Parse
IndexPricing
Sign in
  1. Research
Research
How ChatGPT's gpt-6 shift cut web searches 42.5% while expanding citations 66.7%

How ChatGPT's gpt-6 shift cut web searches 42.5% while expanding citations 66.7%

Published Nov 20, 2026

On this page
  • Matched shift
  • site: collapse
  • Citations per search
  • 10-week timeline
  • 1 vs 5+ searches
  • Source-type shift
  • By industry
  • What to do
  • Study scope
  • How we measured
  • More like this
  • Matched shift
  • site: collapse
  • Citations per search
  • 10-week timeline
  • 1 vs 5+ searches
  • Source-type shift
  • By industry
  • What to do
  • Study scope
  • How we measured
  • More like this

Dimitry ApollonskyFounder, Parse

DataWhat reviewed means

When OpenAI moved ChatGPT Search from gpt-5-6 to gpt-6 in October 2026, did it change how many web searches the engine runs behind the scenes and how many sources it cites?

We ran a controlled matched-question experiment on 24,323 identical organic buyer questions answered by both gpt-5-6 and gpt-6 from October 1 through November 19, 2026, embedded that transition inside a 10-week longitudinal timeline (September 14 through November 19, 2026), and measured fan-out depth across 237,904 gpt-6 answers.

Across 24,323 identical organic buyer questions answered by both gpt-5-6 and gpt-6 in October and November 2026, gpt-6 cut web search fan-out by 42.5% (3.67 to 2.11 queries) and virtually eliminated site: lookups (31.67% down to 0.86%). At the same time, cited URLs per answer jumped 66.7% (6.67 to 11.12) and distinct cited domains grew 58.7% (4.63 to 7.35).

gpt-6 runs 42.5% fewer web searches yet cites 66.7% more URLs

Holding the exact prompt list fixed across 24,323 organic buyer questions answered by both gpt-5-6 and gpt-6 between October 1 and November 19, 2026 removes prompt mix as a variable. The shift is sharp in opposite directions: average web searches per answer fell 42.5% from 3.67 (median 4) to 2.11 (median 2).

Fewer searches did not mean fewer citations. Cited URLs per answer surged 66.7% from 6.67 to 11.12, and distinct cited domains expanded 58.7% from 4.63 to 7.35. gpt-6 runs tighter search fan-out and pulls far more pages out of every query it issues.

Mean web searches per answer (-42.5%)
3.67 to 2.11

Median dropped from 4 queries to 2 queries

24,323 identical questions · gpt-5-6 vs gpt-6

Mean cited URLs per answer (+66.7%)
6.67 to 11.12

Distinct domains jumped from 4.63 to 7.35 (+58.7%)

More citations from far fewer search calls

Share of answers running a site: lookup
31.67% to 0.86%

Targeted domain searches virtually disappeared

2025/2026 year modifiers climbed 44.13% to 68.27%

Average response characters (+63.3%)
5,058 to 8,261

Brands named ticked up from 6.42 to 6.68

Longer, denser answers with deeper source lists

Matched-question shift from gpt-5-6 to gpt-6

  • gpt-5-6
  • gpt-6
  • Web searches per answer
  • gpt-5-63.67
  • gpt-62.11-42.5%
  • Cited URLs per answer
  • gpt-5-66.67
  • gpt-611.12+66.7%
  • Distinct cited domains
  • gpt-5-64.63
  • gpt-67.35+58.7%
  • 051015
Oct 1 to Nov 19, 2026 · 24,323 identical organic buyer questions answered by both models.

Takeaway

Do not assume fewer fan-out searches mean less web reliance; gpt-6 searches half as often and cites two-thirds more pages.

site: domain lookups dropped from 31.67% to 0.86% as fresh and review modifiers surged

The way ChatGPT writes its background search queries changed completely. Under gpt-5-6, 31.67% of answers issued at least one site: operator search to look directly inside a specific brand or publisher domain. Under gpt-6, site: lookups collapsed to 0.86%, a 37-fold drop.

In place of domain-scoped searches, gpt-6 leaned heavily into broad freshness and buyer-intent modifiers. Queries containing 2025 or 2026 rose from 44.13% to 68.27%, review modifiers more than doubled from 4.92% to 10.02%, and pricing/cost modifiers climbed from 8.29% to 11.82%. Reddit query modifiers held nearly flat (1.05% vs 1.13%).

Search query operators and modifiers on 24,323 matched questions

  • gpt-5-6
  • gpt-6
  • 2025/2026 year modifier
  • gpt-5-644.13%
  • gpt-668.27%
  • Domain site operator
  • gpt-5-631.67%
  • gpt-60.86%
  • Pricing/cost/price modifier
  • gpt-5-68.29%
  • gpt-611.82%
  • Review/reviews modifier
  • gpt-5-64.92%
  • gpt-610.02%
  • Reddit modifier
  • gpt-5-61.05%
  • gpt-61.13%
  • 0%20%40%60%80%
Share of answers containing at least one search query matching each operator or modifier.

Takeaway

gpt-6 stopped running direct site: lookups against your domain and shifted to open-web year, review, and pricing searches.

Citation yield per search query nearly tripled from 2.21 to 6.21 URLs

Because search calls fell while cited URLs expanded, the number of cited pages pulled per search query rose from 2.21 under gpt-5-6 to 6.21 under gpt-6. Average response length expanded 63.3% alongside those extra links, growing from 5,058 characters to 8,261 characters.

Meanwhile, the average number of brands named per answer barely moved, ticking up from 6.42 to 6.68. Buyers see roughly the same shortlist size, backed by much longer answers and 66.7% more supporting URLs.

6.21
cited URLs extracted per web search under gpt-6
Up from 2.21 under gpt-5-6 (24,323 matched prompts)

Takeaway

Every individual web query now carries nearly three times the citation weight it did under gpt-5-6.

The model shift flipped cleanly in the week of October 5, 2026

Weekly tracking across ten weeks from September 14 through November 19, 2026 pins down the exact transition. In the weeks of September 28 and October 5, gpt-5-6 averaged 3.57 and 3.79 web searches and 6.38 and 6.88 cited URLs per answer (replacing earlier untagged answers that logged 1.00 recorded query and 5.92 to 6.21 URLs).

In the transition week of October 5, gpt-6 entered with 14,912 answers averaging 2.20 searches and 10.28 cited URLs alongside 24,272 gpt-5-6 answers at 3.79 searches and 6.88 URLs. From October 12 through November 16, gpt-6 held tight bands of 2.02 to 2.19 searches and 10.92 to 12.20 cited URLs every week, while average brands named rose gradually from 5.64 to 7.59.

Weekly search queries and cited URLs across the gpt-5-6 to gpt-6 rollout

  • gpt-5-6 searches
  • gpt-6 searches
  • gpt-5-6 cited URLs
  • gpt-6 cited URLs
Explore the data
2026-09-142.28Not observed4.87Not observed
2026-09-212.54Not observed5.17Not observed
2026-09-283.57Not observed6.38Not observed
2026-10-053.792.26.8810.28
2026-10-12Not observed2.19Not observed10.92
2026-10-19Not observed2.16Not observed12.2
2026-10-26Not observed2.08Not observed11.35
2026-11-02Not observed2.02Not observed10.93
2026-11-09Not observed2.08Not observed11.23
2026-11-16Not observed2.12Not observed11.44
Sep 14 to Nov 16, 2026 · Weeks with >=100 answers per model. Transition overlap occurs in the week of 2026-10-05. Early untagged single-search rows (Sep 14 to Sep 28) are retained in the CSV.

Takeaway

The structural break happened immediately on model cutover in early October, not as a gradual drift.

76.18% of gpt-6 answers run 1 or 2 searches, while 5+ queries pull 15.51 URLs

Across 237,904 post-rollout gpt-6 answers from October 12 through November 19, 2026, ChatGPT settles most buyer prompts quickly: 32.11% (76,386 answers) issue 1 query and 44.07% (104,851 answers) issue 2 queries.

When gpt-6 fans out deeper, citations scale sharply. Single-query answers average 8.58 cited URLs across 6.34 domains and name 5.73 brands, whereas answers running 5 or more queries (5.39% of answers) cite 15.51 URLs across 8.29 domains and name 7.78 brands. Prompt word count rises slightly from 16.7 words at 1 query to 19.8 words at 5+ queries.

Within-gpt-6 fan-out depth vs citations and brand mentions

Avg cited URLs · 5 results

  • 5+ queries15.51
  • 3 queries12.95
  • 4 queries12.93
  • 2 queries12.11
  • 1 query8.58
  • 05101520
Explore the data (5 rows)
1 query76,38632.1116.78.586.345.736,519
2 queries104,85144.071812.117.886.848,612
3 queries27,76611.6719.312.957.857.8410,514
4 queries16,0866.7619.712.937.146.859,160
5+ queries12,8155.3919.815.518.297.7810,409
Oct 12 to Nov 19, 2026 · 237,904 organic gpt-6 answers with at least 1 recorded search query.

Takeaway

Three out of four gpt-6 answers finish in one or two searches, so pages must surface inside the first two query formulations.

Ecommerce, editorial, and directory shares grew while documentation share fell

On the 24,323 matched questions, the overall expansion in cited URLs boosted raw citation counts across every category, but category shares shifted noticeably. Ecommerce jumped from 5.54% of links under gpt-5-6 (8,235 links) to 7.92% under gpt-6 (19,638 links), more than doubling total link volume.

Editorial domains rose from 4.03% to 4.74% and directories climbed from 1.70% to 2.25%. Meanwhile, documentation dropped in relative share from 13.75% to 11.34%, even though absolute documentation links still increased from 20,453 to 28,120.

Source-type diet shift on 24,323 matched questions (gpt-5-6 vs gpt-6)

gpt-6 share (%) · Top 10 of 11

  • General33.43%
  • SaaS30.25%
  • Documentation11.34%
  • Ecommerce7.92%
  • Editorial4.74%
  • Reference3.68%
  • Government3.27%
  • Directory2.25%
  • News1.78%
  • Blog0.75%
  • 0%10%20%30%40%
Explore the data (11 rows)
General49,92633.5682,93033.43
SaaS46,66631.3775,04230.25
Documentation20,45313.7528,12011.34
Ecommerce8,2355.5419,6387.92
Editorial5,9974.0311,7574.74
Reference5,0003.369,1293.68
Government5,1023.438,1093.27
Directory2,5361.75,5822.25
News2,8151.894,4251.78
Blog8740.591,8670.75
Social1,1410.771,4520.59
Oct 1 to Nov 19, 2026 · Top 11 classified source types (148,745 gpt-5-6 links vs 248,051 gpt-6 links).

Takeaway

Because gpt-6 abandoned site: lookups and searched more review and pricing terms, ecommerce, editorial, and directory sites captured the fastest share gains.

Apps and Software trigger 2.6 searches per answer vs 1.10 in Consumer Goods

Search fan-out inside gpt-6 depends heavily on market complexity. Across 22 industry groups with at least 1,000 gpt-6 answers, Apps (2.65 searches, 40.91% running 3+ searches), Software (2.59 searches), Data and Analytics (2.57 searches, 41.37% running 3+ searches), Artificial Intelligence (2.51), and Information Technology (2.50) sit at the top.

Consumer Goods sits at the opposite extreme: 93.62% of answers run a single web search (1.10 average searches, just 1.68% running 3+ searches), followed closely by Food and Beverage (1.31 searches, 77.44% single-search) and Real Estate (1.37 searches, 73.61% single-search).

gpt-6 search fan-out intensity across 22 industry groups

Avg searches · Top 10 of 22

  • Apps2.65
  • Software2.59
  • Data and Analytics2.57
  • Artificial Intelligence2.51
  • Information Technology2.5
  • Content and Publishing2.35
  • Internet Services2.31
  • Education2.24
  • Financial Services2.23
  • Hardware2.19
  • 0123
Explore the data (22 rows)
Apps1,3152.659.8140.9112.66
Software5,4992.5914.4634.6212.77
Data and Analytics2,9732.5716.6241.3713.32
Artificial Intelligence1,6062.5111.5833.3112.09
Information Technology3,9612.515.3533.5512.78
Content and Publishing1,7102.3529.1830.4712.11
Internet Services1,0102.3123.9630.511.21
Education1,5922.2424.6226.8210.79
Financial Services3,3782.2318.8926.6710.46
Hardware1,9852.1933.429.7211.73
Professional Services2,1312.0933.0825.1111.52
Sales and Marketing1,0752.0722.9822.0511.24
Transportation2,1321.9731.5219.659.53
Manufacturing1,0141.935.916.3712.91
Gaming1,4371.7334.034.87.45
Health Care3,2541.7351.0114.419.19
Commerce and Shopping2,8101.6263.1714.279.64
Media and Entertainment1,1961.5261.128.869.58
Community and Lifestyle1,9501.4868.6710.679.73
Real Estate2,2891.3773.616.738.2
Food and Beverage1,3341.3177.445.327.65
Consumer Goods1,6621.193.621.688.13
Oct 12 to Nov 19, 2026 · Primary market industry groups with >=1,000 gpt-6 answers.

Takeaway

B2B software categories still generate 2.5+ searches and 12 to 13 citations per answer, whereas consumer retail prompts settle in a single query.

What to do with these numbers

Stop relying on brand-scoped discovery: gpt-6 virtually ended site: lookups (0.86% of answers), so your owned domain must compete on open-web queries rather than direct domain filtering. Keep fresh 2026 dates, clear pricing tables, and structured comparison pages live, because 68.27% of gpt-6 search queries include a current-year modifier, 11.82% ask for pricing or cost, and 10.02% ask for reviews.

Earn coverage in editorial roundups, directories, and retail listings alongside your own site. Those third-party categories picked up the biggest share gains under gpt-6 while average cited URLs expanded 66.7% to 11.12 per answer.

  • Track how AI model shifts change your visibility over time
  • Inspect live ChatGPT search queries and cited pages via Parse MCP

Study scope and counting rules

The controlled comparison pairs the latest organic ChatGPT Search answer per model on 24,323 distinct questions that logged at least one recorded web search query under both gpt-5-6 and gpt-6 from October 1 through November 19, 2026. Query operator rates count the share of answers in which at least one generated search string matched a site: prefix, a 2025 or 2026 year token, reddit, review/reviews, or pricing/cost/price.

The 10-week timeline spans September 14 through November 19, 2026 across weekly model cohorts with at least 100 recorded answers (early September answers without model tags logged 1.00 search query and are excluded from the labeled gpt-5-6/gpt-6 chart series while preserved in dataset.csv). The within-gpt-6 fan-out cohort covers 237,904 organic gpt-6 answers from October 12 through November 19, 2026 with at least one recorded search query, and the industry breakdown covers the 22 primary Parse industry groups with at least 1,000 answers. The source-type table covers 148,745 gpt-5-6 and 248,051 gpt-6 classified citation-domain links across the 11 main source types in the 24,323-question matched sample.

How we measured

We analyzed 24,323 identical organic buyer questions answered by both gpt-5-6 and gpt-6 (October 1 to November 19, 2026), a 10-week ChatGPT Search timeline (September 14 to November 19, 2026), and 237,904 gpt-6 answers with recorded search queries (October 12 to November 19, 2026).

fewer web search queries per answer under gpt-6
-42.5%

3.67 (gpt-5-6) down to 2.11 (gpt-6) on 24,323 matched questions

more cited URLs per answer under gpt-6
+66.7%

6.67 (gpt-5-6) up to 11.12 (gpt-6) on the same questions

of gpt-6 answers use a site: search operator
0.86%

Down from 31.67% under gpt-5-6

cited URLs extracted per search query under gpt-6
6.21

Up from 2.21 under gpt-5-6

Get the data

Full study dataset (CSV)Matched-question metrics, 10-week rollout timeline, 5 fan-out depth buckets, source-type shares, and 22 industries.

More like this

Cited sources fell after a ChatGPT model update
ChatGPT's flagship upgrade cut cited sources per answer by 48% (23.4 to 12.1) while an unchanged control model held steady.
What does ChatGPT search for?
ChatGPT Search often searched for brands the user had not named, and those brands frequently appeared first in its answers.
Which pages does ChatGPT search for on brand websites?
One in three ChatGPT Search queries ran a site: lookup. Looked-up brands got their own site cited 85% of the time versus 46% without one.
How ChatGPT's writing style changed by model version
Stock AI wording became less common across the ChatGPT model versions we compared on the same prompts.
AI citation Pareto curve across 4.18M links
0.39% of root domains (676 of 175,035) capture 50% of ChatGPT Search citations and 60.85% of cited URLs appear once, yet on matched prompts Google AI Mode concentrates its Top 100 tighter (42.00% vs 35.28% and 29.88%).
For agents: this page as Markdown, https://parse.gl/research/gpt-6-search-fanout-and-citation-shift.md. JSON, MCP · free for the last 90 days, checks from $0.01
Open in ChatGPT (opens in a new tab) Open in Claude (opens in a new tab)

This page

As Markdown: https://parse.gl/research/gpt-6-search-fanout-and-citation-shift.md
As CSV: https://parse.gl/research/gpt-6-search-fanout-and-citation-shift/dataset.csv
Reuse: Figures and tables on this page are free to republish with attribution: a link to https://parse.gl/research/gpt-6-search-fanout-and-citation-shift beside them. A standard followed link (dofollow) is preferred.

Read and search

Any brand's shares and rivals, with 95% ranges: GET https://agent.parse.gl/v1/tools/check_brand?subject=%3Cbrand-domain%3E
Third-party cited pages missing a brand: GET https://agent.parse.gl/v1/tools/search_cited_pages?args=%7B%22pages%22%3A%22third_party%22%2C%22listsBrand%22%3Afalse%2C%22filter%22%3A%7B%22brand%22%3A%5B%22%3Cbrd_id%3E%22%5D%2C%22brandRule%22%3A%22not_named%22%2C%22market%22%3A%5B%22%3Cmkt_id%3E%22%5D%7D%7D
Search the index: GET https://parse.gl/api/public/search?q=<words>

Connect

Connect an assistant: https://agent.parse.gl/mcp

An agent can save work before signup: POST https://agent.parse.gl/v1/workspaces

What it costs

No account: Public figures and up to 5 examples per list, about 120 brands a day per address.
Free account: $5 in credit, your brand's first check (20 buyer questions) and a free weekly check.
After your first top-up: Exports and deeper search results. The first top-up starts at $5.
Live checks: $0.01 per question per engine ($0.05 on Gemini and Claude).
Full public answers and pages: 25 free per day, then $0.005 each.
Public answers older than 90 days: $0.001 each the first time your workspace counts or lists one, then yours.

Prices: https://parse.gl/pricing.md

Map of the site: https://parse.gl/llms.txt

See how ChatGPT's gpt-6 search behavior changes citations for your brand

Start free in your browser.
Parse

Index

  • Brands
  • Markets
  • Topics
  • How we measure

For agents

  • Docs
  • Connect
  • Workflows
  • llms.txt

Parse

  • Use cases
  • Pricing
  • Research
  • Blog
  • Compare

Company

  • About
  • Contact
  • Corrections

© 2026 Parse, Inc. All rights reserved.

Privacy PolicyTerms of Service