AI Visibility Methodology

GEO Radar helps brands understand how often they appear in mainstream AI recommendations, and how that ranking trends over time, so they can improve GEO strategy.

AI visibility measures how often and how prominently AI engines recommend specific products and brands. GEO (Generative Engine Optimization) is the methodology for improving and tracking that visibility, just as SEO tracks search result rankings.

Each collection period, we query AI engines — ChatGPT, Gemini, Grok, Perplexity, Claude, and DeepSeek — using 8 category-specific prompts per engine. All data is collected via official APIs, not web interfaces. Scoring engines are equal-weight. An engine that misses the period validity threshold is excluded from that category's overall score and shown as No data.

Responses are processed by a separate LLM to extract brand mentions and their order of appearance. Extracted brands are matched against a canonical brand database with alias support.

Each brand receives an AI visibility score from 0 to 100:

• Appearance Rate — percentage of valid responses that mention the brand
• Avg Rank Score — exponential decay on average position: 100 × e^(−0.15 × (avgRank − 1))
• Model Coverage — fraction of that category's scoring engines in the period that mention the brand. Historical periods keep the denominator they were published with.

Overall boards only include brands that clear a minimum evidence gate: appearance rate ≥ 10%, or mentions from at least 2 scoring engines in the period. Single-response #1 ghosts stay out of Top 20 and can still appear under Also mentioned.

Individual engine rankings use a simplified formula without model coverage:

Period-over-period rank change (Δ) uses four labels: up, down, new, and flat. If there is no prior published period, the change cell shows —.

Also mentioned lists brands that appear in AI answers for the period but do not make that board's Top 20.

On category pages, the competition chart plots Top 20 brands by appearance rate (X) and average rank (Y, #1 at the top), split into Leaders, Challengers, Niche, and Laggards at the period medians.

Company pages aggregate products by ownership. There is no company-level score or company-level rank; product metrics match Brand and Category Overall boards.

Rankings are not based on a fixed list. Each collection period, brands are dynamically discovered from AI responses. New brands are automatically added and scored. A human review process ensures brand names are properly normalized and deduplicated.

Rankings are updated on each category's collection period. Data is labeled by the period start date (e.g., "2026-07-27").

Some earlier periods may be filled in later using the same prompts, engines, and scoring method as live collection. The front end treats them as ordinary historical periods for trends and Δ; it does not separately label which periods were backfilled.