What Are Prompt-Like Queries in Search Console?

A query string that reads like a sentence spoken to a chatbot, not typed into a search box, is a prompt-like query. Search Console records them as plain text strings in the Performance report, the same as any other search term.

Both sit in the same Queries column. Only the second one gives away where it came from.

Recognizing and Filtering Conversational Search Prompts

Google Search Console logs every search string that triggered an impression or click on a property, whether that string came from a person typing three words into the search box or an AI assistant passing along a full question on a user’s behalf. A prompt-like query is one of the second kind: long, grammatically complete, often first-person, and structured like a request rather than a label.

This isn’t theoretical. A Search Engine Land analysis of leaked data found that searches originating in ChatGPT were showing up inside ordinary Search Console performance reports, and some of those strings contained personally identifiable information the person never meant to expose. That’s a strange thing to sit with: a tool built to show you how people find your site was quietly showing you how people talk to someone else’s chatbot, and that chatbot happened to reference your pages.

The practical fallout is that anyone exporting queries and eyeballing the top 50 rows will likely miss these strings entirely, because they sort low by volume and look like noise. The fix Search Engine Land recommends is mechanical, not conceptual: write a regular expression that filters the Queries field down to entries of 10 or more words.

That threshold isolates the conversational tail without requiring any guesswork about which rows are “real” searches and which are AI-relayed ones. There’s no official word count that defines a prompt versus a query; the 10-word regex approach is a filter convention, not a Google-defined boundary.

Conversational prompts mirror extreme long-tail search behavior

The volume math here should calm some nerves and raise others. Ahrefs reports that keywords getting fewer than 10 searches per month make up almost 93% of its U.S. keyword database, and separately notes that roughly 15% of daily Google searches have never been typed before in that exact form. Most of what shows up in a Performance report was always going to be low-volume and one-off. Prompt-like queries just push that pattern further out.

The option this opens up is treating these strings as directional evidence of intent rather than individual keyword targets. Nobody should build a content calendar around ranking for one 22-word question that appeared twice last month. But fifty variations of that question, all asking some version of the same thing, tell you what the underlying topic cluster needs to cover. Conversational AI traffic didn’t invent the long tail. It stretched it.

The judgment call most teams get wrong: they either ignore these rows completely because no single one hits a meaningful volume, or they panic and try to write a page targeting each exact phrasing. Neither works. Group them by theme, write to the theme.

Search Console omits anonymized queries from filtered reports

Filter by query, and the totals you see won’t always match the totals on the unfiltered chart. Google’s own documentation on the Performance report explains that table data aggregates differently depending on what you group by: data grouped by Queries, Countries, Devices, or Dates aggregates by property, while data grouped by Pages or Search appearance aggregates by page. That distinction alone accounts for some of the gap between chart and table.

A separate factor matters more for prompt-hunting specifically. Google’s help documentation on filtering performance data states plainly that filtering by query or URL can affect report totals because of data truncation and the omission of anonymized queries. Some of the rarest, most specific prompt-like strings, the ones most likely to contain a full sentence or a named detail, are exactly the kind Google anonymizes before they ever reach your export.

So the regex filter for 10+ word queries will always undercount. It can’t show you every conversational search that touched your site, only the subset Google chose not to suppress. Treat exported totals as a sample of prompt behavior, not a census of it.

Dedicated generative AI reports track visibility across AI surfaces

Up to now, every prompt-like string sat mixed into the same undifferentiated Queries list as every other search term, which is why the regex workaround existed at all. Google has since built a more direct path. New Search Generative AI performance reports give a dedicated view of impressions inside generative features on Search, including AI Overviews and AI Mode, plus generative AI features in Discover. That data still feeds into the main performance report too, but now there’s a separate lens built specifically for it.

Paired with that, a branded queries filter inside the standard Performance report now splits query data into branded and non-branded views automatically, without a custom regex.

What changes for the reader is where to start. Instead of building a 10-word regex from scratch every time, check the dedicated generative AI view first to see what’s already separated out, then layer the branded/non-branded split on top to see whether AI-surfaced traffic skews toward people who already knew your name or people encountering it for the first time.

The regex method from Search Engine Land still has a place for digging into the raw query text itself. Once you’ve narrowed a report with the regex filter guidance from Search Console’s own help pages, exporting and reviewing the actual phrasing is still the only way to read what people actually asked, not just how many impressions resulted.

Reading what an SEO strategist does with a Search Console performance report once these numbers are visible matters more than the export itself. Numbers without a next action are just a spreadsheet.

What it is not

A prompt-like query isn’t a new metric, a new report type Google invented for this purpose, or proof that a page is ranking inside an AI Overview. It’s an observation about the shape of a text string sitting in a column that already existed.

It’s also not the same thing as a branded or non-branded query, though the two can overlap; a long conversational prompt can mention a brand name or leave it out entirely. And it isn’t automatically evidence of bot traffic or scraping. Most of these strings represent a real person who asked a chatbot a full question and got your page referenced in the answer.

When it matters

Anyone running content strategy off Search Console exports should care, because treating a 20-word conversational string the same as a two-word head term means misjudging both its intent and its realistic ceiling for rank tracking. Teams managing FAQ content, comparison pages, or anything likely to get summarized by an AI assistant have the most to gain from separating this signal out early, before it skews a content brief toward chasing phrasing nobody will type verbatim twice.

Frequently Asked Questions

How do I filter Search Console for prompt-like queries?

Open the Performance report, go to Search Results, add a query filter, and switch it to custom regex. From there, apply a pattern that isolates queries of 10 or more words, which is the threshold Search Engine Land uses to separate conversational strings from typical keyword searches. There’s no official word-count rule from Google defining this boundary, so the 10-word cutoff is a practical convention, not a platform standard.

Are prompt-like queries the same as AI Overview traffic?

No. A prompt-like query describes the shape of the search string itself: long, conversational, question-like. AI Overview visibility is a separate thing Google now tracks through its dedicated generative AI performance reports, which show impressions specifically within AI Overviews, AI Mode, and Discover’s generative features. A query can be prompt-like without ever touching an AI Overview, and vice versa.

Can I target prompt-like queries with individual content pages?

Not one by one. Any single 20-plus word phrasing will have too little volume to justify a dedicated page. Group recurring prompt-like queries by theme and build content that answers the underlying question across its likely phrasings, rather than chasing the exact string that showed up once.

Does a high volume of prompt-like queries mean my content strategy is outdated?

It depends on whether the queries cluster around topics you already cover. If they repeat themes absent from your content, that’s a gap worth closing. If they’re variations on questions you already answer, it means AI assistants are already finding and using your pages, which is a different problem than a gap.