A Field Guide to Search Console's Generative AI Reports

Google has given publishers a dedicated view of a signal they previously had to infer: how often their URLs appear in supported generative AI features. The new Search Console reports separate these impressions from the familiar performance views, making it easier to inspect visibility in AI Overviews, AI Mode, and generative AI features in Discover. The first rule is about availability. As of July 30, 2026, Google is rolling the reports out to a subset of sites for testing; access is not universal . A missing report does not prove that a site has no AI visibility. Google says a property may lack access because it is not yet in the rollout or because it has not received enough eligible impressions. Treat the interface as an evolving measurement surface, not as a complete census of AI search. Know which report you are opening Search and Discover have dedicated generative AI reports. The Search report includes impressions from AI Overviews and AI Mode. Google says the supported feature list may change, and Search Labs experiments are excluded. The Discover report addresses generative experiences in Discover rather than Search. This separation matters when reporting to stakeholders. Do not add a Search total and a Discover total without labeling them as different environments. Search responds to an active query; Discover is proactively served and can be more variable. A clean monthly document should therefore keep a Search section and a Discover section, even if both roll up to a broader AI visibility discussion. Also distinguish these reports from Search Console's AI powered configuration tool. That tool helps configure filters in the conventional performance report. It is not the same product as the dedicated generative AI performance reports. Read an impression before reading a trend The reports measure the appearance of links to the property in supported generative AI features. They do not provide a transcript of the answer, the user's prompt, a citation quality score, or a list of ranking factors. An impression is evidence of visible inclusion under Google's counting rules, not proof that a user read, trusted, or acted on the linked material. In AI Mode, a follow up question is treated as a new query and its links can generate new performance data. In AI Overviews, the link must be scrolled or expanded into view to count as an impression. Those rules make the metric useful, but they also limit the stories it can support. Report what increased or decreased; do not convert impressions into unsupported claims about brand influence. The newest points can be preliminary and may change while data collection finishes. Google marks them with a dotted line. Leave those points out of consequential week over week conclusions until they settle. Work the dimensions in a fixed order Start with a stable date range long enough to reduce day to day noise. Compare it with the preceding equivalent period, then move through the available dimensions: 1. Dates: Find sustained changes rather than reacting to a single spike. Monthly or weekly granularity can reveal direction more clearly than a daily chart. 2. Pages: Sort for high and low impressions. The report generally credits the canonical URL, so confirm canonicalization before assuming a duplicate page is invisible. 3. Countries: Look for geographic concentration or a market that is moving differently from the property total. 4. Devices: In the Search report, compare desktop, tablet, and mobile. Device is not listed as a Discover generative report dimension in Google's launch description. Export the chart and table when a review needs a durable record. Remember that normal performance report constraints apply, including a 1,000 row table limit. A chart total may also differ from the table total because the chart can aggregate by property while a page table aggregates by URL. Turn page lists into editorial questions A high impression page is not automatically excellent, and a low impression page is not automatically broken. Use the report to select pages for investigation, then bring in evidence from other systems. For a page gaining generative AI impressions, ask whether its facts are current, its main answer remains easy to locate, its sources are still authoritative, and its images or examples still clarify the topic. Check analytics for visits and useful actions, but do not expect the Search Console report alone to establish causation. A rise may coincide with seasonality, news interest, interface changes, or changes elsewhere on the web. For a page losing impressions, check the same fundamentals before rewriting it. Review technical eligibility, indexing, canonical tags, freshness where freshness is genuinely relevant, and whether the page adds information beyond a generic summary. Record the date and scope of any substantial edit so later comparisons have context. For country or device differences, test the user experience that corresponds to the segment. A mobile heavy opportunity may justify examining layout, media, and load behavior. A country difference may expose language, availability, or local context gaps. These are investigation prompts, not conclusions supplied by the report. Build a monthly review sheet Use one row per page and keep the fields small enough to maintain: reporting period and comparison period; Search or Discover report; current and previous impressions; leading country and, for Search, device; relevant business outcome from analytics or conversion systems; factual, technical, or presentation issue found; action, owner, and review date. Preserve rows where the decision is "no change." That prevents a dashboard from becoming a machine for unnecessary rewrites. It also creates an audit trail showing which signals were considered and why editors acted or waited. When evaluating an edit, compare a reasonable period before and after it while noting other events. Search Console itself warns that changes in sentiment, competitors, and other conditions can complicate attribution. A page annotation and a written hypothesis are more honest than claiming an isolated content change caused the movement. Guard against five reporting errors Do not call the rollout universal. Do not treat an absent report as zero impressions. Do not present an impression as a visit. Do not assume the page table and property chart should always match. Do not optimize for the report by manufacturing thin variations of the same page. The practical value of these reports is prioritization. They can show where supported generative features are surfacing a site's URLs and how that visibility varies across time, pages, countries, and, in Search, devices. They cannot explain every selection or replace audience, conversion, and editorial quality evidence. Used monthly, with written hypotheses and restrained language, they provide a useful new lens without becoming a new vanity metric. Sources Introducing Search Generative AI performance reports in Search Console https://developers.google.com/search/blog/2026/06/gen ai performance reports Generative AI performance report Search https://support.google.com/webmasters/answer/16984139 What are impressions, position, and clicks? https://support.google.com/webmasters/answer/7042828 Performance report Discover https://support.google.com/webmasters/answer/9216516?hl=en Cover image credit Cover image by rawpixel , made available under CC0 1.0 . WIKIVISE cropped and converted the source image.