Key takeaways.
- Google AI impressions, recognised assistant visits in GA4 and saved answer citations are separate signals.
- A citation sample needs the question, tool, market and date beside the result.
- Use these signals to inspect specific pages; none proves a qualified lead by itself.
Summarise with AI
When somebody asks whether they show up in AI search, I ask what they mean by “show up”. A page appearing in Google's AI results, a visit from an assistant and a citation in an answer are different signals. You can learn something from each one, but a single visibility score would blur what actually happened.
I would put the numbers beside their sources and dates before deciding whether a page needs work. First, though, it helps to know what each source can really see.
Google AI impressions show exposure on Google.
Google has introduced a dedicated Generative AI performance report in Search Console. It reports impressions for supported Google Search features such as AI Overviews and AI Mode, with views by page, country, device and date. Google says the report reached worldwide rollout by 31 August 2026, although a property with too little eligible activity may not show it.
An impression means a link to the site was shown within a supported feature. It is useful for spotting which pages appear and how that changes over time. It is not a lead, a visit, or a complete count of every time the brand was mentioned by an AI system. I would review those page trends alongside normal Search Console performance, using the same date range, then inspect the page itself before drawing a conclusion.
GA4 assistant traffic shows visits that reached the site.
In May 2026, Google Analytics added an AI Assistant channel for recognised assistants. It can help show sessions that arrived from services such as ChatGPT, Gemini, Claude or Copilot. Google explicitly says that this channel excludes visits from Google's AI Overviews and AI Mode. So it answers a different question from the Search Console report.
Look at the landing pages and, where tracking has been checked, the actions those visitors took. A small number of visits may still matter if they lead to a useful conversation. At the same time, a referral report misses people who read an answer and never click. Some referrals can also be hard to classify when source information is unavailable. I would not turn an empty referral row into “we have no AI visibility”.
Answer citations need a separate, labelled sample.
If the real question is whether an assistant names the business or cites a page, that needs an answer sample. Pick a small set of actual buyer questions, state the assistant, date and market, and save the answer and cited URLs. Repeat the same questions over time. This is a review method, not a universal ranking. Different prompts and answers can produce different results, and a sampled citation does not tell you how many people saw it.
Adam Balestone argues in an article on Neil Patel's site that large, generic prompt sets can look precise while missing real buyer context. I think the useful part of that argument is practical: choose questions your customers genuinely ask, and keep the limitations of the sample visible.
Put the evidence beside the page.
If I were reviewing this each month, I would keep the sources visible:
- Google AI exposure: Search Console impressions and the pages shown in supported Google AI features.
- Assistant visits: GA4 AI Assistant sessions, landing pages and checked actions on the site.
- Answer sample: recorded mentions and citations for a fixed set of relevant questions, with the tool, market and date attached.
Then ask what the evidence suggests about a specific page. Does it clearly answer the buyer's question? Does it show real experience, examples and the detail somebody needs to make a decision? Does it lead naturally to the right service or next step? Google's AI search guidance still points back to useful, distinctive content and the technical foundations of SEO. There is no special markup or volume of AI-written articles that guarantees an appearance.
If a page appears for relevant buyer questions but sends few visits, I would read the answers and the page before changing anything. That is a useful check whether the discovery happened in a traditional result, a Google AI feature or another assistant. It is also how we approach SEO work: look at the page and the evidence together, then decide what is worth improving.
