AI measurement templates Reviewed 28 October 2026 Guide: https://webaloha.co/how-to-measure-geo-ai-visibility/ prompt-observation-log.csv One row per exact prompt, platform and run. The example questions are marked not run. All observation fields are blank. Replace the examples with buyer questions appropriate to your business, then freeze the panel before testing. Record the exact answer surface, API or tool, search/grounding mode, model if disclosed, account state, date, location if known, raw answer and cited URLs. Use complete, partial, unavailable, blocked, or not run for the outcome. Unknown is different from no. Do not infer an absent mention from a blank answer. Keep branded prompts apart from unbranded discovery. A cached repeat does not create a new independent answer. Preserve failed runs and protocol changes. monthly-ai-measurement.csv Leave missing values blank and explain them in Data limits / changes. Enter a verified zero only after an actual observation or complete report establishes it. Use the same period, timezone, filters and population for every compared rate. For a rate, retain both the numerator and denominator; divide only when the denominator is known and positive. Do not sum monthly percentages. Prompt citation rate: complete eligible answers with an owned citation divided by complete eligible answers. Exclude failed/unavailable answers and cached repeats; disclose coverage and keep each platform/surface separately reportable. AI session lead rate: sessions with the chosen lead key event divided by the matching AI sessions. Event count divided by sessions measures something else. Count qualified enquiries only from a verified CRM or enquiry ledger. An analytics key event is not proof that a person is a qualified lead. The additional-source row must exclude sessions already in native AI Assistant if you intend to combine the two. Otherwise report the views separately. Google AI Overviews and AI Mode remain Organic Search in GA4. Do not report ordinary Google organic sessions as isolated AI-feature traffic. source-classification-fixtures.json and verify-classification.mjs Constructed positive and negative string examples, not a GA4 export or real traffic results. Download both into one folder and run: node verify-classification.mjs The script only reads its neighboring fixture file. It makes no network calls. It tests equivalent JavaScript expressions using basic RE2-compatible syntax; it does not simulate Google's default channel rules or GA4 campaign precedence. The patterns cover selected known hosts, not every assistant or Google's full recognition list. Review your recorded values and case handling before use. An empty referrer or Direct source remains unknown. A source tag can be copied, so matching it does not independently prove a recommendation or a human visit. Implementation checks to perform on your own test property Confirm consent defaults and updates, permitted source persistence through redirects, a successful-submission key event, and reconciliation with your own accepted enquiry record. These templates do not establish that those checks have passed. Keep reporting identity and data-quality notices with each export. Primary documentation checked 28 October 2026 https://support.google.com/analytics/answer/9164320?hl=en https://support.google.com/analytics/answer/9756891?hl=en https://support.google.com/analytics/answer/13051316?hl=en https://support.google.com/analytics/answer/12923437?hl=en https://support.google.com/analytics/answer/11242841?hl=en https://support.google.com/analytics/table/13594742 https://support.google.com/analytics/answer/10000067?hl=en https://support.google.com/analytics/answer/11161109?hl=en