---
type: "synthesis"
arc: "measurement"
articles: ["a001", "a005", "a006", "a010", "a011", "a012", "a014", "a025", "a029"]
tags: ["measurement", "observability", "kpis", "auditing"]
id: "cross-day-measurement-observability"
sources: ["geo"]
sourceVaultSlug: "hbr-seg-geo"
originDay: 3
articleStem: "hbr-seg-geo"
sourceUrl: "(unified vault: 13 sources)"
sourceTitle: "HBR — Demand Ⅰ-A · GEO / AI-mediated discovery & agentic commerce"
---
Once the funnel goes dark ([[cross-day-funnel-collapse-dark-funnel]]) and the buyer becomes an algorithm ([[cross-day-new-customer-reframe]]), traditional web analytics measure a human who no longer visits ([[question-web-analytics-replacement]]). Nearly every article proposes a replacement, and together they form a maturing measurement stack.

- **Prompt-based auditing**: [[action-conduct-prompt-audit]] (Ignatius), [[action-measure-som]] via the [[framework-three-prong-ai-perception]] and [[concept-mention-rate]] (Dubois), and B2B's [[concept-generative-listening-systems]] / [[action-conduct-generative-audit]] (thousands of prompts across LLMs).
- **Continuous simulation**: [[concept-continuous-ai-simulation-infrastructure]] / [[action-build-simulation-environment]] (Sabbah & Acar) and PwC's [[concept-synthetic-customers]].
- **Observability of the live ecosystem**: [[concept-agentic-observability]] / [[action-monitor-agent-ecosystems]] (PwC), and WTP experiments [[action-conduct-wtp-experiments]] (luxury).
- **The recursive trick**: [[concept-recursive-ai-probing]] / [[contrarian-use-ai-to-probe-ai]] — use the black box to optimize for the black box.

The honest open problems: measuring mentions inside private, ephemeral chats ([[question-measuring-ai-mentions]]) and version volatility ([[question-som-volatility]]). Shared warning: **LLM visibility ≠ business success** — measure clicks, conversions, branded-search lift too. The new metric vocabulary itself is contested — see [[cross-day-geo-acronym-babel]] on Share of Model vs AI recall share.