← 返回资讯
2026-11-16约 30 分钟阅读责任主体:芯芸达

Reconcile an AI Mention-Rate Change Into Rate, Mix and Interaction Contributions

Reconcile a pooled AI mention-rate change through an exact three-term expansion, ordered substitutions and a symmetric bridge using fictional counts.

A client report moves from 42% to 31.5% overall AI mentions. One engine's rate improves by twenty percentage points and another improves by five. The reporting team has already identified a change in the engine mixture. The next task is to reconcile the exact 10.5-point decline and explain how a chosen decomposition allocates it.

That task requires more than identifying an aggregation reversal. It requires a reproducible bridge between two reported totals, with an explicit convention for the contribution of changing rates and changing weights. Different valid conventions can allocate the same total change differently.

The worked example below develops that bridge from two sets of counts. It is a descriptive accounting exercise. Every observation and engine label is invented, and the calculation makes no claim about actual customer demand or a GEO intervention's causal effect.

Start with compatible counts

Define one positive outcome for each eligible complete answer. In this exercise the label is whether the specified brand appears at least once. Each answer contributes either zero or one positive label, regardless of repeated mentions within its text.

Eligibility, exact questions, collection surfaces and classification rules remain compatible within each engine across the two periods. All planned observations complete. The example changes the allocation between two engine groups while retaining 200 eligible answers in each period.

Actual reports need their own metric contract. Ahrefs documents a response-level mention definition, while citations and its other measures have their respective definitions. The arithmetic here applies to the explicitly defined binary rate in the fictional table. Ahrefs metric definitions

GroupPeriod one mentions / answersPeriod one weight and ratePeriod two mentions / answersPeriod two weight and rate
Engine A80 / 160Weight 0.80, rate 0.5042 / 60Weight 0.30, rate 0.70
Engine B4 / 40Weight 0.20, rate 0.1021 / 140Weight 0.70, rate 0.15
Pooled84 / 200Rate 0.4263 / 200Rate 0.315

Here a group's weight is its share of that period's eligible answers. Multiplying each weight by its group's rate and adding the products reproduces the pooled rate. The table contains no missing-data adjustment or demand weighting.

Choose the accounting convention before interpreting contributions

For each engine, retain the old rate, new rate, old weight and new weight. Define a rate change and a weight change as the new value minus the old value.

An exact expansion of the pooled change has three terms. The first sums each old weight multiplied by its rate change. The second sums each old rate multiplied by its weight change. The third sums the product of the two changes within each group.

The third term is the interaction. It arises from expanding the product of a changed weight and a changed rate. Keeping it explicit allows all three contributions to add to the observed total without hiding a remainder.

This identity can be checked for one group by expanding the new weight times the new rate, after writing each new value as its old value plus its change. Adding the group identities gives the overall result.

The calculation is useful when a client wants to audit a report bridge. Group-composition problems are discussed in Berkeley's statistics text. The particular count table and accounting calculations here are editorial examples developed for this monitoring task. Berkeley statistics text

Calculate the old-weight rate contribution

Engine A's rate rises from 0.50 to 0.70, a change of 0.20. Multiplying by its old weight of 0.80 gives 0.16. Engine B's rate rises from 0.10 to 0.15, a change of 0.05. Multiplying by its old weight of 0.20 gives 0.01.

The sum is 0.17, or positive seventeen percentage points. It describes the rate change evaluated using the first period's weights.

A corresponding intermediate total keeps the old weights and substitutes the new rates. Its value is 0.80 times 0.70 plus 0.20 times 0.15, which equals 0.59. Moving from the old pooled total of 0.42 to that intermediate total contributes 0.17.

That intermediate value is a defined counterfactual summary of the observed group rates. The word counterfactual here identifies the substituted weights. A causal interpretation would require additional assumptions and an appropriate research design.

Calculate the old-rate mix contribution

Engine A's weight changes by negative 0.50. Multiplying that change by its old rate of 0.50 gives negative 0.25. Engine B's weight changes by positive 0.50. Multiplying by its old rate of 0.10 gives positive 0.05.

The mix contribution under the old rates is therefore negative 0.20, or negative twenty percentage points.

Another intermediate total keeps the old rates and substitutes the new weights. It is 0.30 times 0.50 plus 0.70 times 0.10, which equals 0.22. That is 0.20 below the original total of 0.42.

The old-rate and old-weight contributions use a common reference period. Their sum alone still leaves the interaction term to reconcile.

Keep the interaction visible

For Engine A, multiply its weight change of negative 0.50 by its rate change of positive 0.20. The product is negative 0.10. For Engine B, positive 0.50 times positive 0.05 gives positive 0.025.

Adding those products gives negative 0.075, or negative 7.5 percentage points. The complete bridge is positive 17 points, negative 20 points and negative 7.5 points. Together they equal negative 10.5 points, matching the movement from 42% to 31.5%.

The interaction is substantial in this invented example because the larger rate improvement belongs to the engine whose weight shrinks. Reporting it explains why adding two separately calculated old-reference contributions would otherwise leave a gap.

Avoid assigning the interaction to a content team, vendor or engine policy without evidence. It is a term in this specified arithmetic expansion.

Show why the order of substitution changes the allocation

A two-step bridge can first replace the rates and then replace the weights. The first step moves from 42% to 59%, contributing positive 17 points. Replacing the weights at the new rates then moves from 59% to 31.5%, contributing negative 27.5 points.

Reversing the substitution order produces another exact bridge. Replacing weights at the old rates first moves from 42% to 22%, contributing negative 20 points. Replacing rates at the new weights then moves from 22% to 31.5%, contributing positive 9.5 points.

Both paths reach the same final total. Their different allocations follow from the rates or weights held fixed at each step. A reporting contract should identify the chosen order if it presents a two-contribution bridge.

This is especially important when a chart labels a bar as performance. A reader needs its actual mathematical definition to understand whether it uses old weights, new weights or another convention.

Calculate a symmetric two-contribution view

One descriptive convention averages the two substitution orders. Equivalently, it evaluates the rate changes using the average of the old and new weights, and the weight changes using the average of the old and new rates.

Engine A's average weight is 0.55 and Engine B's is 0.45. The rate contribution becomes 0.55 times 0.20 plus 0.45 times 0.05, which equals 0.1325, or positive 13.25 percentage points.

The average rates are 0.60 for A and 0.125 for B. The mix contribution becomes negative 0.50 times 0.60 plus positive 0.50 times 0.125. The result is negative 0.2375, or negative 23.75 percentage points.

These two contributions sum to negative 10.5 points. In this convention, each receives half of the interaction from the old-reference three-term expansion. The convention supplies a reproducible allocation. It does not establish a uniquely correct business explanation for the change.

Decide what the report bridge should retain

Keep the original integer counts, group definitions, time windows and outcome rule. Store full-precision rates for calculation, then round the displayed contributions. A small rounding remainder should be identifiable from the retained inputs.

Give the bridge a method name and version in the report specification. Explain whether it uses a three-term old-reference expansion, an ordered substitution or the symmetric convention above. Compare periods using that stated method.

If a required group has no eligible answers in one period, its rate is unresolved. Apply the agreed missing-data policy before claiming a complete decomposition. A group introduced later may also change the comparison population and require a separate expansion analysis.

The existing market-mix guide explains the earlier decision of detecting an aggregation reversal and selecting a fixed-weight comparison. This article addresses the subsequent accounting task of allocating the exact total movement under declared conventions. Market-mix guide

Use the arithmetic in a procurement review

Give a reporting supplier the fictional table and request the pooled totals, three-term bridge and symmetric view. Ask for the underlying group table and formulas in the format the client will actually receive.

Then trace one authorized project record to its group membership and outcome label. Confirm that consumer surfaces, API samples, languages and account contexts are grouped according to the intended design. A field called engine can conceal different collection environments unless its meaning is explicit.

Semvi by Xindar is Xindar's monitoring offering. Define the required reporting fields, grouping and export demonstration in the engagement. Evaluate a later recommendation through the actual entity and supported reason in the complete answer. Semvi introduction, Product scope and buyer verification

Questions about a report bridge

Why are three terms needed in the old-reference expansion?

Both the weight and the rate change. Expanding their product produces a term for each individual change and their interaction. All three add to the exact pooled movement.

Why do two substitution orders allocate different amounts?

Each order evaluates one change using a different set of fixed values. Retaining the intermediate totals makes the allocation reproducible.

Does the symmetric convention remove all judgment?

It defines one transparent allocation by averaging the two orders. Choosing that convention and defining the groups remain reporting decisions.

Can the contribution labeled rate measure a GEO campaign's effect?

A rate contribution is a descriptive calculation under its stated weights. A causal campaign claim needs a suitable design and evidence about concurrent changes.

Which inputs should the client keep?

Retain the counts, group definitions, full-precision values, eligibility rules, reporting windows and method version, together with the underlying observation evidence needed for review.

Sources and method

The linked metric definition and statistical background were reviewed on November 14, 2026. All counts and bridge calculations are fictional. Their arithmetic follows from the displayed inputs and defined substitutions. The article reports no vendor-account test, population visibility estimate or measured recommendation gain.