How to Read Google's New AI Search Reports in Search Console

What Google Has Added

Search Console now includes dedicated Generative AI performance reports for Search and Discover. This is a meaningful improvement because marketers previously had to infer AI Overview and AI Mode visibility from blended search data, third-party tracking or manual checks.

The new reports do not answer every question. They do give site owners a first-party view of how pages are appearing across Google's generative experiences, with dimensions such as page, country, device and date. Rollout is phased, so two properties managed by the same team may not receive access at the same time.

The most important habit is to treat the report as a scoped dataset: it describes Google's surfaces, not “AI search” as a whole.

Find the Right Property Before Reading the Numbers

Search Console data belongs to a property. A domain property can include protocols and subdomains, while a URL-prefix property covers only the specified prefix. If your team switches between them, the totals may differ without either report being wrong.

Before analysing the AI report:

1. confirm the production domain is selected
2. check that important HTTPS and subdomain variations are covered
3. note the date range and any active filters
4. record when the report first became available
5. avoid comparing a partial rollout period with a complete prior period

This simple control prevents many false “visibility dropped” conclusions.

What the Core Dimensions Tell You

Pages identify which URLs are being surfaced. Look for concentration: is visibility confined to one article, or distributed across service, product and educational pages?

Countries show where visibility occurs. For an Australian business, overseas impressions may be useful, irrelevant or a sign that a page lacks local context. Interpret the geography against the business model rather than assuming more countries are always better.

Devices help you see whether exposure is concentrated on mobile or desktop. Device mix can influence what happens after the impression, so compare it with page experience and conversion behaviour.

Dates reveal direction and volatility. Use weekly or 28-day comparisons where possible. Daily data is useful for diagnosing a specific change, but it is easy to overreact to normal movement.

Google's launch documentation also notes that this data continues to be included in overall performance reporting. The dedicated report is a clearer lens, not an entirely separate universe of Search activity.

Five Questions the Report Can Answer

1. Which content themes are earning AI visibility?

Export the page data, group URLs by topic and compare their share of impressions. A cluster of practical implementation guides may outperform broad opinion pieces, or the reverse. That is a content signal grounded in your own site.

2. Did a technical change affect visibility?

Overlay deployment dates, indexing issues and template changes. If impressions fall across many unrelated pages on the same date, investigate a site-wide cause before rewriting content.

3. Are new pages being discovered?

Filter to recently published URLs and record the time between publication, indexing and first AI impression. Do not turn one example into a universal promise; build a distribution from several posts.

4. Which pages deserve improvement?

A page with sustained impressions but weak downstream engagement may have an expectation mismatch. Review its title, opening answer, visual quality and connection to the next useful action.

5. Is visibility aligned to the market?

Country and device breakdowns can show whether the audience matches the commercial footprint. For a local service firm, strong Australian visibility is usually more valuable than a larger volume of irrelevant global impressions.

What the Report Cannot Prove

The report cannot tell you that a user saw a positive recommendation, understood your brand correctly or chose you over a competitor. It does not measure ChatGPT, Perplexity, Claude or every Gemini interaction. It also does not replace revenue or lead tracking.

Avoid these common interpretations:

• “AI impressions increased, therefore leads increased.”
• “This URL has the most impressions, therefore it is our best article.”
• “Search Console shows no data, therefore we never appear in AI answers.”
• “One week's decline means an algorithm penalty.”

Each statement requires additional evidence.

Build a Useful Weekly View

Create a small scorecard with five lines:

Keep external AI referrals in a separate block. This prevents a growing ChatGPT referral line from being accidentally attributed to Google's report.

A Sensible Diagnostic Sequence

When the data changes, investigate in this order:

1. Scope: property, date range, filters and rollout coverage.

2. Technical health: indexing status, robots rules, canonicals, server errors and rendering.

3. Page mix: whether one high-volume page created most of the movement.

4. Content change: material edits, freshness, accuracy and whether the page still satisfies the task.

5. External context: seasonality, news cycles and changes in customer demand.

This order stops teams from rewriting a useful article when the actual issue is a property filter or a failed deployment.

The First 30 Minutes in the Report

Start by exporting the available page and date data. Mark the ten pages with the highest visibility, the ten with the largest recent increase and any important commercial pages with no visibility. Then compare those groups with indexing status and GA4 landing-page outcomes.

Finish with three actions only: one technical fix, one content improvement and one measurement improvement. A report becomes valuable when it changes a decision, not when it creates another dashboard tab.

Official Sources

• Google: Introducing Search Generative AI performance reports
• Google: Optimising for generative AI features
• Google: Search Console performance documentation