---
title: "How to Read the New Search Console Generative AI Performance Reports and What Bing Webmaster Tools Adds"
canonical: https://snezzi.com/blog/search-console-generative-ai-reports-vs-bing-ai-performance/
source: https://snezzi.com/blog/search-console-generative-ai-reports-vs-bing-ai-performance/
published: 2026-09-02
modified: 2026-09-08
author: "Gautham Seshadri"
category: "AI Visibility"
---

> Canonical page: https://snezzi.com/blog/search-console-generative-ai-reports-vs-bing-ai-performance/

# How to Read the New Search Console Generative AI Performance Reports and What Bing Webmaster Tools Adds

Google Search Console's generative AI performance report tracks impressions from AI Overviews and AI Mode but supplies no clicks, CTR, or query data. Bing Webmaster Tools reports citations, grounding queries, and intent signals that Google currently omits. Site owners who review both reports obtain a clearer view of presence in generative answers than either tool delivers alone.

The new Google reports rolled out worldwide in 2026. They mark the first time Search Console surfaces direct data on how often site URLs appear inside AI-generated answers. Many site owners now check these reports to understand visibility in this channel. The reports stop short of showing whether those impressions produced clicks or which exact queries triggered them. Bing Webmaster Tools already supplies several of the missing pieces, including total citations and the specific queries that caused content to be referenced.

This guide explains what each report measures, how to access and interpret the data, and how the two tools complement each other. The focus remains on practical steps and realistic expectations rather than speculation about future features.

## What the Google generative AI performance report actually measures

The report counts impressions, defined as the number of times site URLs appear inside AI Overviews and AI Mode on Google Search. It does not count clicks, conversions, or any form of engagement beyond the initial display. An impression simply records that a URL entered a generative answer at some point during a search session. That definition is narrower than classic Search performance metrics, and treating it like a ranking or traffic score will produce the wrong conclusions.

[Google's 2026 developers blog post on gen-AI performance reports](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports) states that the data covers AI Overviews and AI Mode only. Experiments run inside Search Labs are excluded, so lab tests never inflate the totals you see in production reporting. The same rollout note confirms that data became available to all websites worldwide on August 31, 2026. From that date forward, property owners can review the same impression series without waiting for a staged invite.

Impressions break down by page, country, device, and date. You can select hourly, daily, or monthly granularity depending on how tightly you need to inspect a spike or a drop. The report still shows no query-level detail and no information on where a URL appeared inside the answer, whether as a main link or a footnote. Because the metric is limited to impressions, a high number does not indicate traffic impact on its own. Site owners who treat these numbers as rankings or click predictions misread the data.

The report also excludes any pages that have opted out of generative features through the dedicated toggle. Once a property opts out, those URLs stop appearing in AI Overviews and AI Mode, and the impression series for those pages goes quiet. That tradeoff matters when you weigh short-term control against long-term visibility in generative answers.

## How to open and configure the report in Search Console

Open Google Search Console and select the Performance section in the left menu. Two tabs appear: Search and Discover. Choose the generative AI view under either tab to load the new report. The interface presents a line chart of total impressions over time and a table that updates when dimensions change. Start with a short recent window so the chart is readable, then widen the range once you know which days matter.

Apply the Pages dimension to see performance for individual URLs. Switch to Countries or Devices to compare where generative impressions concentrate. Set custom date ranges that respect the standard data-retention window for Search Console. Filters can be added for specific countries or device types, but the table still caps at 1,000 rows. If your property has more than 1,000 qualifying pages in the selected window, export early and continue the analysis offline rather than assuming the on-screen table is complete.

Export the chart and table through the download control at the top of the report. The exported file converts approximate values shown as ~ or - into zeros. Review the exported data in a spreadsheet to spot trends that the web interface compresses, especially when you need side-by-side period comparisons. Keep date ranges consistent across exports when you compare weeks or months, and label each file with the property, view (Search or Discover), and exact range so later audits stay clean.

A practical configuration habit is to save one baseline export before a major content change and a second export after the change settles. Because the report is impressions-only, you are looking for directional movement in which URLs surface inside generative answers, not for a finished traffic story. Pair that habit with the Bing workflow described later so you can attach query and intent context to the same pages.

## Reading the chart and table correctly

The chart displays property-level totals and updates when the date range changes. The table aggregates differently depending on the selected dimension. When Pages is chosen, each row represents one canonical URL. When Countries or Devices is chosen, rows represent aggregated property-level data. Mixing those views without noting the dimension will make totals look inconsistent even when both views are correct.

Preliminary data appears as a dotted line on the chart and can shift for several hours after the initial recording. Final values replace the dotted line once processing completes. If you make a decision on the same day a spike appears, wait for the dotted segment to solidify before you rewrite a page or change internal links. Early readings are useful as alerts, not as locked baselines.

Discrepancies between the chart total and the table total are normal because the two views use separate aggregation rules. Page-level rows always use the canonical URL that Google selected. Country and device rows roll up to the property level even when multiple pages contributed. Values shown as ~ in the interface become zeros in the downloaded file, which can affect percentage calculations if you do not handle them in the spreadsheet. Replace those zeros with explicit nulls when you calculate share of impressions so empty cells do not drag averages downward.

A reliable reading sequence looks like this:

1. Confirm the date range and whether you are in the Search or Discover generative AI view.
2. Read the chart for property-level direction, ignoring dotted segments until they finalize.
3. Switch the table to Pages and note which canonical URLs drive the bulk of impressions.
4. Switch to Countries and Devices only after you understand the page set, so geography and device splits stay anchored to real URLs.
5. Export both the chart and the table, then reconcile ~ and - values before any percentage math.

Following that order reduces false alarms caused by preliminary data or mismatched aggregation.

## Limitations that affect decision-making

The report contains impressions only. [Google's own announcement of the reports](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports) describes the available breakdowns as pages, countries, devices and dates; clicks, CTR, average position and query data are absent. Without those fields, impressions alone cannot reveal whether visibility produced visits or revenue. A page can collect generative impressions and still contribute little to sessions if users never leave the answer surface.

No grounding-query data appears, so you cannot see which user questions caused a URL to enter an answer. The report also does not indicate whether a link sat inside the main response or appeared only as a footnote. Those placement differences matter for brand presence, yet they remain invisible in the current Google view. An opt-out toggle removes the site from all generative features and forfeits any future impressions, which is a blunt control rather than a selective filter.

Click data is listed as planned for a later version, yet no release date has been announced. Until that arrives, treat Google’s series as a visibility volume signal, not a full funnel. Site owners who need query-level insight today must look elsewhere. Treating the current impression numbers as a complete performance picture leads to incomplete conclusions about content quality, internal linking, and commercial priority.

Decision-making improves when you write explicit questions the Google report cannot answer, then assign those questions to another source. Examples include “Which prompts cite this URL?”, “Is the intent commercial or informational?”, and “Did citation share rise after the rewrite?” The last of those is a share-of-voice question rather than a console question; [how to measure share of voice and track competitor mentions in AI search](/blog/how-to-measure-share-of-voice-and-track-competitor-mentions-in-ai-search/) sets out the method. Bing’s AI Performance report is built to answer several of those questions today, which is why the next sections treat the two tools as a pair rather than substitutes.

## What Bing Webmaster Tools AI Performance report adds

Bing Webmaster Tools began reporting AI performance data earlier in the generative-search cycle and now exposes fields Google still withholds. The report tracks total citations and the specific pages cited across Microsoft Copilot and Bing AI summaries. The optimization side of that surface is covered separately in [Bing Copilot optimization](/blog/bing-copilot-optimization-boost-microsoft-search-citations/). [Bing's own help documentation on AI Performance](http://bing.com/webmasters/help/ai-performance-9f8e7d6c) lists these citation counts as a core metric, so you can move from “a URL appeared” to “this page was referenced” with less guesswork.

Grounding queries are included, showing the exact searches that caused content to be referenced. [Search Engine Land's June 2026 coverage of Bing Webmaster Tools AI reporting](https://searchengineland.com/bing-webmaster-tools-updates-ai-reporting-with-intents-topics-citation-share-and-compare-480277) describes intent classification, topic-level visibility, and citation-share comparisons added in that update. Intent labels group queries as informational, commercial, navigational, or related categories, which helps separate research-oriented citations from purchase-oriented ones. Topic views and citation-share trends make it easier to see whether a refresh improved relative presence, not only raw counts.

Historical depth is another practical difference. Bing’s series extends back to November 2025, which supports longer trend lines when you evaluate seasonal content or multi-month rewrites. That window is useful when Google’s generative impression history is shorter or when you need a pre/post comparison that spans more than a few weeks.

These fields let you see not only that content appeared but which questions triggered the citation and how the share of citations changed after updates. The combination of citation counts, grounding queries, intent labels, and topic share supplies context that impression totals alone cannot provide. Even properties with modest Bing traffic can use the report as a diagnostic layer for query and intent patterns that remain hidden in Search Console’s generative view.

## What neither console can tell you

Both reports share a limit that is easy to miss because each tool is comprehensive about its own surface: Google reports on Google, Bing reports on Microsoft. Neither says anything at all about ChatGPT or Perplexity.

That is not a small blind spot. Across two independent B2B brands we track, the citation volume by engine breaks down like this:

| Engine | Brand A | Brand B |
|---|---|---|
| Google AI Mode | 31,421 | 5,508 |
| ChatGPT | 5,060 | 1,183 |
| Perplexity | 2,538 | 3,391 |
| Google AI Overviews | 256 | 318 |

For Brand A, the two consoles between them cover the large majority of citations, and a Search Console plus Bing workflow is close to sufficient. For Brand B, Perplexity alone produces nearly three times the citations that ChatGPT does and more than half the volume of AI Mode, and none of it appears in either console. A team in Brand B's position that reports only from Search Console and Bing Webmaster Tools is describing a minority of its own citation footprint while believing the picture is complete.

The practical consequence is a sequencing one. Set up both consoles first, because they are free, authoritative about their own surfaces, and give you the grounding-query layer described above. Then establish separately what share of your citations they can actually see. If ChatGPT and Perplexity are material for your category, console data is a component of your reporting rather than the whole of it. [AI brand monitoring across ChatGPT, Gemini and Perplexity](/blog/ai-brand-monitoring-how-to-track-your-brand-across-chatgpt-gemini-and-perplexity/) covers what that additional layer involves.

One caution on interpreting any of these numbers. An impression inside an AI feature is not a ranking, and it is not a citation. Being included in the grounding set and being named in the answer are different events, and only the second reliably sends anyone to you. Bing's citation field measures the second. Google's impression field does not distinguish between them.

## How to combine both tools for a fuller picture

Use the Google report to monitor impression volume trends inside AI Overviews and AI Mode. Use the Bing report to examine citation context, intent labels, and competitive share. Pages that appear in both reports often represent high-value content worth maintaining, because dual presence suggests the material is eligible for generative surfaces across ecosystems rather than a one-off anomaly.

A simple weekly workflow keeps the comparison honest:

- Pull Google generative impressions by page for a fixed seven-day window and note the top canonical URLs.
- Pull Bing citations, grounding queries, and intent labels for the same window and the same URL set where possible.
- Mark pages that rise in Google impressions but stay flat in Bing citations, and the reverse pattern.
- Attach the dominant Bing intents to each shared URL so content briefs reflect real question types.
- After publishing updates, repeat the paired export and record whether impression volume, citation count, or citation share moved first.

Cross-reference dates after content changes to observe whether impression or citation counts moved. An impression in Google signals that a URL entered the generative answer. A citation in Bing signals that the content was explicitly referenced. These are separate events, and only the Bing report currently distinguishes them with query and intent detail. When both move together after a rewrite, you have stronger evidence that the change improved generative eligibility rather than random fluctuation.

Even sites that receive little Bing traffic benefit from checking the Bing report for the query layer that Google withholds. Treat the two reports as complementary measurement surfaces. Google answers “How often did our URLs appear in AI Overviews and AI Mode?” Bing answers “Which pages were cited, on which prompts, with which intents, and how did share change?” Together they give a more complete measurement of presence in generative answers than either tool supplies on its own.

## Conclusion

Google's report delivers the first direct impression counts from AI Overviews and AI Mode, while Bing's report supplies citations, grounding queries, and intent data. Reviewing both sources produces a clearer picture of how content performs inside generative answers. Site owners who treat the reports as complementary rather than interchangeable avoid the gaps each tool leaves on its own.

Build a repeatable habit: configure the Search Console generative views carefully, read chart and table aggregation rules before acting, accept the impressions-only ceiling, then fill query and intent gaps with Bing’s AI Performance fields. Regular checks after content updates help track changes across the two ecosystems and keep decisions tied to what each report can actually prove today.