AI Visibility Audit
Consensus definition
An AI Visibility Audit is a structured, time-bounded methodology for measuring a brand's presence, accuracy and competitive standing in AI-generated answer outputs across a defined set of providers and prompts1. Its scope spans five components: a prompt set of typically 20–30 queries per core topic (brand-direct, category-level and buyer-scenario formulations)2; provider coverage across the major answer surfaces (ChatGPT, Gemini, AI Overviews and AI Mode, Perplexity, Copilot)3; four scored dimensions per response — mention rate, citation rate, sentiment, and competitive share of voice against a 3–5 competitor benchmark — incomplete sets inflate the brand's share4; a source-authority mapping of which domains the systems pull as citation inputs; and a gap analysis translating findings into a prioritised fix list covering content structure, entity clarity, schema markup and off-site signals5. Because AI answer content is non-deterministic, each prompt is run several times per platform; repeats stabilise the aggregate, while single segments keep swinging, so the larger lever is the choice of prompts, not the number of repeats2. The output is a deliverable — typically a scored report plus implementation specifications — not a subscription dashboard6. Brand-direct and category prompts are reported apart: in a 2026 test of 175 brands on eight platforms, 96 % were described accurately when asked about directly, yet 89 % never appeared in category answers4 — the audit's job is to show which of the two the brand has.
rhinegold operator refinement
Rhinegold's reframe: monitoring tools answer what changed — useful only once a credible baseline exists and weekly optimisation cycles are running. An audit answers the prior question — where the brand stands, what is being said, and what to fix first. For operators without dedicated AI-visibility staff, a quarterly audit often delivers more cost-per-insight than a monitoring subscription accessed four times a year. The audit is also the correct instrument before a rebrand, product launch or vertical entry6.
Operational use
Commission a baseline audit with a 30-prompt set (10 brand-direct, 10 category, 10 buyer-scenario), run across at least four major AI surfaces, three repetitions per prompt2. Score each response on citation presence, mention position, sentiment and share of voice. Translate the gap analysis into a prioritised content + technical action list. Re-audit quarterly or after major brand changes to track delta on identical metrics5.
Measurement boundary
AI answer content is non-deterministic — back-to-back runs differ substantially. An audit produces a snapshot, not a trend, which is why repeat runs per prompt and platform and a defined re-audit cadence are structural, not optional2. Provider sampling is bounded: low-frequency or enterprise-specific query formulations may go unrepresented. An audit does not substitute for ongoing monitoring when competitive dynamics shift weekly. The audit design must also fix and disclose the access path — API-based and UI-observed sweeps of the same provider return different numbers (API vs UI Measurement). Two first-party sources can be read alongside the audit, both partial: Search Console's Generative AI report (impressions in AI Overviews and AI Mode, combined)7 and Bing Webmaster Tools' AI Performance (citations and grounding queries across Copilot and Bing)8. A re-audit is comparable only if prompt set, providers, access path and counting rules are unchanged — see Measurement Consistency.
Distinct from
From Mention Rate and Citation Rate: those are constituent metrics an audit produces and contextualises across providers and prompt categories. The audit is the methodology that yields them. From Brand Recommendation Share: a narrower sub-dimension (head-to-head category queries) within the audit's share-of-voice analysis. From Generative Engine Optimization: GEO is the optimisation discipline informed by audit findings — the audit sets the agenda, GEO executes it.
Common mistakes
- Treating a single manual prompt run as an audit. Non-deterministic outputs need repeat runs per prompt and platform; single-run snapshots systematically misstate visibility.
- Building the prompt set only from branded queries. Category-level and buyer-scenario queries reveal organic citation behaviour that brand-direct prompts cannot capture — and that is where the biggest competitive gaps usually sit.
- Conflating provider coverage with completeness. The same brand can show very large citation-volume differences between providers, making single-provider audits misleading for cross-channel strategy.
- Reporting one visibility figure over brand-direct and category prompts together — the two differ by an order of magnitude and answer different questions.
- Delivering a score without implementation specifications. Source-authority mapping, schema gaps and entity-clarity defects are the actionable outputs; a visibility score without a prioritised fix list converts to no intervention6.
Sources & deeper reading
- 1Aggarwal et al. — "GEO: Generative Engine Optimization" (KDD 2024, arXiv 2311.09735)
- 2Passionfruit — "Brand Visibility Audit in ChatGPT, Perplexity & Gemini" (prompt-set + redundancy-sampling methodology)
- 6Metricus — "AI Visibility Monitoring vs One-Time Audits" (audit-vs-monitoring distinction, quarterly cadence rationale)
- 7Search Console Help — "Generative AI performance report (Search)" (impressions only, AI Overviews and AI Mode combined; all websites from 31 August 2026)
- 8Bing Webmaster Tools — "New AI Visibility Insights: Intents, Topics, Citation Share, Compare" (16 June 2026; AI Performance report since 10 February 2026: citations and grounding queries across Copilot, Bing AI answers and partner surfaces)
Update history
- 2026-10-08 — Replaced a dead source (Visiblie) with Victorious' Q2 2026 report and added its recognition-versus-mention finding (96 % / 89 %). The 'three runs, stable median' rule is now stated as a principle: repeats stabilise the aggregate, prompt choice is the larger lever. Added the first-party reports from Search Console and Bing, and the link to Measurement Consistency.