GeoScore Methodology

Geo Score runs three buyer-style prompts across GPT-4o, Claude and Perplexity, producing up to nine web-grounded responses. The three prompts ask for a direct recommendation, a shortlist and a comparison, each built from the exact search phrase you enter, so the check reflects genuinely different ways a buyer might ask.

The check runs across three live AI models:

• ChatGPT (GPT-4o)
• Claude (claude-sonnet)
• Perplexity (perplexity-sonar)

That is up to nine web-grounded responses per check (3 prompts × 3 models). Each response records whether your brand appeared, where in the response it appeared, which competitors appeared, and the sentiment of the mention.

GeoScore is a number from 0 to 100.

GeoScore combines how often your business appears across the checked prompts, how prominently it appears relative to competitors and the context in which it is mentioned.

Higher model coverage, stronger placement and positive context all improve your score. Missed mentions and competitor displacement lower it.

The score is an index of observed AI visibility based on the specific checks GeoScore ran, not a probability or a guarantee of future AI behaviour.

AI Readiness is a second, independent number from 0 to 100. Geo Score measures how often current AI responses actually recommend your business. Readiness measures something different: how strong, clear and independently supported the evidence behind your business is for earning that recommendation. The two scores are never combined. A low Geo Score alongside a high Readiness Score is a genuine opportunity (the groundwork is there, it isn’t converting into recommendations yet), not a contradiction.

Readiness runs a separate, deliberately trimmed evidence check, not the deeper multi-page audit used in the paid report. It fetches a fixed, bounded set of resources: your homepage, an About page (when one can be found), up to two further pages selected for relevance to your search phrase, robots.txt, and sitemap.xml. It never follows links found inside a fetched page.

From that evidence, Readiness checks:

• Entity clarity: does the homepage clearly state what the business is and does
• Structured data: is there schema.org / JSON-LD markup an AI model can parse directly
• An About page: findable, machine-readable context about the business
• Content depth for the search: pages that speak directly to the buyer search checked
• Independent citations: third-party sources, reusing the citation domains the Geo Score model calls already returned wherever possible, rather than paying for a second search

At most one additional targeted web search runs, only when Geo Score’s own citations and the site’s own authority signals are both thin. At most one cheap model call classifies the evidence into plain categories (clear/partial/unclear, strong/moderate/thin, and so on). It never assigns the number itself. A fixed set of weights in code turns those evidence signals into the 0–100 score, the same way Geo Score’s formula is fixed code, not a model’s opinion.

Readiness results are cached for 7 days per website and search phrase, since the evidence describes the site rather than any one run. If the evidence check can’t complete (for example the homepage doesn’t respond in time), Readiness is marked unavailable for that check. Your Geo Score is never affected.

GeoScore analyses the AI responses to find business names that appeared instead of yours. It uses an LLM to extract named entities from each response. If the LLM extraction is unavailable, a heuristic fallback extracts sequences of two or more consecutive capitalised words (e.g. "Acme Software", "Widget Corp").

Competitors are ranked by how often they appeared across your checked responses. The most frequently appearing names are shown first.

The recommended next action is generated by rule-based logic, not an LLM. Rules examine your GeoScore, which models mentioned you, which models did not, the topic, and which competitors appeared most often, then select the highest-priority action from a fixed set of recommendations.

GeoScore does not claim to know the internal reasoning behind why a specific AI model chose a particular answer. AI models do not expose their ranking or source-selection logic; GeoScore observes their outputs.

Each check stores one row per prompt per model. Each row records:

• Date, brand name, prompt text, prompt category
• Which AI model answered
• Whether your brand was mentioned (true/false)
• Position of the mention (1–5)
• Competitors that appeared in the response
• Sentiment of the mention
• A short snippet of the response (not the full text)

The check record also saves your email address, website, company name and topic. The saved record does not include the full AI response text. If you request a deeper report, a separate record of that request is also saved.

A GeoScore is a snapshot of what GeoScore observed in one run. It cannot tell you:

• Why a specific AI model chose the sources it used: AI models do not disclose their internal reasoning or source weighting
• Whether the result will be the same next week: AI model outputs change as models update and retrain
• Your organic search ranking: GeoScore measures AI recommendation visibility, not traditional SEO position