How to Read Google Search Console’s Generative AI Performance Report

Google Search Console’s Generative AI performance report tells you how often links to your site were shown in AI Overviews and AI Mode on Google Search. It is an impression report, not a separate traffic or conversion report. Start with the property-level trend, use the Pages table to identify the URLs being surfaced, and segment by country, device, and date before deciding what to update. Do not treat the number as clicks, citations across all AI assistants, or searches for your brand.

That distinction matters because the same page can gain AI visibility while its ordinary search clicks remain flat. The two signals are useful together, but they do not measure the same event.

Table of Contents

16 sections Show

Read the Search Console report in five passes

1. Open Search Console > Performance > Generative AI for the verified property. If you cannot see the report, check your property access and whether Google has data to display before assuming AI visibility is zero.
2. Set a complete date range and record the property-level impression total. Avoid treating the newest dotted segment as final.
3. Open Pages and identify the canonical URLs gaining or losing impressions. Do not add page rows together and call the result the property total.
4. Check Countries and Devices for the same date range. A gain outside your target market may be less useful than a smaller gain inside it.
5. Export the report, label its impression data accurately, and compare it with ordinary Search clicks and GA4 outcomes in separate columns.

The rest of this guide explains the counting rules and the places where those five steps are easiest to misread.

What counts as a Generative AI impression?

The report currently includes two Google Search experiences: AI Overviews and AI Mode. An impression is counted when a link to your site is shown to a user in a supported generative AI feature. It does not mean that the person clicked, read your page, or saw a mention of your brand in the generated prose.

It is also narrower than the phrase AI visibility. The report does not cover ChatGPT, Claude, Gemini’s standalone app, or every place a model might quote your content. Google says the Search report uses Web search data. The separate Discover experience is not included here, and experiments in Search Labs are excluded.

Use this sentence in a stakeholder update: “Our site’s links received X reported impressions in Google’s supported generative AI Search features during the selected period.” Avoid “AI sent us X visitors.”

Start with a complete period, not today’s dotted line

Open the report from your verified Search Console property and choose a date range that has finished collecting data. Google’s newest values may be preliminary. On the chart, preliminary data appears as a dotted line and can change over the next few hours.

For a weekly review, compare the last seven complete days with the preceding seven complete days. Keep the property, date range, and filters identical. Search Console dates use Pacific Time, so a report exported from another time zone may appear to begin or end on a different calendar day.

Record four things before opening any page row:

1. Property-level impressions for both periods.
2. The absolute change and percentage change.
3. Whether any day in the comparison is preliminary or affected by a known reporting anomaly.
4. The first day the trend changed, if the movement is sustained.

One day of movement is a prompt to investigate, not proof that an algorithm update or content edit worked.

Why the chart total will not always equal the Pages table

The chart normally aggregates impressions at the property level. If two pages from the same site are shown in one generative AI result, the property chart can count one impression. The Pages table groups the data by URL, so each page can have its own impression. Adding all page rows can therefore produce a different figure from the chart.

This is not, by itself, a tracking failure. Use the chart to report total property visibility. Use the Pages table to decide which content is gaining or losing visibility. Google also applies normal performance-report table limits, including the 1,000-row limit, so a large site’s exported page rows are not a substitute for the property total.

The Pages table generally assigns performance to the canonical URL and groups by the final linked URL after redirects. If you are surprised to see a different URL than the one you edited, check canonicalization and redirects before assuming the report is wrong.

Read the four dimensions as four separate questions

• Pages: Which URLs appeared as links in supported AI Search features? This does not reveal the prompt or query behind an individual impression.
• Countries: Where did the search originate? A country alone does not identify your target buying audience.
• Devices: Was visibility reported on desktop, mobile, or tablet? The split does not prove your page design caused the difference.
• Dates: Is the change persistent or confined to a short interval? The pattern does not identify the product or ranking change that caused it.

Start with Pages and sort by impressions. Mark your important product, comparison, and help URLs separately from broad informational content. A large rise on a low-value definition page is not equivalent to a rise on a purchase-intent page. Then check Countries and Devices to see whether the change occurred where your audience actually searches.

Do not assume that a page absent from the table is blocked from AI Search. It may have too few impressions, be outside the selected range, or be represented by another canonical URL. Check its normal indexing and Web performance before drawing a conclusion.

Three tempting calculations that the report does not support

AI click-through rate. The dedicated report documents impressions, not an AI-specific clicks metric. Dividing all Web Search clicks by AI impressions mixes different populations and does not produce an AI CTR.

AI queries by page. The report’s documented dimensions are Pages, Countries, Dates, and Devices. The normal Search performance report has queries, but its Web totals combine AI features with other Search activity. A query and a page moving in the same week does not prove that query triggered the page’s AI impression.

AI Overview versus AI Mode share. Google currently groups both features in this report without a documented feature-type dimension. Do not label a chart segment as “AI Mode” unless Google actually exposes such a filter in the interface you are using.

External clicks on links in AI Overviews and AI Mode can be included in ordinary Search Console Web click data. That still does not turn the dedicated impression report into a source of AI-only click attribution.

Export a weekly review that somebody can audit

The report’s Export control downloads chart and table data. Note a small but important behavior: values displayed as ~ or - in the interface become zeros in the export. A spreadsheet formula that treats those as measured zero impressions can mislead an analyst.

Use a compact review sheet, one row per period, with these fields:

• Property and period: Verified property, start date, end date, and Pacific Time basis.
• Property AI impressions: Current and previous complete-period totals.
• Changed URLs: Top gaining and declining canonical URLs.
• Geography and device: Where the movement is concentrated.
• Normal Web performance: Clicks and impressions for the same URLs, labeled as all Web Search rather than AI-only clicks.
• Business outcome: GA4 landing-page key events or qualified leads, separately labeled.
• Confidence and next action: Observation, hypothesis, owner, and review date.

Export before and after a material change, retain the date range, and annotate known reporting issues. Do not quietly replace a historical export after a later backfill.

A worked example without false attribution

Hypothetical example: A site reports 12,000 Generative AI impressions for one complete week versus 9,000 the week before. Four URLs account for most of the growth in the Pages table. Their page rows sum to 15,400 impressions, which exceeds the property chart’s 12,000.

First, do not call the difference a tracking bug. The chart and the page rows use different aggregation. Second, inspect countries. If most of the gain came from a target market, the change is more commercially relevant than a gain in an unsupported market. Third, check ordinary Web clicks to those four URLs and relevant GA4 outcomes.

Suppose Web clicks also increased. The defensible conclusion is: “Reported AI Search visibility rose, led by four URLs in our target market. Total Web clicks to those URLs rose too, but Search Console does not isolate the AI-generated portion of those clicks.” The tempting conclusion, “AI Overviews generated the extra clicks,” is not supported.

Keep GA4 and Bing measurements in their own columns

GA4’s AI Assistant channel tracks recognized referral traffic from assistants such as ChatGPT and Claude. Google says that channel excludes Google AI Overviews and AI Mode, which remain within Organic Search. It is a traffic measure, not the same denominator as Search Console’s Generative AI impressions.

Bing Webmaster Tools’ AI Performance report measures visible citations across its supported AI experiences. A Bing citation is not a Google impression, a click, or a GA4 session. Keep these three measures separate in client or executive reporting. If you use BrandJet to monitor how your brand appears in AI answers, treat that as another observation layer, not as a replacement for first-party Search Console or GA4 data.

What to change when impressions fall

First compare complete periods and check Google’s Search Console data-anomalies record. Then inspect which URLs, countries, and devices changed. For affected URLs, verify that they are indexable, canonicalized as intended, accurate, and still satisfy the query intent they were built for. Check ordinary Search performance and competing pages before rewriting copy.

An impression decline alone does not prove a penalty. Search demand, result layout, model behavior, competitor coverage, and reporting changes can all alter the count. Write down the likely explanations, make one clear change at a time, and review the same metrics after enough complete data accumulates.

FAQ

Does Search Console show AI Overview impressions?

Yes. The Generative AI performance report for Search includes impressions from AI Overviews and AI Mode together. It does not currently document a dimension that separates the two.

Can I see the exact AI queries that showed my page?

Not in this dedicated report. The regular Search performance report shows queries for Web Search overall, but it does not let you label one of those queries as the cause of an AI impression simply because the dates and page match.

Does a rise in AI impressions mean AI traffic rose?

No. Impressions mean your links were shown in a supported AI feature. Check overall Web clicks and landing-page outcomes separately, and do not claim AI-specific clicks unless a first-party report actually isolates them.

Why is the sum of page impressions higher than the chart total?

The property chart and page table use different aggregation. Multiple URLs can earn page-level impressions while the property appearance is counted once in the chart.

Is the GA4 AI Assistant channel the same report?

No. GA4’s channel measures recognized assistant referrals and excludes Google’s AI Overviews and AI Mode. The Search Console report measures impressions of your links within Google’s supported AI Search features.