Graphic · AI Search · 2025–2026

ChatGPT Query Fan-Out Statistics (2026)

The Fan-Out Phrasebook: what ChatGPT actually types when it searches. Only 1 in 7 responses contained fan-out queries, and in one month those queries leaned harder on "official", "price" and "reviews", while citations moved from Reddit and Wikipedia to the sources themselves.

Fig. 1

The Fan-Out Phrasebook

What words ChatGPT puts in its fan-out queries: share of fan-out queries containing each word, August vs. September 2026. Ring: August. Dot: September. Rows run from the biggest gain to the biggest drop.

August 2026 September 2026 Gap (September 2026 − August 2026)
05%10%15%20%25%Gap↑ Used more in September↓ Used less in September202617.6%22.9%+5.3 ptsofficial10.1%13.1%+3 ptsprice6%8.9%+2.9 ptsreviews2.7%3.6%+0.9 ptssite4%4.8%+0.8 ptswhat2.3%3%+0.7 ptsnew1.4%2.1%+0.7 ptshow3.3%3.6%+0.3 ptsowner1.5%1.8%+0.3 ptsbest5%5.3%+0.3 ptscompany2.2%2.5%+0.3 ptsingredients1.6%1.9%+0.3 ptsbrand1.8%1.9%+0.1 ptsreview2.2%2.3%+0.1 ptscost2%2.2%−0.2 ptsmeaning2.3%2.9%−0.6 pts

The 16 non-place words in Ahrefs' top 20 for September 2026, from 15 million fan-out queries in free, logged-out GPT-5.6 Luna responses. Left out: three country names (india, australia, philippines) and "september", which Ahrefs marks with an unexplained asterisk. Ahrefs describes the "official" figure as a share of English fan-out queries. Gaps here are September minus August as printed, so a few differ by 0.1 pts from Ahrefs' change column (e.g. price: +2.84 pts from the unrounded 6.04% and 8.88%).

ChatGPT responses with fan-out queries
14.4%
About 1 in 7 of 66.6M free, logged-out responses, Aug–Sep 2026
Fan-out queries per response that has them
1.6
15.0M fan-out queries across 9.6M responses
Fan-out queries using the site: operator
6.8%
Aimed at ~116K different domains
Drop in US ChatGPT answers citing Reddit
−88%
19.1% of answers in March 2026 to 2.3% in September
Fig. 2

Three in four ChatGPT site: searches target a brand's own site

ChatGPT site: search operator use: share of site: searches to the ~160 most-searched domains, by type of site, Aug–Sep 2026

75.9%11.9%
  • Brand sites75.9%75.9%
  • Retailers11.9%11.9%
  • Government and health authorities7.3%7.3%
  • Reddit and Facebook1.9%1.9%
  • Review platforms1.2%1.2%
  • Other1.9%1.9%
  • All site: searches to the top ~160 domains100%100%

A brand site is "a company website like apple.com or ikea.com, as opposed to a retailer, government site, forum, or review site." Shares cover only the ~160 most-searched domains, not all ~116K domains that site: queries pointed at. Values are read from the labels on Ahrefs' chart; 75.9%, 11.9% and 1.9% also appear in its text.

Fig. 3

Consumer brands hold 8 of the 10 most site:-searched domains

Domains ChatGPT searches most with site: in fan-out queries, number of site: mentions, Aug–Sep 2026

apple.com9,442ikea.com7,361nike.com5,210adidas.com4,619samsung.com4,432mi.com3,303lego.com3,106dior.com2,662reddit.com2,586gov.br2,583

Ranks come from Ahrefs' dataset of around 1 million site: queries. Reddit, ninth, is the only one of OpenAI's 17 licensing partners in the top 150. support.apple.com (2,005) ranks 14th as a separate domain.

Fig. 4

How many fan-out queries ChatGPT runs depends on who measures

Average ChatGPT fan-out queries per prompt or response in four studies. Each label shows how many times the lowest figure it is.

Tracked real promptsTest of a fixed prompt set
024686.01 between the lowest and highest averageAhrefs1.6 1.0× lowestPer response · Aug–Sep 2026Peec AI2.1 1.3× lowestPer prompt · Apr 2026Nectiv (2025)2.17 1.4× lowestPer prompt · 2025Nectiv (2026)7.61 4.8× lowestPer prompt · 2026

The studies use different models, account types and prompt sets, so their averages are not comparable as a trend. Ahrefs counts only responses that contained fan-outs. Peec's tracking of 20M+ fan-outs in five countries put the range at 2.4 to 2.8 per prompt in early October 2025 and 2.3 to 2.8 in late January 2026.

Fig. 5

OpenAI's publisher partners doubled their citation share by July

ChatGPT citation share of OpenAI licensing partners among the 200 most-cited domains, US, 2026

05%10%15%20%MarAprMayJunJulAugSep12.5%7.5%

Partners include Dotdash Meredith, Condé Nast, Hearst, News Corp and others with OpenAI content deals. March, July and September are in Ahrefs' text; April, May, June and August are read from its chart labels. September is "late September". Shares are of citations to the top 200 domains, not of all citations.

Fig. 6

ChatGPT leans on official sources less for comparison questions

ChatGPT citation sources by question type, 20,000 US prompts. Ring: comparison questions ("A vs B"). Dot: factual questions (pricing, features, company details).

Comparison questions Factual questions Gap (Factual questions − Comparison questions)
025%50%75%100%GapPrompts citing at least one official source68.3%84.4%+16.1 ptsCitations that went to official websites46.4%63.4%+17 pts

A separate Ahrefs test of 20,000 US prompts, split by intent. Ahrefs does not give the test's exact dates or the share of prompts in each group.

Fig. 7

ChatGPT cites Reddit in 1 in 43 answers, down from 1 in 5

Reddit ChatGPT citations vs. Wikipedia and YouTube: share of US ChatGPT answers citing each site. Ring: March 2026. Dot: September 2026.

March 2026 September 2026 Gap (September 2026 − March 2026)
05%10%15%20%GapReddit2.3%19.1%−16.8 ptsWikipedia2.9%14.1%−11.2 ptsYouTube0.2%0.3%−0.1 pts

Ahrefs Brand Radar, US ChatGPT citations, 2.4M answers in March and 3.8M in September 2026. Relative drops per Ahrefs: Reddit −88%, Wikipedia −80%, YouTube −27% (from unrounded shares). Promptwatch separately measured Reddit's share of all ChatGPT citations falling from 3.8% (July 18–Aug 7) to 0.5% (Aug 14–17), an 86% drop, which it calls provisional.