AI Visibility Measurement

Tooling and coverage

• The measurement stack covers every priority assistant and every priority market — coverage is not silently capped by a tool’s pricing tier.
• Bing Webmaster Tools AI Performance is wired up for every market property — the only first-party citation data available. (Set Up AI Visibility Measurement)
• Third-party tool estimates are reconciled against first-party data on a schedule, so each tool’s error bars are known.
• Coverage gaps (smaller markets, reasoning modes, marketplace assistants) are filled by in-house sampling that produces comparable metrics.

Prompt corpus and metrics

• The prompt corpus is versioned, journey-staged, split by market and audience segment, and has a named owner and review cadence. (Build and Maintain a Prompt Library)
• Written, tool-independent definitions exist for mention, citation, and share of voice — numbers survive a tool change.
• Executive metrics report on rolling 90-day windows with volatility bands — never point-in-time readings, since 40–60% of cited sources change month to month.
• The natural noise floor per surface is quantified, and gains or losses are judged against it.
• Both ChatGPT reasoning modes are sampled — they cite largely different domains.
• Visibility is tracked against the same fixed prompt set over time, not improvised queries.
• Each platform is measured independently — visibility on one surface is not treated as predictive of another.

Accuracy and sentiment monitoring

• Hallucination detection diffs assistant answers against canonical product data as ground truth — not against a tool’s generic heuristics.
• A severity scale with owners and routing exists for AI misinformation incidents, with price and safety errors at the top.
• A truth-set of verifiable facts per flagship product per market exists and runs against each assistant on schedule.
  • evidence: a per-surface accuracy score reported on a fixed cadence
• The sentiment and complaint themes assistants attach to each product line are tracked per market and reconciled against real support and review data.

Competitive benchmarking

• Distinct named-competitor panels exist per product archetype, benchmarked separately per market — one blended panel misrepresents everyone.
• Prompts where competitors are cited and you are absent flow into the editorial backlog through a standing process.
• A documented surface × market priority matrix says which assistant/market combinations matter most, so effort concentrates.

Attribution and referral

• AI referral segmentation is implemented in analytics for every market property and maintained as user agents and referrer patterns evolve. (Set Up AI Visibility Measurement)
• AI-referred cohorts are tracked through conversion, retention, and repeat behavior — not just session counts.
• Zero-click influence is estimated with survey or panel data rather than assumed to be zero.
• When measurement drops, crawl, rendering, and trust failures are ruled out before content changes are made. (Diagnose an AI Visibility Drop)