive answerDistinct eligible observed questions, not every repeated answer
Shop-link rateA supported shop link accompanies the shop recommendationShop recommendations where link presence can be assessed

A mention is not a recommendation or a first position

'Shop A sells accessories' can mention Shop A without endorsing it. A list of product brands does not establish retailer ordering. For first position, use explicit retailer ranking or an otherwise supported sole-retailer case. The first textual occurrence of a name, the first citation, and the first product heading are not automatically first retailer recommendation.

Retailer wins, joint recommendations, and the purchase distribution

A strict win requires the own shop to be ahead of the other identified recommended retailers. A sole own-shop recommendation can be a win; competitor-only recommendations are losses. When several retailers are recommended without a supported order, joint recommendation is observable while a strict win may be undecidable. A mutually exclusive purchase distribution applies precedence: strict own win, joint recommendation, competitor only, no clear retailer recommendation, then unresolved evaluation. A joint-rate metric can overlap other metrics, but pie-chart segments must not overlap.

Why '0 unclear' and '65 unclear' can both appear

Brand-status uncertainty and retailer-order uncertainty are different dimensions. Every answer might have an assessable brand mention while many purchase answers lack a defensible retailer order. The first card can then show zero unknown brand statuses while the win-rate card excludes many undecidable retailer outcomes. A clear label should state which assessment is unclear and which cohort it counts. If both cards claim the same dimension and cohort yet disagree, investigate it as a defect.

A worked example

The following artificial sample illustrates why a large brand-presence number and a small purchase-win share are compatible. It does not describe a real merchant.

Example
Synthetic sample
100 assessable brand answers: 30 recommend the brand -> 30%
 60 purchase answers: 3 strict shop wins -> 5% of purchase answers
 20 decidable retailer comparisons: 3 wins -> 15% win rate

The 30%, 5%, and 15% answer different questions.
Do not label all three simply 'recommendation rate'.

Conditional rates can exceed a broad mention rate

A category's mention rate may use all applicable category answers while recommendation or purchase relevance uses a smaller eligible subpopulation. A 50% rate from one of two eligible observations is possible alongside a lower broader rate. It does not mean a recommendation occurred without evidence. Inspect the relevant base and show a small-sample notice. Never reconstruct a conditional denominator by dividing unrelated rounded percentages.

Rounding and comparisons

A change from 20% to 25% is five percentage points and a 25% relative increase. These are different units. Raw counts, weighted values, and percentages rounded to one decimal can differ slightly after display rounding. A competitor rate from another module is comparable only when the identity, period, question cohort, method, weights, and eligible assessment agree.

Commerce metrics: answers, recommendation units, and sharesAEO and purchase-visibility scores: formulas and publication rulesHow the KI recommendation funnel is counted