Data Methodology

Why we publish methodology

The core risk in GEO is 'high-impact but hard-to-verify' data. Our rule: the more specific the number, the more traceable it must be. This page documents every key metric's definition, sampling, personalization handling and error range for third-party review.

Core metric definitions

• Brand Mention: whether the brand or product name appears in an AI answer.
• Brand Citation: whether the AI provides a link or explicit pointer to the brand's site/product page. Citations are stronger than mentions.
• Qualified Citation Rate: share of the target prompt set that yields a qualified citation.
• Citation Share: the brand's share of citations among prompts where competitors also appear.

Prompt design principles

• mirror real buying intent (best supplier / manufacturer / alternative / compare)
• cover industry × country × buyer-role combinations
• neutral wording, no brand steering
• store full prompt text for review

Platform sampling

For each platform, run the prompt set in target country/language; record: mentioned?, cited?, cited URL, cite snippet. Platforms differ in retrieval/generation, so results are reported separately, not compared directly.

Personalization & chance

• run each prompt multiple times; separate chance from stable appearance
• record login/geo variation; default = anonymous + target-country IP
• flag low-confidence prompts

Share & attribution

Among competitor-co-occurring prompts, count citations per brand and compute share; attribution uses the cited domain and topic, with 'uncertain' flagged at boundaries.

Cadence & error

Benchmarks update every [TODO: cadence, e.g. quarterly]. Single samples carry platform volatility; conclusions rely on trends and stable multi-run appearance, not one-off results.

Sources & evidence-card format