Google’s AI Cites “Best” and “vs” Pages 11 Times as Often as ChatGPT

GEOforge Team
October 13, 2026

GEOforge Research · Citation Data

One in seven of Google’s AI citations points to a “best X”, “top N”, “X vs Y” or “alternatives” page. For ChatGPT it is about one in 77. The gap has not moved in 13 weeks.

15% vs 1.3%
Share of citations pointing to listicle-style pages: Google AI vs ChatGPT
11×
How much more often Google AI cites them
208 of 247
Same prompts where Google AI’s share beat ChatGPT’s
18 of 18
Brands with enough data where Google AI cited them more

Marketers often talk about “AI search” as one channel. For one kind of page it is two. When Google’s AI answers back a claim, they reach for roundups and comparisons far more often than ChatGPT does, and that has been true every week for a quarter.

What is the short answer?

  • Google AI cites listicle-style pages 11 times as often as ChatGPT. Across 3,850,973 citations, 14.7% of Google AI Mode and AI Overviews citations and 1.3% of ChatGPT citations point to a URL that signals a best-of, top-N, versus or alternatives page.
  • Nearly half of Google’s AI answers cite at least one. 47% of AI Mode answers and 50% of AI Overviews answers did, against 10% of ChatGPT answers.
  • It is not one brand, one week or one prompt set. Google’s share beat ChatGPT’s in all 18 brands with enough data, in 208 of 247 prompts compared head to head, and in every one of 13 weeks.
  • We can say what each engine cites. We cannot say why.

What did we measure?

GEOforge runs each brand’s tracked prompts against ChatGPT, Google AI Mode and Google AI Overviews on a schedule and stores every source each answer cites. For this piece we used the 13 full weeks from 13 July to 11 October 2026: 327,613 answers that came with a source list, from 786 prompts across 26 brands, citing 3,850,973 sources.

We counted a citation as listicle-style when the path of its URL contains best-, top- followed by a number or word, -vs-, -versus- or alternatives. That is a deliberately narrow, mechanical rule: it reads the address, not the page. It will miss roundups with plain URLs, and it can catch an odd page that happens to use one of those words.

How big is the gap, and has it moved?

Each week, between 14% and 17% of Google AI citations were listicle-style. For ChatGPT the range was 1.0% to 1.6%. The two lines never came close to each other.

Two engines, one stable gap

Citations to best-of, top-N, versus and alternatives pages as a share of all citations, by week. Hover a point for the counts.

Share of citations that point to a best-of, top-N, versus or alternatives page, by week Google AI Mode and AI Overviews combined stayed between 14% and 17% of their citations every week from 13 July to 11 October. ChatGPT stayed between 1.0% and 1.6%. 0 5 10 15 20 31 Aug: AI Mode lists shrink 13 Jul 27 Jul 10 Aug 24 Aug 7 Sep 21 Sep 5 Oct Week starting (2026). Share of all citations that carry a listicle-style URL. Google AI, week of 13 Jul: 14.6% of citations (43,842 of 301,294) Google AI, week of 20 Jul: 16.6% of citations (58,056 of 350,284) Google AI, week of 27 Jul: 14.8% of citations (37,311 of 252,154) Google AI, week of 3 Aug: 14.4% of citations (43,303 of 300,612) Google AI, week of 10 Aug: 13.8% of citations (48,302 of 350,797) Google AI, week of 17 Aug: 14.0% of citations (53,837 of 385,027) Google AI, week of 24 Aug: 15.0% of citations (47,176 of 315,273) Google AI, week of 31 Aug: 14.5% of citations (22,667 of 156,348) Google AI, week of 7 Sep: 14.8% of citations (21,525 of 145,419) Google AI, week of 14 Sep: 15.2% of citations (19,989 of 131,168) Google AI, week of 21 Sep: 14.3% of citations (21,225 of 148,208) Google AI, week of 28 Sep: 14.1% of citations (18,701 of 132,496) Google AI, week of 5 Oct: 14.3% of citations (26,654 of 186,885) Google AI 14 ChatGPT, week of 13 Jul: 1.2% of citations (586 of 48,725) ChatGPT, week of 20 Jul: 1.0% of citations (578 of 56,917) ChatGPT, week of 27 Jul: 1.1% of citations (606 of 53,598) ChatGPT, week of 3 Aug: 1.6% of citations (818 of 52,728) ChatGPT, week of 10 Aug: 1.3% of citations (799 of 60,185) ChatGPT, week of 17 Aug: 1.4% of citations (915 of 66,520) ChatGPT, week of 24 Aug: 1.2% of citations (829 of 67,702) ChatGPT, week of 31 Aug: 1.4% of citations (656 of 46,979) ChatGPT, week of 7 Sep: 1.3% of citations (613 of 47,710) ChatGPT, week of 14 Sep: 1.3% of citations (621 of 47,601) ChatGPT, week of 21 Sep: 1.4% of citations (605 of 44,729) ChatGPT, week of 28 Sep: 1.4% of citations (614 of 43,072) ChatGPT, week of 5 Oct: 1.3% of citations (776 of 58,542) ChatGPT 1
Share of citations that point to a best-of, top-N, versus or alternatives page, by week Google AI Mode and AI Overviews combined stayed between 14% and 17% of their citations every week from 13 July to 11 October. ChatGPT stayed between 1.0% and 1.6%. 0 5 10 15 20 31 Aug 13 Jul 3 Aug 24 Aug 14 Sep 5 Oct Google AI 14 ChatGPT 1
Google AI Mode + AI OverviewsChatGPT
Show the weekly data
Week ofGoogle AI citationsGoogle AI listicle shareChatGPT citationsChatGPT listicle share
13 Jul301,29414.6%48,7251.2%
20 Jul350,28416.6%56,9171.0%
27 Jul252,15414.8%53,5981.1%
3 Aug300,61214.4%52,7281.6%
10 Aug350,79713.8%60,1851.3%
17 Aug385,02714.0%66,5201.4%
24 Aug315,27315.0%67,7021.2%
31 Aug156,34814.5%46,9791.4%
7 Sep145,41914.8%47,7101.3%
14 Sep131,16815.2%47,6011.3%
21 Sep148,20814.3%44,7291.4%
28 Sep132,49614.1%43,0721.4%
5 Oct186,88514.3%58,5421.3%

Google AI figures combine AI Mode and AI Overviews. In June the same measure was 17% for AI Overviews and 1% for ChatGPT, so the pattern predates this window.

Is it the same prompts, or just different questions?

That was our first suspicion: Google and ChatGPT might simply be answering different questions. They are not. The same prompts run on every engine. Taking the 247 prompts with at least 50 citations on both ChatGPT and AI Overviews in the last four weeks, AI Overviews’ listicle share was higher for 208 of them and ChatGPT’s was higher for 17. Pooled across those prompts the shares were 15% and 1.4%.

Is this just one or two brands?

No. Of the 26 brands we track, 18 had at least 500 citations on both ChatGPT and Google AI between 31 August and 11 October. Every one of them showed a higher listicle share on Google. The smallest gap was 5.3 times. Brands in different industries and with different prompt sets all look the same.

Share of each brand’s citations that point to listicle-style pages, 31 Aug–11 Oct
BrandChatGPTGoogle AI
Brand A1.3%29.2%
Brand B4.5%23.7%
Brand C1.4%20.9%
Brand D1.2%18.3%
Brand E2.2%17.6%
Brand F1.6%17.3%
Brand G1.2%16.1%
Brand H2.8%15.2%
Brand I1.9%13.3%
Brand J0.7%13.2%
Brand K0.4%10.2%
Brand L0.6%9.5%
Brand M0.7%6.6%
Brand N0.3%6.5%
Brand O0.2%4.7%
Brand P0.0%3.5%
Brand Q0.5%3.4%
Brand R0.3%2.9%

Which kinds of page drive it?

All four patterns we looked for are far more common in Google’s citations. “Best” pages are the biggest group on Google, and “versus” and “alternatives” pages show the largest relative gaps, at more than 20 times ChatGPT’s rate.

Every listicle pattern is more common on Google

Share of all citations whose URL carries each pattern, 14 September to 11 October. A page can match more than one pattern.

“best-” in the URL
1%
8%
“top-N” in the URL
1%
3%
“-vs-” in the URL
0%
4%
“alternatives” in the URL
0%
0%
ChatGPTGoogle AI Mode + AI Overviews

Bars are scaled to 9%.

Could our own collection explain it?

We checked. The share did not move when Google AI Mode attached about half as many sources to each answer from the end of August (about 20 per answer before, about 8 after). AI Mode’s listicle share was 15% before and 15% after. Excluding the most recent week, Google’s weekly share stayed within 14–17%. If a collection fault were creating the gap, it would have to leave ChatGPT untouched, hold steady through a change in list length, and show up in every brand. We found no sign of that.

A caveat on ChatGPT. About 46% of ChatGPT answers in our data (13 July to 11 October) come with no source list, and we count only answers that have one. If ChatGPT cites listicles more often in answers where it shows no sources, we cannot see it.

What should you do with this?

  • Treat “best X” and comparison pages as a Google AI channel first. Being on the third-party roundups Google cites is a different job from getting your own pages cited, and ChatGPT rewards it far less.
  • Measure the engines separately. A single blended AI-citation number would average a 15% pattern with a 1% one.
  • Check which roundups Google AI actually cites for your category. The pages are specific to each market. Our data shows the pattern, and we have not tested whether getting added to a given list changes your own citations.
  • Do not read this as “ChatGPT ignores comparisons”. It may cite them under plainer URLs, which our mechanical rule would miss.

How did we measure this?

Data. Every answer GEOforge collected between 13 July and 11 October 2026 that carried a list of cited sources: 327,613 answers across 26 brands and 786 prompts. We counted the sources each engine attached to its answer, not links that appear only in the answer text.

Classification. By URL path, as described above. Pages from Google’s own panels were not treated specially because they carry none of these words. Answers whose sources came back as unresolved Google redirect links were excluded.

Normalization. Shares of citations, not counts, because weekly collection volume varies. Weeks start on Monday; the last week runs 5–11 October.

Checks. We split the window by week, compared the same prompts across engines, repeated the comparison brand by brand (18 of 18 agree), looked at each engine on its own, checked that the AI Mode source-list change did not move the share, dropped the latest week, and looked back to June.

Limitations. The prompts are the ones our clients track, so they lean B2B and are not a random sample of all searches. URL classification is mechanical and undercounts listicles with plain URLs. We cannot see why the engines differ, and we did not test whether any listicle placement changes a brand’s own visibility. Every figure is aggregated; no brand, prompt or cited page is identified.

Your citations are not one number. They change by engine.

GEOforge tracks every source your brand and your competitors earn across ChatGPT, Google AI Overviews and Google AI Mode, week by week. 14-day free trial, no sales call.

Start a free trial →

Sources & method. All figures come from the GEOforge production citation database: 13 July to 11 October 2026, 26 brands, 786 prompts, 327,613 AI answers with source lists and 3,850,973 cited sources. Related: Engagement Is a Threshold, Not a Ladder and Google’s AI Cites Your Website Without Naming You 43% of the Time.