Why GSC AI page totals exceed the chart
899 impressions in the property chart. 916 impressions when the rows in Pages.csv are added together. Same date range, same Generative AI view, different total.

That sort of gap can make a report look as though someone has found an extra source of AI traffic. Usually, they have found an aggregation difference. For a business looking at AI visibility in Dubai, the UAE, the UK or Liverpool, it is worth getting this right before a small spreadsheet sum becomes a big claim about leads.
The supplied export: 899 versus 916
The example export covers 4 August to 3 November 2026 with the Generative AI appearance filter applied. The property chart reports 899 impressions. Adding the impression values in Pages.csv produces 916. Country and device totals each return 899.
That does not mean the page export has discovered 17 extra searches, 17 visitors or 17 prospective customers. It means the report is being read at two different levels.
| Report view | Total in the export | What it represents | What it does not prove |
|---|---|---|---|
| Property chart | 899 impressions | Aggregated performance for the property and selected filter | Visits, unique people or enquiries |
| Pages.csv summed | 916 impressions | URL-level rows added together | A replacement property total |
| Country totals | 899 impressions | Property performance split by country | Separate traffic on top of the chart |
| Device totals | 899 impressions | Property performance split by device | Separate traffic on top of the chart |
Property reporting and page reporting are not the same sum
Google’s documentation says the performance chart is aggregated by property while page tables are aggregated by URL, and that totals can differ. It also confirms that Generative AI data sits within Web Search performance reporting. You can read the relevant Google Search Console guidance on Generative AI reporting directly.
In practical terms, a property chart answers: how much eligible visibility did the site receive under this filter? A page table answers: which URLs were associated with that visibility? Those are useful questions, but they are not interchangeable.
A URL-level report is for spotting patterns: perhaps a service page, comparison page and supporting article appear repeatedly. It is not a licence to add every page row and present the result as a new property-wide total.
Direct answer: a Search Console Generative AI page impressions total can exceed the property chart because the chart is aggregated at property level and the Pages report is aggregated by URL. Page rows should not be treated as unique searches, users, sessions or additional AI visibility. The property chart is the cleaner headline figure for overall exposure within the selected period and filter.
Keep Web Search exposure separate from referral visits
There is another wrinkle that gets muddled in cheerful reporting decks. Generative AI data in Search Console is part of Web Search performance. It is not a separate analytics channel that can be added to ordinary organic search as though the business has acquired a second internet.
Referral sessions are different again. If analytics records a visitor arriving from a named AI product or another referring source, that is a session measurement in your analytics platform. It has its own attribution limits, consent effects and tracking setup. It should be reported alongside Search Console visibility, not merged into it.
For example, an agency might download Pages.csv, put a SUM formula beneath the impressions column, see 916, then add that to a monthly organic-search figure. That creates a neat-looking number and a rather less useful one. The 916 is not an additive Web-plus-AI traffic total.
What the export can, and cannot, tell a service business
The export can help a consultant, clinic, real estate firm or hospitality business see whether its pages are receiving Google AI-related exposure. It can show which URLs surface in the page-level report and whether that picture changes across periods.
It cannot identify the exact prompts behind an impression. It cannot name the person who saw it. It cannot tell you whether that impression produced an enquiry. Search Console impressions are visibility measures, not a lead ledger.
That last bit is where reporting often goes wonky. A contact form success message does not prove an email arrived either, especially when an old inbox, SMTP change or spam rule is involved. Likewise, an impression does not prove a commercial outcome. Both need their own evidence trail.
Use three reporting lanes, then compare them honestly
A more useful review keeps the measures apart before looking for a relationship between them. This is also a sensible starting point for an AI readiness audit compared with an SEO audit, because visibility measurement and broader website improvement are related but not the same job.
- Exposure: Record the property-level Generative AI impressions for the agreed date range and filters.
- Visits: Review analytics referral sessions and organic landing-page sessions separately, with source definitions written down.
- Observed enquiries: Check form submissions, booked calls, phone tracking where used, email delivery and CRM records. Keep the original source data where possible.
Then ask a modest question: are the pages gaining exposure also appearing in the journeys that produce attributable enquiries? Sometimes there will be a useful connection. Sometimes there will not. Pretending the latter cannot happen is how budget drifts towards a metric with no demonstrated commercial value.
When content-preview controls are the actual issue
If the question is not the arithmetic but whether Google may show content previews from particular pages or passages, that is a different diagnostic boundary. Review Google snippet controls, data-nosnippet and content preview settings before altering directives. Those controls may affect what Google can use or preview, but they do not turn page-table sums into unique visitor counts.
The practical recommendation is simple: use the property chart for the overall Generative AI exposure figure, use page rows to investigate URLs, and keep referral sessions and recorded enquiries in separate columns. Add a note beside every report stating the date range, filter, platform and definition. Boring admin, admittedly, but it saves the person presenting the report from having to row back later.
Measure exposure before claiming outcomes
AI visibility is worth watching, particularly for a service business with clear expertise, useful pages and decent enquiry tracking. It is not worth inflating. Measure the visibility you have before claiming the traffic or enquiries you cannot attribute.
For a wider review of content clarity, technical access and how your site is interpreted, see the AI readiness audit for business websites.
Request an AI visibility review that separates exposure, visits and observed enquiries.