Sentence-length, context-loaded questions are appearing in ordinary Google Search, not just inside chat tools. We have two of them in a ten-row report. This changes which pages can realistically win, and it is visible in Search Console today if you look for it.
Our actual rows
Three-month view, Web search type, top ten by impressions, taken 17 September 2026. Unedited:
| Query | Words | Impressions | Position |
|---|---|---|---|
| does my business need geo | 5 | 153 | 68.7 |
| borvis | 1 | 56 | 5.1 |
| does an llms.txt file on the website do anything for ai visibility, or is that a myth? | 17 | 6 | 84.3 |
| autowithai | 1 | 4 | 81.0 |
| ai brand citations vs traditional backlinks | 6 | 4 | 86.0 |
| i'm a paid media director. is there a way to measure whether our earned media is what ai models cite about our candidate? | 23 | 4 | 87.3 |
| ai brand citations vs traditional backlinks seo | 7 | 3 | 87.0 |
| ai citations vs backlinks | 4 | 3 | 87.0 |
| llc service - geo test | 5 | 2 | 58.0 |
| is geo worth it for small businesses | 7 | 2 | 75.5 |
Mean length across these ten rows is 7.6 words. Two rows are twelve words or longer. Total impressions across the visible ten: 237.
What the average looks like
For context, reported United States Google query length sat flat at roughly 3.33 to 3.36 words from June 2024 until AI Mode launched in May 2025, then climbed to over 3.51 words by May 2026.
Google Ads data shows the same shift more sharply. Three-to-four word queries went from 33% of impression share in January 2025 to 48% in August 2026, while one-to-two word queries fell from 42% to 24%. On conversions the move is larger still, with one-to-two word queries dropping from 76% to 41%.
Separately, SOCi reported in its 2026 Local Visibility Index that LLM queries average 23 words, almost six times traditional search.
Our longest query is exactly 23 words, matching that reported LLM average. With one query that is a coincidence and nothing more. We mention it because it is the kind of detail that gets turned into a finding by someone else, and it is not one.
The shape of the long ones
Length is not the interesting part. Structure is. Look at what the 23-word query actually does:
- States a role. "i'm a paid media director." Nobody types their job title into a search box out of habit. They type it when they expect the answer to be tailored to them.
- Supplies context before asking. "our candidate" tells you the sector without naming it.
- Asks a compound question. Not "AI citations" but whether a specific input is producing a specific output.
- Uses natural punctuation. A full stop mid-query, and a question mark at the end. Keyword-shaped queries do not carry punctuation.
The 17-word one has the same signature: "does an llms.txt file on the website do anything for ai visibility, or is that a myth?" It offers its own counter-position. That is conversational phrasing, and it is arriving in Search rather than only in a chat window.
Why this is happening
The plausible mechanism is habit transfer. People who have spent a year describing their situation to a chat interface start phrasing questions that way everywhere, including in a search box that historically rewarded three keywords.
Google has also built surfaces designed for exactly this, describing AI Overviews and AI Mode as experiences for longer and more complex questions. Behaviour and interface are reinforcing each other.
We are not claiming these two queries came from an AI surface. The Web performance report does not tell you that, and the separate generative report has no query dimension at all, so nobody can make that link from Search Console today. What we can say is that queries shaped like chat prompts are landing in ordinary Search results.
What it means for pages
- These queries are winnable in a way head terms are not. Nobody has optimised for a 23-word question about measuring earned media citation for a candidate. There is no incumbent. For a site with no domain authority, that is the only kind of query currently within reach.
- The buyer is more qualified, not less. Someone stating their role and situation is further into a decision than someone typing two words. Four impressions from that person are worth more than four hundred from a definition query.
- A page has to answer the whole question. Compound questions need compound answers. A page covering only half of "is there a way to measure whether X is what Y cites" is not a match.
- Keyword tools will not show you these. A 23-word query has no meaningful reported volume. It exists in your own Search Console and nowhere in a research tool, which makes your own data the only place to find them.
Our response to the two long queries above was to write the pages they deserved, rather than a page about long-tail queries in general. One of those is now published.
The unflattering part
Every row above has zero clicks. Average positions run from 68 to 87 on the non-brand queries. That is page seven to page nine.
We are publishing this because the honest version is more useful than a screenshot of a rising line. What these numbers show is a site that is correctly matched to its topic and not yet competitive on it. Impressions at position 80 mean Google understands what the page is about and is not yet prepared to show it to anyone.
The one positive signal is movement on the largest query, which went from 120 impressions at position 69.9 to 153 at 68.7 over the period in which the page was substantially rewritten. That is a small directional change on a single query, with no control, and we would not present it as proof of anything.
Because anyone can show a chart going up. A dated record of a site at position 80 with no clicks, published by the people running it, is checkable later against whatever happens next. That is a harder thing to fake than a case study.
Limits
- This is one small site with 237 visible impressions across ten rows. It is an observation, not a study, and nothing here generalises on its own.
- The ten rows are the top ten of twenty by impressions, so the sample is selected rather than random. Our 7.6-word mean is not comparable to a population average.
- The query-length figures for Google and Google Ads come from third-party analyses of advertiser data, reported in September 2026. We were not able to reach the underlying dataset directly.
- The SOCi 23-word figure is from its 2026 Local Visibility Index as reported in February 2026. The methodology and sample size for that specific figure are not disclosed in the reporting we could access.
- Nothing here establishes that long queries originate from AI surfaces. Search Console does not expose that, and we are describing shape rather than source. We later found reasons to think the 23-word query is an AI tracking tool rather than a person.
If you want to know whether this is happening to you, sort your own query report by length rather than volume. The interesting rows are usually at the bottom.