Original research · Rankability

Does ranking in search predict getting cited by AI?

We mapped 2,824 AI citations against organic rankings across 37 keyword markets and 8 AI engines. The short answer: ranking high helps a lot — but it was never the whole story.

37
keyword markets
2,273
ranking domains
2,824
AI citations
4
search indexes
8
AI engines
Jun 18, 2026
snapshot date
If you rank #1–3
99%cited by AI
If you don't rank at all
76%still cited
FINDING 01

If you rank, rank high.

Among pages that rank in Google, position is strongly correlated with AI citation. The decline from the top of page one down to position 21+ is steep and monotonic — every band lower means materially fewer AI citations.

Spearman ρ
−0.50
ranked pages only · strong negative
Significance
p<.001
bootstrapped across reports
95% CI
−.55,−.44
tight — robust effect
Share of pages cited by at least one AI engine, by Google position band
n = 2,273 domains · "Not ranked" shown in amber as the exception to the trend
100 75 50 25 0 99% 1–3 111 pages 92% 4–10 241 pages 48% 11–20 310 pages 38% 21+ 279 pages 76% Not ranked 1,332 pages

The twist: "Not ranked" sits above the 11–20 and 21+ bands. Directory and listicle pages get cited by AI without ranking organically at all — the thread we pull on next.

FINDING 02

But ranking isn't required.

Sort every citation by where the cited page ranks and a clean split appears. About a third of AI citations go to pages that don't rank in Google at all — directories, listicles and review sites the model surfaces directly.

~1 in 3

citations go to non-ranking pages. That's why the overall correlation collapses once you include them:

ρ = −0.50 ρ = 0.03 (n.s.)
ranked-only → all pages included
NON-RANKING 32.8%
  • 37.3% Both — ranks and AI-cited
  • 32.8% AI-only — cited, doesn't rank
  • 29.9% Search-only — ranks, not cited
FINDING 03

Each AI engine tracks the search index it's built on.

We measured how much each engine's citations overlap the top-10 of four search indexes (top-10 set overlap, Jaccard). The backing-index hypothesis held for 4 of 8 engines. Hover any row for the exact overlap and significance.

Matches backing index Tracks a different index

A long bar with a redirect icon isn't a contradiction — Perplexity and Grok overlap Google's top-10 the most, but each was expected to track its own hybrid index, so it counts as tracking a different one.

EngineBest-matched indexTop-10 overlap (Jaccard)Match
Brave AI logoBrave AI Brave logoBrave 0.560
Perplexity logoPerplexity Google 0.589
Copilot logoCopilot DuckDuckGo logoDuckDuckGo 0.462
Grok logoGrok Google 0.342
Gemini logoGemini Google 0.283
Google AI Mode logoAI Mode Google 0.260
Claude logoClaude Brave logoBrave 0.208
ChatGPT logoChatGPT No index 0.068

Want Claude to cite you? Rank in Brave. Want AI Mode or Gemini? Rank in Google. The index behind the engine is the lever — optimize for it, not the chatbot.

ChatGPT is the exception. It tracks no index in a statistically meaningful way (best overlap just 0.068, n.s.) and is driven almost entirely by directories and listicles.

FINDING 04

A spectrum from search-followers to directory-scrapers.

The share of an engine's citations that go to non-ranking pages sorts the engines into two camps. Low share = it mostly cites what ranks. High share = it pulls from directories the model trusts directly.

Search-follower Directory-scraper
Perplexity 12.2%
Grok 19.9%
Brave AI 21.2%
Gemini 42.3%
AI Mode 49.0%
Claude 51.1%
Copilot 61.7%
ChatGPT 77.0%

Percent of each engine's citations pointing to pages that do not rank in Google's top results. ChatGPT pulls more than three-quarters of its citations from non-ranking sources.

WHAT THIS MEANS FOR YOU

Four moves to get cited by AI.

Rank top-3, not just page one

Positions 1–3 are cited 99% of the time vs 38% at 21+. The top of page one is where AI citation becomes near-certain.

Earn directory & listicle placements

A third of citations go to non-ranking pages. Getting listed in the round-ups AI trusts is a parallel path to ranking yourself.

Optimize for the backing index

Target Claude via Brave, Gemini & AI Mode via Google, Copilot via the Bing family. Pick the index, not the chatbot.

Treat ChatGPT as its own channel

It ignores ranking signals — 77% of its citations are non-ranking pages. Win it through directories and third-party mentions.

Share this study

Found this useful? Send it to your team.

If this changed how you think about AI citations, pass it on. Marketing leads, agency owners, and in-house SEO teams all need to see it.

Tip: highlight any sentence in the article to share it as a quote.

METHOD & LIMITATIONS How we ran this study
01

Unit of analysis: domain. Each keyword's per-index exports are merged on domain (exact URLs vary across exports), taking each domain's best position per index and best citation position per engine.

02

Index coverage. Google (37/37), Bing (19/37), Brave (37/37), DuckDuckGo (37/37). Bing appears in only 19 of 37 reports, so the ChatGPT/Copilot→Bing test is under-powered; Copilot's link to the Bing family shows mainly through DuckDuckGo, a Bing-powered index.

03

"Not ranked" handling. Kept, not dropped, to avoid selection bias. The overall correlation is reported both with these pages (ρ = 0.03, n.s.) and without them (ρ = −0.50, p<.001).

04

Significance. Per-engine×index overlaps carry 95% bootstrap CIs; the backing-index winner is tested with a paired Wilcoxon across reports; the overall correlation CI is bootstrapped by resampling reports (the independent unit).

05

Snapshot & causation. A single date with no time dimension. Causation is not claimed — rankings and citations may share common drivers such as authority and relevance.