How we measure the answer layer.

Research is produced from aggregated, de-identified answer and citation records collected by AthenaHQ across major AI answer engines. Studies can include ChatGPT, Perplexity, Google AI Overviews, Google AI Mode, Gemini, Claude, Copilot, Grok, and other supported systems.

Every published result identifies its observation window and eligible cohort. We do not present a single prompt run as a stable model-level finding.

Unless a study states otherwise, flagged responses and deleted records are excluded. Longitudinal comparisons use like-for-like cohorts so changes in AthenaHQ’s customer mix do not masquerade as changes in model behavior.

Source counting

A source’s citation share is the percentage of eligible answers citing its normalized root domain at least once. Repeated links to the same domain in one answer count once.

Entity mentions

Mention studies count resolved entity references in eligible answers. Mention frequency is not equivalent to sentiment, recommendation, or commercial impact.

Some studies group source domains into maintained media categories. For example, YouTube and TikTok are classified as video; Reddit, Quora, G2, and Trustpilot as community and reviews. Unmatched domains can be categorized as text and articles.

Because one answer may cite multiple source formats, format prevalence percentages are not mutually exclusive and should not be added into a single whole.

We distinguish observations from interpretations. Citation frequency describes retrieval behavior in the measured cohort; it does not, by itself, prove that a source caused a recommendation or conversion.

AthenaHQ’s corpus reflects the industries, prompts, markets, and models observed by its platform. Findings should not be treated as a census of all global AI use.

Material corrections are noted on the affected article with a description and date. Reproducible data extracts are retained internally, and publication numbers are checked against final research assets before release.

Questions about methodology or corrections can be sent to research@athenahq.ai.