Queries that look like AI fan-out made up 6.1% of 27 sites' visible Google impressions, ranked higher than keywords and drew 3 clicks

Findings

1. Search Console showed the query behind 358,745 of the 27 sites' 641,159 impressions (56.0%), from 8.3% to 88.3% per site, so every figure below covers visible impressions only.
2. Long conversational queries made up 6.1% of visible impressions (21,977), short question queries 5.0% (17,824) and keyword queries 88.9% (318,944).
3. 24 of the 27 sites had at least one long conversational impression, the median site drew 2.5% of its visible impressions from them, and 5 sites drew 20% or more.
4. Long conversational queries averaged position 14.9 against 33.1 for keyword queries (n = 21,977 and 318,944 impressions), and they ranked higher on 13 of the 16 sites with at least 50 impressions in each class.
5. They produced 3 clicks from 21,977 impressions, and none from 10,326 impressions on the free tool sites, where keyword queries had a click-through rate of 2.4% (n = 145,348).
6. 92.6% of their impressions came from desktop searches, against 77.7% for keyword queries (n = 21,977 and 318,944), and their desktop share was higher on 14 of 16 sites.
7. Their share of visible impressions rose from 4.8% in the first four full weeks (n = 115,112) to 7.4% in the last four (n = 116,958), and from 4.8% to 7.2% on the 21 sites with visible impressions in every full week.

What we measured

Strandline Research No. 02 followed one B2B site that published 34 new pages in September 2026. Long conversational queries made up 21.0% of the new pages' visible Google impressions (669 of 3,181), at an average position of 6.9 against 51.0 for keyword queries. Those queries had the shape of searches run by AI assistants while they write an answer, or by AI rank trackers testing a prompt: long, phrased as a question, and sometimes carrying the prompt text itself.

Google describes the same technique for its own AI features. It says AI Overviews and AI Mode may use a "query fan-out" technique, "issuing multiple related searches across subtopics and data sources" to develop a response, and that sites appearing in these features "are included in the overall search traffic in Search Console" (Google Search Central).

This study measures how common such queries are across 27 small and mid-size websites in one portfolio, and how they rank against keyword queries. Search Console does not say who ran a search, so the classes describe the shape of a query. A person who types a full question into Google lands in the same class as a tool that sends one. Each query gets the first class whose rule it matches:

1. Long conversational: 9 or more words, ending in a question mark, or containing prompt text such as "answer the following".
2. Question: starting with who, what, where, when, why, how, which, can, does, is, are or should, with fewer than 9 words.
3. Keyword: everything else.

Search Console shows the query for 56.0% of impressions

Search Console leaves rare queries out of its query tables. Google says it "might not track some queries that are made a very small number of times or those that contain personal or sensitive information" (Search Console Help). On small sites this hides a large part of the data. Across the 27 sites, rows with a query carried 358,745 of 641,159 impressions (56.0%) and 4,102 of 8,246 clicks (49.7%). The median site showed the query for 57.3% of its impressions, the lowest for 8.3% and the highest for 88.3% (n = 27 sites). The share for every site is in sites.csv.

The hidden queries are the rare ones, and a long query is more likely to be rare than a short one. The share of long conversational queries among all impressions may therefore be higher than their share of the visible ones, and Search Console offers no way to check.

Long conversational queries made up 6.1% of visible impressions

The 27 sites had 23,315 visible queries, counted once per site: 1,735 long conversational, 1,621 question and 19,959 keyword queries. Of the 1,735 long conversational queries, 1,564 had 9 or more words and 171 were shorter questions ending in a question mark. Fourteen carried prompt text, and every one of them also had 9 or more words or a question mark, so the prompt-text rule added no query of its own. Thirteen of the 14 open with one of two prompts written for an AI model and reached Google whole, with the question after them. Six begin "answer the following user question naturally and concisely. if relevant, include specific tools/companies and explain why. question:" and averaged position 2.0 (30 impressions). Seven begin "context: location: [country] (not for language). do not include location references in your response. question:" and averaged position 7.0 (41 impressions). Other long conversational queries read like a buyer's request, such as "what are the best free [product type] tools available and how do they compare to each other?", with the product type left out here.

Single-topic guide sites15.3%

Free online tools and templates6.5%

All 27 sites6.1%

B2B services4.5%

Directories and databases1.1%

Content sites and blogs1.0%

Source: Strandline Research analysis of Google Search Console data for 27 sites, 24 June to 21 September 2026, web search, final data. Base: impressions whose query Search Console shows (n = 358,745). Sites and visible impressions per type are in site-types.csv.

How common these queries are depends on the site. Single-topic guide sites, each built around one search question, drew 15.3% of their visible impressions from long conversational queries and another 20.4% from short questions (n = 56,797). Free online tools and templates drew 6.5% (n = 157,717) and B2B services sites 4.5% (n = 41,462). Directories and databases drew 1.1% (n = 48,597), and content sites and blogs 1.0% (n = 54,172).

The median site drew 2.5% of its visible impressions from long conversational queries, and the highest 36.4% (n = 27 sites). A few sites and queries carry most of the volume. The site with the most drew 8,682 of the 21,977 long conversational impressions (39.5%), and the three with the most drew 14,696 (66.9%). The 10 long conversational queries with the most impressions drew 7,776 of them (35.4%), while 862 of the 1,735 queries had a single impression in 90 days.

Long conversational queries rank higher than keyword queries

Long conversational queries averaged position 14.9 (n = 21,977 impressions), against 33.1 for keyword queries (n = 318,944) and 41.3 for short questions (n = 17,824). A lower number is closer to the top of the results. 78.1% of long conversational impressions came from queries averaging position 10 or better, against 41.1% of keyword impressions and 4.4% of question impressions.

Pooled averages mix sites that rank very differently, so we also compared the classes within each site. Of the 16 sites with at least 50 visible impressions in both classes, long conversational queries averaged a higher position than keyword queries on 13, and the median gap was 15.1 positions. The count holds at other cut-offs: 13 of 17 sites at 30 impressions and 12 of 15 at 100. Short questions went the other way and ranked below keyword queries on 11 of the 19 sites with 50 impressions in both classes.

All 27 sites, long queries14.9

All 27 sites, keywords33.1

Free tools, long queries7.9

Free tools, keywords17.4

Guide sites, long queries13.4

Guide sites, keywords24.9

B2B services, long queries13.9

B2B services, keywords31.5

Content sites, long queries72.2

Content sites, keywords71.6

Directories, long queries119.5

Directories, keywords47.5

Source: Strandline Research analysis of Google Search Console data for 27 sites, 24 June to 21 September 2026. Long queries are long conversational queries. Impression-weighted mean of each query's average position. The site type and impressions behind each bar are in chart-position-by-type.csv.

Two site types break the pattern. On the content sites and blogs, the two classes ranked about the same: 72.2 for 532 long conversational impressions and 71.6 for 50,492 keyword impressions. On the directories and databases, 533 long conversational impressions averaged position 119.5 against 47.5 for keyword queries, and 476 of them came in the week of 29 June. Most were variations of one question in which only a city, and often an industry, changed, which points to one batch of generated searches. On the site that drew most of them, S20 in sites.csv, long conversational queries averaged position 123.6.

Two explanations fit the gap, and Search Console cannot separate them. A long, specific query has fewer pages competing for it, so a site that covers the topic can rank near the top. And an impression counts only when a result appears on a page of results that was loaded: a tool that reads only the first results can record impressions near the top and nowhere else.

The word count shows where clicks stop.

Clicks fall away before the 9-word line. Queries of 7 words had a click-through rate of 0.1% (n = 8,087), and queries of 8 words produced 1 click from 21,660 impressions at an average position of 9.7. Three 8-word queries drew 15,327 of those impressions (70.8%). All three were strings of keywords with no question form, and the 9-word rule counts them as keyword queries. Searches written by software do not all read like questions, and a word-count rule misses the shorter ones.

They almost never produce a click

Long conversational queries produced 3 clicks from 21,977 impressions in 90 days. Keyword queries produced 4,092 clicks from 318,944 impressions (1.3%), and short questions 7 clicks from 17,824 impressions.

Part of that gap comes from the site mix. Long conversational queries take their largest share of visible impressions on the single-topic guide sites, where keyword queries also drew almost no clicks: 9 from 36,491 impressions. The comparison holds within the site types that do get clicks. On the free online tools and templates, keyword queries had a click-through rate of 2.4% (3,541 clicks from 145,348 impressions) and long conversational queries produced no clicks from 10,326 impressions. On the B2B services sites, the figures were 1.0% (393 from 38,830) against no clicks from 1,868.

Long conversational impressions came from desktop searches more often than keyword impressions in every site type. Within sites, their desktop share was higher on 14 of the 16 sites with 50 impressions in both classes, by a median of 10.9 percentage points. High desktop shares and almost no clicks fit software running searches better than people reading results, though Search Console records neither who searched nor why.

One free tool site went the other way, and shows what people asking full questions look like. On S07 in sites.csv, long conversational queries were common, most of them came from mobile devices, and a smaller share of them than of the site's keyword impressions came from desktop. That fits people typing whole questions about the tool's topic on their phones.

The share rose from July to September

• Long conversational queries
• Short question queries

Chart data

week_start	Long conversational queries	Short question queries
2026-06-22	3.3%	4.8%
2026-06-29	4.8%	4.8%
2026-07-06	3.3%	4.7%
2026-07-13	5.1%	4.4%
2026-07-20	6.2%	4.0%
2026-07-27	6.0%	4.9%
2026-08-03	7.7%	5.0%
2026-08-10	6.6%	5.0%
2026-08-17	6.2%	6.3%
2026-08-24	4.2%	6.6%
2026-08-31	10.4%	5.4%
2026-09-07	7.1%	4.8%
2026-09-14	7.4%	4.1%

Source: Strandline Research analysis of Google Search Console data for 27 sites. Weeks run Monday to Sunday and are labeled by their Monday. The first week covers 24 to 28 June only, and the last full week ends 20 September 2026. Visible impressions per week, from 17,837 to 32,462, are in weekly.csv.

Long conversational queries made up 4.8% of visible impressions in the first four full weeks, 29 June to 26 July (5,533 of 115,112), and 7.4% in the last four, 24 August to 20 September (8,602 of 116,958). Short questions stayed between 4.0% and 6.6% a week (n = 17,837 to 32,462 visible impressions a week). The number of sites with any long conversational impression ranged from 15 to 21 a week. Long conversational queries averaged a higher position than keyword queries in 12 of the 13 weeks in the chart. The exception was the week of 29 June, when the directories' batch pulled their average to 54.6.

The rise does not depend on the largest sites or on the six sites that first appeared in search during the window. Leaving out the three sites with the most long conversational impressions, the share went from 3.5% to 5.0% (n = 58,115 and 64,720). On the 21 sites with visible impressions in every full week, a group that leaves out the six newer sites, it went from 4.8% to 7.2% (n = 113,542 and 102,862).

Single weeks jump because a few queries carry large volumes. The week of 31 August reached 10.4% (3,327 of 31,855 visible impressions), and one query drew 1,768 of that week's long conversational impressions. Single-topic guide sites drew 2,344 of them.

Method

Source. Google Search Console performance data for 27 websites, pulled through the Search Console API (searchanalytics.query) on 23 September 2026 for 24 June to 21 September 2026. That is 90 days, ending on the last date with final data. Search type web, final data only.

Sample. The Search Console account lists 46 properties and has data access to 43. One of the 43 belongs to another company and is left out before any analysis. Of the other 42, the 27 with at least 1,000 impressions in the 90 days are included, and the 15 with fewer are left out. The included sites had 641,159 impressions and 8,246 clicks. Six of them first appeared in search results after 24 June, the last on 18 July, and count from that date. Site types:

• Free online tools and templates (7): sites built around one free tool, such as a calculator, a diagram maker or a document template.
• Single-topic guide sites (11): sites built to answer one search question in depth.
• Directories and databases (4): searchable directories, databases and a job board.
• B2B services (2): company sites for services sold to businesses.
• Content sites and blogs (3): article sites and a personal site.

Sites are not named. The per-site table labels them S01 to S27 and gives total impressions only as a band.

Pulls. For each site: daily totals (dimension date), 90-day query rows (query), daily query rows (date and query) and query rows by device (query and device), 25,000 rows per request, paged to the end. Google aggregates rows grouped by query, date or device by property, so query rows and daily totals count impressions the same way. The daily and device query rows add up to the same 358,745 impressions as the 90-day query rows. The full parameters are in method.txt.

Classes. Queries are lower case as Search Console returns them, and words are split on spaces.

1. Long conversational: 9 or more words, a closing "?", or one of 16 prompt phrases, among them "answer the following", "user question" and "question:". method.txt lists all 16.
2. Question: not long conversational, and the first word is who, what, where, when, why, how, which, can, does, is, are or should.
3. Keyword: everything else.

Metrics. Impressions and clicks are sums, and CTR is clicks divided by impressions. Average position is the impression-weighted mean of each query row's average position, which Search Console defines as the average position of the topmost result from the site. "At position 10 or better" counts impressions from query rows whose average position is 10.0 or lower. Queries are counted once per site. Weeks run Monday to Sunday and are labeled by their Monday. The weekly chart covers the 12 full weeks from 29 June to 20 September and the partial week of 24 to 28 June, and 21 September counts only in the 90-day figures. The four-week comparisons use full weeks only. Comparisons within a site use sites with at least 50 visible impressions in both classes, with 30 and 100 as checks.

Limits.

• Visible rows only. Rare queries are hidden, and long queries are more often rare, so the long conversational share of all impressions cannot be measured.
• Query shape only. Search Console does not say who or what ran a search. People asking full questions and AI tools that search Google fall in the same class, and Google counts AI Overviews and AI Mode in the same totals as other results.
• The rules are simple so that anyone can reproduce them. Keyword strings written by software can have fewer than 9 words (the 8-word row of word-count.csv), and short questions ending in a question mark count as long conversational.
• 27 sites from one portfolio, most of them small. One site supplied 39.5% of the 21,977 long conversational impressions and three supplied 66.9%.
• Average positions are averages of Search Console's own averages, and a position is recorded only as deep as a searcher loads results.
• The site types are broad groups of 2 to 11 sites.

Data

Every table and chart in this report comes from the CSV files listed below, and each file's description says what one row holds. Counts are plain integers, and percentages and positions are plain numbers rounded to one decimal. Sites appear only as types or under anonymous labels (S01 to S27), and the files contain no query text. The pull parameters and classification rules are in method.txt.

• Dataset totals CSV, 1 rows, 0.7 KBOne row: properties listed and with data access, the property left out because it belongs to another company, portfolio properties with data access, sites included and excluded, date range, total and visible impressions and clicks with their shares, queries, the lowest, median and highest share visible per site, quartiles of the long conversational share per site, sites with long conversational impressions, and sites that first appeared after the start date.
• Query classes CSV, 9 rows, 0.9 KBOne row per class and subgroup, plus all visible queries: queries, impressions, share of visible impressions, clicks, CTR, average position and share of impressions at position 10 or better.
• Query classes by site type CSV, 6 rows, 1.1 KBOne row per site type plus all sites: sites, total and visible impressions, share visible, and for each class the impressions, share of visible impressions, clicks, CTR and average position, with the number of sites that had long conversational impressions.
• Sites, anonymized CSV, 27 rows, 2.2 KBOne row per site under a label (S01 to S27): site type, band of total impressions, share visible, long conversational and question shares of visible impressions, and the average position and desktop share of long conversational and keyword queries where both classes have at least 50 impressions.
• Sites by long conversational share CSV, 6 rows, 0.2 KBNumber of sites in each band of long conversational share of visible impressions.
• Comparisons within each site CSV, 9 rows, 0.7 KBFor sites with at least 30, 50 or 100 impressions in both classes: sites compared, sites where the first class ranks higher (average position) or has the higher desktop share, sites where keyword queries do, and the median gap (keyword position minus class position, or class desktop share minus keyword desktop share).
• Queries by number of words CSV, 11 rows, 0.7 KBOne row per word count: queries, impressions, share of visible impressions, clicks, CTR, average position, share of impressions at position 10 or better, and the impressions and share of the 3 queries with the most impressions.
• Device split by class CSV, 18 rows, 1.0 KBOne row per site group (all sites and each site type) and class: impressions and the desktop, mobile and tablet shares.
• Weekly shares CSV, 13 rows, 1.1 KBOne row per week, Monday to Sunday, labeled by its Monday, from the partial week of 22 June (24 to 28 June in the window) to the week of 14 September: days in the window, sites with long conversational impressions, visible impressions, long conversational and question impressions and shares, keyword share, average positions of long conversational and keyword queries, long conversational clicks, and the impressions of the week's top long conversational query.
• Weekly long conversational impressions by site type CSV, 13 rows, 0.6 KBOne row per week, as in weekly.csv: days in the window and long conversational impressions for each site type and for all sites.
• First and last four weeks CSV, 6 rows, 0.8 KBVisible and long conversational impressions and the long conversational share in the first and last four full weeks, for all 27 sites, for the 24 sites left after removing the 3 with the most long conversational impressions, and for the 21 sites with visible impressions in all 12 full weeks.
• Concentration of long conversational impressions CSV, 6 rows, 0.4 KBLong conversational impressions and their share for the top 10 queries, the top 1% of queries, queries with a single impression, and the one and three sites with the most.
• Prompt text in queries CSV, 3 rows, 0.4 KBQueries that contain a prompt phrase, grouped by the prompt they open with (generic text only, the questions left out): queries, impressions, clicks and average position.
• Average position by class and site type CSV, 12 rows, 0.9 KBData for Exhibit 2: average position of long conversational and keyword queries for all sites and each site type, with the chart label, site group, class and the impressions behind each figure.