A paired predictive-plus-measured AI visibility corpus. (1) Measured side: GA4-connected client sites where sessions, purchases and revenue are attributed back to the specific AI model that referred them (ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot, DeepSeek), reported as monthly AI share of traffic, revenue and purchase rate, and ranked per page for the top 50 pages. (2) Predictive side: a large recurring prompt-survey corpus — thousands of category prompts run across every major AI model, scored for how each answer describes, cites and recommends each tracked brand across every stage of the buying cycle. The pairing is the point: the same brands have both what-the-models-would-say labels and what-they-actually-drove outcomes.
Equities in US. Related tickers SHOP, AMZN, MELI, ETSY, WMT, NKE, BBY, DG.
Tabular (GA4-derived aggregate session, purchase and revenue tables keyed by AI model and by URL; monthly time series) + text (stored AI model answers per prompt with brand mention/citation/recommendation scoring). Explicitly aggregate-level: the product states it uses read-only access to aggregate metrics and no user-level data.
Per reference account the dashboard shows 10,840 AI sessions, 7.1% AI share and $76,480 AI revenue with 50 tracked pages, implying roughly 150K-160K total sessions per account per period; the prompt side runs 'thousands of category prompts across every major AI model' per client, so with a seven-figure-plus aggregate session base the prompt corpus is the larger artifact at potentially hundreds of thousands of scored prompt-answer pairs
Free with the mechanic: AI-model label on each session, purchase and dollar (filter-by-model is a core interaction); per-page AI-referred session, purchase and revenue labels for the top 50 URLs; month-over-month AI-share labels making model-launch effects time-indexed (the page notes 'when AI share starts climbing... you see exactly which model drove it'); from the prompt side — brand mentioned yes/no, cited yes/no, recommended yes/no, per model, per prompt, per buying-cycle stage; and page-type labels distinguishing product pages, collections and editorial journal articles, which reveals whether AI drives commerce or research traffic.
Centium — A paired predictive-plus-measured AI visibility corpus. (1) Measured side: GA4-connected client sites where sessions, purchases and revenue are attributed back to the specific AI model that referred them (ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot, DeepSeek), reported as monthly AI share of traffic, revenue and purchase rate, and ranked per page for the top 50 pages. (2) Predictive side: a large recurring prompt-survey corpus — thousands of category prompts run across every major AI model, scored for how each answer describes, cites and recommends each tracked brand across e…
Not yet on file: Sample · Data dictionary · Data frequency · Reporting lag · Point-in-time · License