Google Search Console’s Web performance data can include eligible activity from Google AI features such as AI Overviews and AI Mode. It does not present that activity as a dedicated, all-platform AI visibility report. That distinction matters when a team compares it with a ChatGPT visibility test and assumes the two datasets should agree.
They should not. Search Console answers a Google Search question. A ChatGPT test answers a prompt-level discovery question. Referral analytics answers whether a visit arrived with a detectable referrer. Each is useful. None can substitute for the others.
For marketing teams in Dubai, the UAE, the UK or elsewhere, this is a familiar reporting trap: Google-related visibility looks healthy, while commercially important unbranded prompts produce weak mentions elsewhere. The wrong response is to average everything into one reassuring AI number.
What Google Search Console’s AI feature data actually covers
Google documents that traffic from AI features is included in Search Console’s standard Web search reporting. That includes eligible impressions, clicks and positions associated with Google AI experiences, rather than a separate reporting area that isolates every AI feature interaction.
Google’s documentation on AI features in Search Console is deliberately narrow on this point: it describes reporting for Google Search. It does not claim to measure appearances in ChatGPT, Gemini, Perplexity, Copilot or any other independent assistant.
Google Search Console AI visibility is evidence of how your site performs in Google’s Web search environment, including eligible AI feature activity. It is not evidence that a brand was cited, described or recommended across the wider AI assistant market. To assess broader AI visibility, teams need separate prompt observations, referral data and commercial outcome data, interpreted as different evidence types.
This also means you should be cautious with a rise or fall in Web impressions. It may be meaningful, but it is not automatically an AI Overview trend. Search Console does not provide a clean switch that says this specific impression came from one named AI answer.
Three datasets, three different questions
| Evidence source | Question it can help answer | What it cannot prove |
|---|---|---|
| Google Search Console Web data | Whether pages received eligible Google Search impressions and clicks, including relevant AI feature activity | Visibility in ChatGPT, Perplexity, Copilot or other non-Google platforms |
| Referral analytics | Whether identifiable visits arrived from an AI platform or referral path | Whether the platform mentioned you without generating a click |
| Prompt-level testing | Whether a platform currently names, describes, cites or omits you for a defined prompt | Market-wide traffic, demand or stable long-term platform behaviour |
| Commercial outcome data | Whether observed visits and enquiries contribute to qualified pipeline or revenue | Which answer, citation or feature caused every conversion |
Referral data is about visits, not mentions
Analytics platforms can show visits attributed to known referrers, where that information is passed through. This is useful for finding traffic that arrives from an AI assistant, an AI-powered search result or a linked source surfaced in an answer.
It is not a visibility census. A person may read an answer, note your company name and search for it later. Another may open a native app that does not pass a useful referrer. A third may be given your name but decide not to visit. Zero recorded ChatGPT referrals does not prove zero ChatGPT visibility.
Equally, a handful of referrals do not prove broad recommendation. Check the landing pages, source dimensions, engagement and enquiry route before treating a referral spike as a commercial trend. A common operational issue is a referral landing on a campaign URL that immediately redirects and strips useful parameters. Fix that tracking path before writing a grand conclusion about AI demand.
Prompt tests observe discovery and description
Prompt testing is closer to the question most teams really mean: does this assistant surface our business when a buyer asks a realistic question? It can also reveal whether the assistant gets your category, location, service boundary or supporting evidence wrong.
That requires a controlled prompt set, recorded dates, platform conditions and a scoring rule. It should include unbranded discovery prompts alongside branded checks. Being correctly described after typing your company name is a different capability from appearing in a query for a provider in your category.
A structured ChatGPT visibility test is useful for recording that distinction. It is not a replacement for analytics, and it should not be treated as a ranking tracker. One answer is an observation under particular conditions, not a verdict on the whole market.
A measurement model that does not pretend the sources match
Keep the datasets separate first. Then compare their direction and commercial relevance.
- Use Search Console for Google: review Web queries, landing pages, impressions and clicks. Mark significant site releases, content changes and technical incidents so you do not confuse a broken canonical tag or accidental noindex rule with an AI visibility shift.
- Use referral analytics for attributable visits: create a consistent source view for known AI referrals, inspect landing pages and distinguish referral traffic from branded search that may have followed an AI interaction.
- Use prompt tests for observed answers: test a maintainable set of buyer-like prompts across priority platforms, locations and languages where relevant.
- Use CRM or enquiry data for commercial context: ask how leads found you, retain source detail where possible and look for patterns rather than forcing precise attribution where none exists.
The wider AI share of voice measurement checklist provides the broader model. Search Console belongs inside that model as a Google evidence source, not above it as the master metric.
How to read a conflicting result without panicking
Consider a B2B consultancy that sees strong Web impressions and clicks in Search Console after publishing clearer service pages. Its team then finds no identifiable ChatGPT referrals and poor performance on unbranded prompts such as which consultancy helps with a particular operational problem in Dubai.
There is no contradiction to solve. Google may be finding and serving the pages effectively, while ChatGPT has insufficient, inconsistent or poorly corroborated evidence to surface the business for that discovery task. The pages may also use a service name that differs from the organisation’s structured data or third-party profiles, making the entity harder to resolve.
Start by checking the basics before rebuilding the site: the relevant page is indexable, its rendered body copy is present without relying on JavaScript, its Organisation and service relationships are unambiguous, and the offer is described in the terms buyers actually use. Then assess external corroboration and rerun the same prompt set later.
Record a starting point before changing several things at once. An AI visibility baseline makes it easier to separate a real change from a reporting mismatch.
Questions about Google Search Console and AI visibility
Does Google Search Console include ChatGPT data?
No. Search Console reports Google Search data. It does not report your brand’s appearances, citations, recommendations or traffic inside ChatGPT. You may see ChatGPT-related visits in analytics where a referral is passed, but that is separate from Search Console and does not measure every mention.
What does Google’s AI reporting measure?
Google includes eligible activity from its AI search features in standard Web performance reporting. The data can contribute to impressions, clicks and position reporting under Google’s documented rules. It should be interpreted as Google Search performance, not as a platform-neutral measure of AI visibility.
What other AI visibility data should we collect?
Collect a documented prompt set across priority assistants, identifiable referral sessions and landing pages, branded search context, enquiry source information, and changes to pages or technical access. Keep the records separate, then use them together to investigate patterns that matter to discovery and commercial outcomes.