In plain English
Measure AI-search visibility as a chain of observable signals: crawl and index health, provider-reported appearances or citations, visits from identifiable referrals, useful on-site actions, qualified enquiries and sales outcomes. Keep each source separate, record coverage limits, and compare trends over time. Prompt spot-checks can diagnose errors, but they are not a reliable market-share metric.
Which metrics belong in an AI-visibility ladder?
Begin with eligibility and move towards commercial evidence. Record whether priority pages are accessible, indexable and represented in relevant webmaster tools. Above that, capture provider-reported AI appearances, cited pages or grounding queries where available. Then measure identifiable referral sessions, useful page actions, enquiries, qualified opportunities and client-confirmed sales. Each rung answers a different question and should retain its own data source and definition.
Do not combine citations, impressions, visits and leads into one visibility score unless the calculation and limitations are explicit. A citation can occur without a click; a referral can arrive without a known prompt; an enquiry can have several earlier influences. For a South African service business, segment by genuinely served geography and service where the data supports it, but keep low-volume or withheld data labelled unknown rather than filling gaps with estimates.
Sources for this section: Introducing Search Generative AI performance reports in Search Console, Introducing AI Performance in Bing Webmaster Tools Public Preview.
What do provider-native reports actually show?
Google announced dedicated generative AI performance reports for Search and Discover with views of impressions, pages, countries, devices and dates. Its June 2026 announcement also says the reports are being rolled out to a subset of sites. Availability must therefore be checked for the property being reported; an absent report is not evidence of zero visibility. Check the account before reporting. Preserve the report name, export date and selected filters.
Bing's AI Performance public preview reports total citations, average cited pages, sampled grounding queries, page-level citation activity and trends across supported Microsoft and partner experiences. Bing explicitly says those counts do not show placement, authority, ranking or a page's role in an individual answer. Keep Google and Bing measures in separate columns. Their surfaces, definitions and coverage are not interchangeable, even when both use the word visibility.
Sources for this section: Introducing Search Generative AI performance reports in Search Console, Introducing AI Performance in Bing Webmaster Tools Public Preview.
How can referrals connect visibility to business outcomes?
OpenAI says ChatGPT adds utm_source=chatgpt.com to referral URLs from ChatGPT search results, allowing publishers to identify that inbound traffic in analytics. Review landing pages, session source and useful events rather than reporting the referral count alone. Retain the landing URL and timestamp. Google Analytics defines session-source dimensions around the origin of a session, but those fields do not reconstruct every earlier exposure or conversation that influenced the visitor.
Carry the available source into the enquiry record, then connect it to qualification and sales outcomes under agreed definitions. Preserve direct, unknown and unattributed values instead of forcing every record into AI search. Consent choices, copied links, browser behaviour, phone calls and cross-device journeys can interrupt the chain. A useful dashboard shows the observed referral and downstream outcome while stating how much of the journey remains unavailable.
Sources for this section: Publishers and Developers - FAQ, Traffic acquisition report.
How should the monthly review avoid false certainty?
Compare consistent periods and annotate releases, tracking changes and provider-report changes. Review which pages and subjects gained or lost observed visibility, whether cited passages remained accurate, and whether referral sessions produced useful actions. Keep raw counts beside rates and denominators. Where a provider supplies only a sample or limited rollout, repeat that limitation in the chart rather than hiding it in an appendix.
Use a small, versioned prompt set only as a diagnostic observation: record platform, model or mode, location setting, date and output. Results can vary and do not reveal population-level share, so do not average manual prompts into a market metric. End the review with one testable action, an owner and a recheck date. No dashboard can prove future ranking, citation, lead volume or revenue; it can make the next decision better evidenced.
Sources for this section: Optimizing your website for generative AI features on Google Search, Introducing AI Performance in Bing Webmaster Tools Public Preview.
Official references
- Introducing Search Generative AI performance reports in Search Console (Google Search Central Blog)
- Introducing AI Performance in Bing Webmaster Tools Public Preview (Microsoft Bing Webmaster Blog)
- Publishers and Developers - FAQ (OpenAI Help Center)
- Traffic acquisition report (Google Analytics Help)
- Optimizing your website for generative AI features on Google Search (Google Search Central)