Guide: How to Use Google's New AI Impressions Report to Find Your Hidden Search Visibility

Understanding AI Search Visibility Reporting

Google added a dedicated Generative AI view to Search Console in June 2026 that separates out impressions from AI Overviews and AI Mode — showing you when your content was surfaced inside Google's AI-generated answers. This is distinct from the Discover AI report and lives within the Performance section of Search Console.

The critical thing to understand before building any workflow around this: it's an impressions report, not a performance dashboard. It shows where Google is willing to cite your content inside AI answers and how often. It does not tell you which query triggered the answer, whether anyone clicked, or whether it drove a sale. Its real value comes from comparing AI visibility against your conventional organic visibility — that comparison is what drives every actionable insight in this guide.

Important caveats to verify before relying on this data: the report is still rolling out and may not be available on your property yet. It currently shows impressions only — no clicks, CTR, position, or query data. Recent data is preliminary and may revise, so avoid reacting to single days. The data table caps at 1,000 rows, which matters for large catalogues. And AI impressions are already counted inside your overall Performance totals — this view is a separated lens on the same data, not an additive number.

What You'll Need

• Search Console access to the property (confirm you actually have the Generative AI view — not all properties have it yet)

• Ability to export both the AI report and the standard Performance report

• A spreadsheet or Claude to join the two exports

• A consistent date range decided before you start (use a rolling 28-day or full-month window — short ranges are actively misleading due to rollout noise)

• A page-type taxonomy for your store (this is the single most important prep step for ecommerce — covered in Step 3)

Step 1: Access and Export the Data

Find the Report

Open Search Console, navigate to Performance → Search results, and look for the Generative AI tab. Some accounts display it as a dedicated view. The Discover AI report is separate — this process covers Search only.

Set Your Date Range

Apply the window you decided on during setup. Use a meaningful period — weeks, not days. Whatever range you choose, you must use the identical range for both exports or the comparison falls apart.

Export AI-Visible URLs

Pull the Pages table, which gives you URL plus impressions. Segment by country and device only if strategically relevant (for example, if you sell into multiple markets). Otherwise keep it simple.

If you have a large catalogue, watch the 1,000-row cap. If your AI-visible URLs exceed that, the export is truncated — you're seeing the top of the list, not the complete picture. For large stores, filter by path (such as /blog/ versus /products/) and export in segments rather than assuming the top 1,000 tells the whole story.

Export Conventional Performance Data

From the standard Performance report, pull impressions, clicks, CTR, average position, and top queries for the same URLs and the same dates. This is where click and query context live — the AI report doesn't have them, so you borrow them from here.

You now have two exports keyed on URL. That pairing is the foundation of everything that follows.

Step 2: Read the Numbers Correctly

Three rules. Break them and every downstream conclusion is contaminated.

Property-level and page-level impressions don't add up, and that's expected. In the chart (property level), if two of your URLs appear in one AI response, that counts as one impression. In the Pages table, each URL may get its own impression. Summing page-level impressions won't reproduce the chart total — the aggregation dimension is different. Don't try to reconcile them, and don't report a page-level sum as property-level exposure.

AI impressions are not traditional impressions — never blend them. In classic search, an impression roughly means a listing was shown for consideration. In an AI answer, the answer itself is the product and your link might be prominent, buried among citations, hidden until expansion, or only surfaced after a follow-up. The counting rules even differ by feature — in AI Overviews a link must be scrolled or expanded into view to count, while in AI Mode a follow-up is a new query that can generate fresh impressions. Do not pour AI and organic impressions into one "total visibility" number.

No blended CTR. Since the AI report has no clicks, any "AI CTR" you construct is invented, and any blended CTR across both reports smears two incompatible things together.

Step 3: Build the Core Diagnostic — Join AI to Organic

This is the engine of the entire process. Everything before it was setup.

Join and Tag the Data

Join the two exports on URL so you have one row per page with: AI impressions, organic impressions, clicks, average position, and top queries.

Then categorise every page. For ecommerce this matters more than for a generic site, because page type largely predicts whether Google can extract an answer from it. Tag each URL with:

• Page type — product (PDP), collection/category (PLP), buying guide, comparison, how-to/educational, blog, homepage, other

• Topic or category

• Intent — research, comparison, ready-to-buy

• Template — which theme template renders it (templates, not individual pages, often explain visibility patterns)

• Funnel stage

• Last substantial revision date

A list of URLs and impression counts is inventory. The tags are what turn it into analysis.

Optional: Hand to Claude for the Outlier Pass

Give Claude the combined export and your taxonomy, and ask it to bucket pages into the four quadrants below, flag the outliers, and group them by template, type, and topic. It won't replace reviewing the actual pages, but it handles the categorisation grunt-work fast and consistently.

Here is a joined dataset of AI impressions and organic search data for [STORE NAME], tagged by page type, template, topic, and intent.

Bucket each page into one of four quadrants:
A — High organic, low AI
B — Modest organic, high AI
C — High organic, high AI
D — Low organic, low AI

For each quadrant:
- Flag outlier pages (unusually high or low relative to peers in the same page type)
- Group pages by template, page type, and topic
- Note any patterns in which templates or content types concentrate in each quadrant

Return as a structured summary with the outlier pages listed explicitly and pattern observations for each quadrant.

Step 4: Interpret the Four Quadrants

Plot each page on two axes — organic visibility and AI visibility — and read the four resulting states.

Quadrant A — High organic, low AI. Ranks well but rarely appears in AI answers. Not automatically a problem — not every query triggers an AI response, and not every page type is useful for synthesis. But review whether the page answers questions directly near the top, whether key information is in plain HTML rather than trapped in tabs or images, and whether headings are descriptive enough for a passage to stand alone if extracted.

Quadrant B — Modest organic, high AI. Punches above its ranking weight in AI answers. Study these first — they're your most instructive pages. Look for recurring traits: direct answers at the start of sections, strong heading structure, original first-party data (your review counts, test results, sizing data), clean tables and lists, tight topical scope, and language that mirrors how customers actually ask.

Quadrant C — High organic, high AI. Working as intended. Use these as your reference pattern for what "good" looks like on your store, and protect them during redesigns or replatforms.

Quadrant D — Low organic, low AI. A more foundational visibility problem. AI optimisation isn't the lever — basic relevance, indexing, and ranking are. Don't confuse this with an "AI problem."

Run a Concentration Check

If most AI impressions come from a handful of URLs, group them by topic, type, template, and intent. You're often looking for a template story — one template exposes content cleanly in HTML, another hides it behind scripts or decorative markup. Or you've published hundreds of category articles and only a few contain anything original.

Step 5: Remediation — Hypotheses to Test

One hard rule: you cannot prove a single change caused an AI-visibility shift. Demand, seasonality, competing sources, and Google's own systems all move during the same window. Treat everything below as hypotheses to test and monitor over weeks, not guaranteed levers.

For the rest of the SOP — including ecommerce-specific remediation candidates mapped by page type (buying guides, PLPs, PDPs, and template-level fixes) plus the monitoring framework — you can find it here, a free gift from me :) 

• New 2026 data shows 43% of US shoppers used AI assistants for product research in 90 days, and 46% now start research on AI platforms, up from 25% in 2024. Traditional search fell from 43% to 24%.

• Amazon received $600M in tariff refunds in Q2 and will automatically reimburse customers in a limited set of traceable cases. The rest funds lower prices, echoing Walmart and Costco's approaches.

• OpenAI builds a new ad type that launches a business agent conversation instead of opening a website. It scrapes the advertiser's site for context, taps product feeds and MCP tools, and captures leads via forms.

• Salesforce finds agentic search as the shopping journey's first step grew 200% year over year, with AI-referred traffic up 150% to 428% per quarter. Just 28% of commerce orgs use agentic AI today.

• Reddit turns community conversations into ad signals, powering shopping formats and AI creative tools. Q2 ad revenue hits $762M, up 64%, with active advertisers up about 70% year over year.

• A 9th Circuit panel rules Amazon cannot block Perplexity's Comet browser from shopping on its site, finding the user, not Perplexity, accesses Amazon's servers. Amazon says it disagrees and is weighing next steps.

• A marketer details running personal-brand marketing across eight channels via AI agents that triage comments, review performance, and draft content. Nothing publishes without human approval by design.

🌪️ n8n - Open-source, self-hostable workflow automation with custom code nodes.

🎥 Kling AI - Filmmaker-friendly AI video with strong character consistency and motion control.

🖥️ Langfuse - Trace, evaluate and debug your AI agents and LLM apps in production.

🤖 Replit - Cloud IDE with an autonomous agent that builds, hosts and deploys apps.

🌐 Bardeen - Automate repetitive browser tasks without code.

😍It was great, all I needed

🙂Good, not great

🙄It sucked