Brand Recommendation Share
- A mention is not a recommendation — being one of ten options contributes only one fifth as much as being one of two.
- Read the gap between Mention Rate and Brand Recommendation Share, not the absolute number.
- Use it as a steering KPI; pair it with Citation Rate for authority.
Consensus definition
There is no industry-standard definition for a competition-weighted recommendation metric in LLM answers. In current practice, observers fall back to mention share or raw Share of Voice, neither of which distinguishes a brand named alone from a brand named in a long aggregate list.
rhinegold operator refinement
Brand Recommendation Share treats each valid recommendation answer as the unit of analysis. If the brand is named, it receives a contribution of 1 divided by the number of distinct providers listed. If the brand is not named, its contribution is zero. The metric is the average of these contributions across all valid recommendation answers — which makes it decomposable as Mention Rate × conditional concentration.
2 Illustrative of the weighting mechanism; not measured client data.Source: rhinegold weighted-recommendation methodology v1.0.
- Numerator
- For each valid recommendation answer: 1 divided by the number of distinct providers listed if the brand is named in that answer; 0 otherwise.
- Denominator
- All valid recommendation answers (N).
- Unit of analysis
- A valid recommendation answer (an LLM answer to a discovery prompt that contains a provider list).
- Inclusion
- Answers to discovery prompts that produce a provider list; provider extraction successful.
- Exclusion
- Non-recommendation answers (brand prompts, informational prompts). Answers without a provider list. Answers where provider extraction failed.
- Missing data
- Brand not named → contribution 0 (the answer still counts in the denominator). Provider extraction failed → answer is excluded from both numerator and denominator.
- Aggregation
- Within an answer: 1/|providers| weighting if named, 0 otherwise. Across answers: equal-weighted arithmetic mean over the valid recommendation set.
- Worked example
- Across 100 valid recommendation answers, the brand is named in 40. Average list size when named: 2.5. Mention Rate = 0.40. Mean conditional contribution = 0.40. BRS = 0.40 × 0.40 = 0.16. Brand named in only 8-provider lists: BRS = 0.40 × 0.125 = 0.05.
- Known sensitivity
- Sensitive to competitor-set completeness (incomplete set deflates the denominator of the conditional concentration). Sensitive to list-length distribution within the prompt set. Volatile at small N. Span-aware brand detection matters: substring matches inflate the numerator.
- Must not be read as
- A position-aware ranking metric (it is not). A causal claim about commercial outcome (it is not). Equivalent to Share of Voice (it is the weighted refinement).
- Comparability
- v1.0 baseline. Any change to the formula requires a new metric_version.
Operational use
Brand Recommendation Share is the steering KPI for visibility programs1 where the right question is not whether the brand is mentioned at all, but how exclusively. It separates weak visibility — one of ten in an aggregate list — from stronger visibility among a small number of recommended providers.
Measurement boundary
The metric is volatile at small n. It says nothing about the brand's position within a list — a rank-weighted variant is needed for that2. It is not a replacement for Citation Rate: a recommendation is not a source citation.
Distinct from
Against Share of Voice — the naïve count share — this is the competition-weighted refinement. Against Mention Rate, this is no longer binary. Against Citation Rate, this measures recommendation, not source attribution.
Observed pattern in practice
Common mistakes
- Confusing it with raw Mention Rate (overstates strength when lists are long).
- Averaging over non-recommendation answers (distorts the denominator).
- Working with an incomplete competitor set, which deflates the conditional concentration and inflates the metric.
Where consensus is missing
No industry-standard definition for a competition-weighted recommendation metric exists. This is a deliberately chosen, more substantive aggregation than naïve Share of Voice. Treat related research on position bias2 as methodological context, not as primary evidence for this formula.
Sources & deeper reading
- rhinegold Insights, Episode 04 — How to Measure What You Can't See (GEO metrics for B2B)
- rhinegold cross-industry B2B recommendation benchmarks — shared with engaged clients and partners.