Assistant-Shaped Search: What Conversational Search Queries in GSC Tell Us

Conversational search queries, the long, question-shaped searches that read like sentences, now account for a third of the query rows in our own Google Search Console data. They rank substantially better than our short keywords, they earn impressions steadily, and in ninety days they produced zero clicks. That pattern is no malfunction. It's consistent with what search becomes when people ask AI surfaces full questions and the answer gets assembled without a visit, and there's a good chance your GSC report carries the same fingerprints.

Key takeaways:

• In our 90-day GSC sample, 6+ word queries were 34.2% of the non-anonymised query rows but only 13% of impressions
• Those long queries ranked at an average position of 27 against 42 for short keywords, and still earned no clicks
• AI Mode prompts and follow-ups flow into GSC as their own query rows; hidden fan-out subqueries are not exposed as query rows at all
• The response is editorial, not technical: answer the questions these queries ask, in headings, on pages built to be cited

What Are Conversational Search Queries?

Conversational search queries are searches phrased the way people talk: full questions, first-person requests, and multi-clause briefs rather than two-word keywords. "B2B SEO agency" is a keyword. "How can a b2b tech company optimize its content to be cited by llms like chatgpt" is a conversational query, and it's a real one from our report.

They arrive in Search Console like any other query, which is exactly why they're easy to misread. The instinct is to treat them as long-tail keywords with tiny volume. The data says they behave like something else entirely.

Three shapes dominate our sample. Question queries open with how, what, or why and made up 57.3% of our six-plus word tail. First-person briefs describe a situation and ask for help, complete with company size and role. And option-menu strings paste in choices no person would type, the clearest machine tell of the three.

What Does Our GSC Data Show?

We pulled 91 days of our own Search Console queries, 19 May to 17 August 2026, with the freshest day or two still subject to GSC's reporting lag: 877 query rows and 15,756 impressions for tenpointlabs.com, analysed by query length with a small script we kept for the next run. The pattern was stark enough to write about, and small enough to be honest about: this is one B2B site's window, offered as a worked example rather than an industry study. Run the same split on your own property before believing any of it.

One caveat sits underneath every percentage below. Google anonymises rare queries and can truncate additional rows, so 877 is the count of query rows Search Console exposed to us, not the count of every search that generated an impression. Read the shares as proportions of the visible export.

What the data shows:

• 6+ word queries made up 34.2% of the non-anonymised query rows in our GSC export, but just 13% of impressions
• Those queries ranked at an impression-weighted average position of 26.7, against 41.8 for queries of four words or fewer
• Their combined click-through rate across ninety days was exactly zero

The long tail gets strange when you read it. One 19-word query embeds a demographic menu, "i am a 25-34, 35-44, 45-54, or 55-64 year old marketing manager or sales manager", which reads like nothing a person would type into a search box. Another opens "i'm a content marketer at a b2b software company and need a tool that shows me which of our blog..." and runs to 21 words. These are briefs, not searches, and automated tools or agent-style sessions are the likeliest authors of the strangest of them.

Where Do These Queries Come From?

Start with what conversational queries aren't. AI systems expand one prompt into many hidden sub-queries, a mechanism called query fan-out; Google's AI Mode typically makes 5 to 11 such searches per query, over 95% of which have no recurring human search history, documented in Ahrefs' query fan-out explainer. Those hidden fan-out subqueries are not exposed as query rows in Search Console. GSC reports queries from the user-facing surfaces; a page read during fan-out earns no row of its own, a distinction Search Engine Roundtable sets out clearly.

So the rows come from nearer the surface. Prompts people type into AI Mode flow into Search Console as user queries, and Google documents that a follow-up in an AI Mode conversation is treated as a new query, with its own impression, position and click data. Google's June 2026 generative-AI performance reports notably carry no query dimension, one more reason the classic report's long tail is worth mining. Add long-phrasing humans on classic Search and automated third-party tools running prompt-shaped checks, and you have the realistic author list for a strange tail like ours.

Honesty requires a caveat, and it's a big one. The standard query table mixes classic Search and AI Mode together with no label separating them, so no individual row can be attributed to a surface with certainty. Conversational phrasing is a useful heuristic, not proof of provenance. What we can say is that our long tail carries machine fingerprints, option menus, persona preambles, briefing-document phrasing, and that its behaviour, better rankings with zero clicks, is consistent with assistant-mediated search without demonstrating it.

Part of the tail is also simply people. Voice search and chat-style searching have pushed humans toward longer phrasing for years, and Google's documentation on AI features confirms that appearing in AI Overviews and AI Mode runs on the same indexing and eligibility as classic Search, meaning the machine layer reads the same index your SEO built. The two streams mix in the same report, which is precisely why the impressions deserve attention rather than deletion: some of those sentences are your buyers talking to an assistant about the problem you solve.

Why Do They Earn Zero Clicks?

Here it's worth separating what we measured from what we think it means, because the two get conflated constantly in writing about AI search.

• Observed: our six-plus word queries ranked substantially better than our short keywords and drew zero clicks across ninety days.
• Plausible: some of those impressions are AI Mode or conversational-search sessions where the answer arrives assembled from pages nobody visits, and the visit your analytics would have counted gets absorbed before it happens.
• Not established: that AI answers caused all, or even most, of those zero clicks. At an impression-weighted average position of 26.7, zero clicks are entirely ordinary in classic Search too.

That's the tension in our table: we rank fifteen spots better on conversational queries than on keywords and see nothing for it in traffic. Some of that is simply page three. Some of it, on the rows that read like assistant turns, looks like something else, where being read and being visited have come apart. Position was built to predict clicks on a results page, and on assistant-shaped queries it does less of that work than it used to.

Either reading leads to the same conclusion: treating the impressions as waste gets it backwards. Question-shaped demand is reaching pages you own, and the questions are usable whether or not an AI surface produced the row.

What Should B2B Teams Do?

Four responses put this data to work, in rising order of effort:

1. Mine the queries as intent research. Filter your own GSC to queries of six-plus words and read them like interview transcripts. Ours contain complete buyer briefs, budgets of anxiety, and phrasing we now reuse in content. It's free voice-of-customer data arriving daily.
2. Answer the questions in question-shaped headings. In a study of 18,012 verified ChatGPT citations covered by Search Engine Land, 78.4% of citations tied to questions came from headings. The long queries in your report are literally telling you which headings to write.
3. Reframe impressions as visibility. A conversational-query impression with no click can be an AI-search signal, particularly when the phrasing resembles an AI Mode turn, but Search Console doesn't tell us which individual query rows came from AI Mode. Report the pattern as a visibility signal alongside classic traffic, not as a failing keyword.
4. Build pages worth citing. Fan-out queries feed answers that cite sources, so the durable move is becoming a source: clear claims, question-led structure, and the content signals that make AI tools trust and cite your brand. Bear in mind where assistants prefer to look: a 2025 University of Toronto preprint found the AI engines it studied favour earned, third-party authoritative sources over brand-owned pages, so citability work extends past your own domain. The production side sits in our GEO playbook for B2B companies.

Is This Good or Bad News?

For B2B teams the pattern is mostly good news wearing an unfamiliar costume. The queries prove demand exists and name it in the buyer's own sentences. They show your pages surfacing against question-shaped searches on the surfaces that now broker a share of B2B discovery through AI. What they end is the comfort of measuring everything in clicks.

The teams that adapt will read their long tail monthly, write toward it deliberately, and measure being-cited as seriously as being-visited. The teams that don't will keep reporting a traffic metric that measures a shrinking slice of how buyers meet them.

FAQs

A prediction to hold us to: within a couple of years, the query report B2B content teams watch most closely won't be the keywords humans typed, it will be the conversational turns users make inside AI-assisted Search. Ours is already a third of the way there. Go read your own long tail this week; the assistant-shaped future is sitting in it, ranked around position 27, waiting for someone to answer properly.