Google Search Console has been including AI Mode queries in its standard performance data for a while, with no built-in way to separate them from ordinary searches. Google has since added an official AI report and rolled it out to all sites worldwide by August 31, 2026, but it shows impressions only, by page, country and date, with no click data and no query text. An AI Mode conversation and a typed search query are fundamentally different kinds of visitor intent, so if you want to see which actual queries look like AI Mode prompts, and what they earn in clicks, the filtering workaround below is still useful.
Table of Contents
Why This Data Is Genuinely Hard to Separate
Google confirmed through John Mueller that AI Mode prompts do show up as queries in Search Console, but the underlying query strings are almost exclusively anonymized before they reach the reporting interface, there’s no clean, official tag distinguishing “this was a conversational AI Mode prompt” from “this was a normal typed search” in the raw data. That’s why site owners and SEO practitioners have had to build their own workarounds rather than wait for an official solution, and it’s also why any approach to isolating this traffic is inherently an approximation, not a precise, guaranteed-accurate filter.
The Practical Workaround: Filtering by Query Shape
Since AI Mode queries tend to look conversational, longer, more natural-language, often phrased as a full question or instruction, rather than the short keyword fragments typical of a traditional search, filtering by query characteristics gets you a reasonable approximation. A straightforward version: filter Search Console’s Performance report down to queries of 10 or more words, which disproportionately captures conversational AI Mode-style prompts compared to typical short-tail search terms.
A more refined version uses a custom regex filter on the Query field, matching against conversational markers, prompt verbs like “write,” “draft,” “generate,” “summarize,” “explain,” “act as,” greetings, acknowledgements like “yes” or “sounds good,” and follow-up phrasing like “show me more” or “any other options,” patterns that show up in a genuine back-and-forth AI conversation but rarely in a typed search box.
What to Actually Do With This Once You’ve Filtered It
Once you’ve isolated a reasonable approximation of AI Mode-driven queries, the useful comparison is against your standard organic queries: are the topics and intent behind AI Mode traffic meaningfully different from what drives your typical search traffic, and is your content actually structured in a way that answers those more conversational, specific questions directly, rather than requiring an AI system to piece an answer together from content written for a different kind of query entirely. A page written to rank for a short keyword phrase and a page written to directly answer a specific, detailed conversational question aren’t always the same page, even when they’re nominally about the same topic.
The Official Search Console AI Report: What It Does and Doesn’t Show
Google announced the AI-specific report for Search Console on its Search Central blog in June 2026, began with a subset of properties starting in the UK, and finished rolling it out to every property on August 31. It reports impressions from AI Overviews, AI Mode and generative AI features in Discover, but not clicks, and not the queries behind them. A separate property-level Search generative AI control lets a site exclude its content from those surfaces, which Google says is not a ranking signal and has no impact on AI training.
Check the new report first, since it tells you whether you appear in AI features at all, then use the query-shape filtering above to get a rough picture of the prompts and clicks the official report leaves out.
Step by Step: Setting Up the Filter
In Search Console, open Performance, then Search results, and set the date range to at least the last three months so there is enough data to see a pattern. Click Add filter, choose Queries, then choose Custom (regex) and leave Matches regex selected, which is the default. Paste the pattern from the Practical Ecommerce article, which matches queries of ten or more words: ([^" "]*\s){10,}?, and click Apply. Switch on the comparison view to set the period against the one before it, and export the query list if you want to read through the actual wording.
Google’s help page warns that filtering by query or page can affect the report totals because of data truncation and the omission of anonymized queries, so use the filtered numbers for direction rather than as a precise count.
How to Read What You Find
- Treat the numbers as a floor, not a total. Search Console hides some rare queries for privacy, so the list is incomplete.
- Long queries aren’t all AI Mode. Voice searches and detailed typed questions also run long, so the filter is an approximation, as noted above.
- Look at topics rather than individual queries. Group them by theme and see which pages they land on.
- Compare trends over several weeks. A single week of low impression counts says very little.
- Check which pages these queries reach. If the same pages also earn short-keyword traffic, you can use the long-query wording to improve them.
Turning the Findings Into Content Decisions
The point of this exercise isn’t the report itself but what it tells you about wording. Long, conversational queries show the exact questions people ask, often with their constraints included, such as budget, platform or deadline. Use them to add a direct answer near the top of the matching page, to write a short section for a question you haven’t covered, or to combine thin pages that each answer half of it. Don’t rewrite whole pages to chase a few dozen impressions, and don’t draw conclusions about AI Mode specifically, since the filter can’t confirm where a query came from.
Pair this with the official AI report, which tells you whether a page appeared in AI features at all, and you have a reasonable picture from two imperfect sources.
Getting Started
Try the simple version first, filtering Performance data down to queries over 10 words, before building a more elaborate regex filter, since even that basic cut usually reveals whether this is a meaningful enough slice of your traffic to warrant deeper analysis. For a lot of smaller sites, the honest answer will be “not yet significant,” which is itself useful information, it means the bigger optimization opportunity is still elsewhere, at least for now.
Related reading: Your Search Console Is Quietly Logging AI Chat Fragments and Google Search Console Can Now Show You Image-Based Search Traffic.