Competitive Mention Map
Consensus definition
A Competitive Mention Map is constructed by running a representative prompt set — from a few dozen queries reflecting real buyer research intent upward — across multiple AI platforms (ChatGPT, Perplexity, AI Overviews, Gemini) and recording every brand that co-appears with the tracked brand in each response1. The raw output is a co-occurrence matrix: rows represent the tracked brand's queries, columns represent competitors, cells record joint mention frequency. Clustering by query type — category, comparison, use-case and evaluation queries2 — then produces segmented views, and comparison prompts may surface different competitor peer groups than category prompts. Visual forms include weighted network graphs (nodes = brands, edge weight = co-mention frequency), platform-breakdown heat maps and segment-by-segment frequency tables. The analytical lineage connects to classical brand-association mapping — Aaker's brand equity framework treats associations as network-like mental connections, and Keller's CBBE model frames them as competitive points-of-parity and points-of-difference. The Competitive Mention Map operationalises those concepts for AI-mediated discovery environments, where the platform — not the consumer — constructs the initial association set3.
rhinegold operator refinement
Rhinegold's reframe: when an AI model answers "Which providers should I evaluate for [category]?", it synthesises a shortlist that functions as the buyer's consideration set — the subset of brands actively evaluated before a purchase decision. Consumer research puts the typical consideration set at two to five options4. The Competitive Mention Map makes this AI-constructed shortlist explicit: which peers appear with you, how frequently, on which platform. Appearing alongside aspirational enterprise brands on one platform while being grouped with low-cost alternatives on another is a split-positioning signal that demands strategic intervention, not more content.
Operational use
Run a defined prompt set — practitioner guides start at 20–30 buyer research queries across category, comparison, use-case and evaluation intents, on a monthly cadence, across four AI engines2 — and keep it stable between waves. Build the co-occurrence matrix from raw response logs; aggregate by query cluster and platform. Track over several months which competitors co-appear most frequently — shifts in AI share of voice typically take 60 to 90 days to show5 — and treat them as leading indicators of repositioning in AI-mediated channels.
Measurement boundary
A Competitive Mention Map is only as valid as its prompt set — queries that skew toward product categories, geographies or maturity stages the brand does not serve will populate the map with spurious pairings. Results vary meaningfully by platform: citation behaviour differs substantially between providers (Perplexity links to sources in more than three quarters of its responses5), producing different co-mention patterns for the same queries. Competitor naming (parent vs. product vs. trade name) requires explicit normalisation. Maps are point-in-time snapshots — LLM training cycles and retrieval shifts can change the competitive frame without any action by the tracked brand.
Distinct from
From Co-Mentions: that is the underlying signal — the raw observation that brand X and brand Y appear in the same AI response. The Competitive Mention Map is the analytical product built from aggregated co-mention data across a prompt set, not a single observation. From Share of Voice: SoV is a scalar per brand (brand mentions ÷ all brand mentions); the map is a relational matrix showing pairwise co-occurrence across the full competitive set. From Platform Divergence: that is a descriptor of variance across AI engines — one dimension visible within the map (the platform axis), not a separate construct. From Brand Recommendation Share: that tracks explicit recommendations rather than any mention or co-appearance; recommendation share is a subset of the signals that feed the map.
Common mistakes
- Treating the map as static: AI-platform co-mention patterns shift with LLM updates and content changes, making single-run snapshots misleading without longitudinal comparison.
- Using branded prompts only: querying "How does [Brand X] compare to [Brand Y]?" by name will artificially inflate co-mention frequency; the map should be built predominantly from category-level and use-case prompts reflecting real buyer queries1.
- Collapsing platforms into a single aggregate: different engines construct different competitive peer groups for the same queries due to differing retrieval and citation architectures5; platform-level disaggregation is essential for actionable diagnosis.
- Conflating co-mention frequency with endorsement: high co-mention frequency with a budget competitor may signal commoditisation framing, not validation — context and sentiment in the surrounding response must be read alongside frequency counts1.
Sources & deeper reading
- 1Sight AI — "Brand Monitoring in LLM Outputs" (competitive co-mention tracking methodology; consideration-set framing; sentiment in context)
- 2Shadow Inc. — "How to Measure AI Share of Voice" (20–30 prompts; category, comparison, use-case and evaluation queries; monthly cadence; four engines)
- 5Nightwatch — "AI Share of Voice: How to Track and Grow" (Perplexity links sources in 77 %+ of responses; improvements show in 60–90 days)
Update history
- 2026-10-09 — Aligned prompt-set sizes, cadence, platform counts and citation figures with what the cited guides actually say (Shadow: 20–30 prompts, monthly, four engines; Nightwatch: Perplexity links in 77 %+ of responses, 60–90 days to show change). The two-to-five-option consideration set is now cited to its source. Removed a dead source.