Do AI citations come from the first 30% of a page? Yes, at 1.47x, and the peak sits at 10-20%

A rule circulating in agency playbooks says 44.2% of AI citations come from the first 30% of a page. The advice built on it is to put the answer in the first 100 to 200 words. The evidence rates that claim Overstated. The number is quoted correctly: across 18,012 verified ChatGPT citations, 44.2% sat in the first 30% of a page Goodwin · Third Door Media / Search Engine Land · 2026B. Adjusted for band width, that is a 1.47x lift over an even spread, and the finer data puts the peak at 10-20% of the page.

AI citations in the first 30% of a page: a 1.47x lift once band width counts

The headline compares three unequal bands as if they were thirds. The study's bands cover 0-30%, 30-70% and 70-100% of a page, so an even spread would give 30%, 40% and 30% Goodwin · Third Door Media / Search Engine Land · 2026B. Against the observed 44.2%, 31.1% and 24.7%, that is 1.47x expected at the top, 0.78x in the middle and 0.82x at the bottom.

The correction shrinks the effect without removing it. A 1.47x lift across 18,012 citations is a real within-page position effect, and the top band clearly wins Goodwin · Third Door Media / Search Engine Land · 2026B. The middle band still runs at 0.78x, a little over half the top band's rate per unit of page.

One limit applies to every figure from this study. Search Engine Land, which reported it, is owned by Semrush, and Semrush is Kevin Indig's research partner Goodwin · Third Door Media / Search Engine Land · 2026B. The Growth Memo primary is paywalled, so the band definitions rest on the trade-press account.

The most-cited slice of a page is the 10-20% band, past the opening

The advice turns a page-level finding into a word count, and the finer data does not support that step. Indig's second positional study drew roughly 98,000 citation rows from about 1.2 million ChatGPT responses across seven verticals Goodwin · Third Door Media / Search Engine Land · 2026B. The 10-20% band performed best in all seven, ahead of the opening 10% Goodwin · Third Door Media / Search Engine Land · 2026B.

The first study points the same way at a finer grain. At paragraph level its pattern inverts: 53% of citations come from mid-paragraph and only 24.5% from first sentences Goodwin · Third Door Media / Search Engine Land · 2026B. The opening line of a block is not where most citations land.

On a 3,000-word guide, the 10-20% band covers words 300 to 600. That sits well past the 100-to-200-word target the circulating advice names. The finding supports a complete, self-contained answer in the first section of a page. It does not support compressing that answer into the first sentence.

The position effect on AI citations varies by vertical and by engine

Where position matters depends on the category, which no version of the circulating rule mentions. In the second study, Finance was steepest, with 43.7% of citations in the first 30%, while Education peaked later at around 30-40% depth Goodwin · Third Door Media / Search Engine Land · 2026B. A universal first-200-words rule is contradicted by the study series that produced its headline number.

Engine choice moves the picture as much as vertical does. Of 11,647 domains cited across five engines, 69.6% were cited by only one Khallad · SurfacedBy · 2026B. Both positional studies measured ChatGPT alone, so their split describes one retrieval pipeline.

The model behind that pipeline keeps changing. Both studies predate the GPT-5.6 retrieval shift and declare neither model variant nor reasoning mode Goodwin · Third Door Media / Search Engine Land · 2026B. In October 2026, Profound tied a drop in tagged ChatGPT referrals to the GPT-6 Intelligent UI rollout Southern · Search Engine Journal · 2026B. A positional split measured on one model version needs re-measuring on the next.

A controlled test finds a U: page conclusions beat the middle

The only controlled manipulation of within-page position found a lopsided U rather than a slope. MaxAEO rotated one identical claim through five page depths in a Latin-square design Han · MaxAEO · 2026C. Survival into AI answers was 72% at the intro, 24% in the dead middle and 46% in the conclusion Han · MaxAEO · 2026C. Because only position changed, content type cannot explain the gap.

The study is Tier C and provisional. It comes from one vendor that sells AI-visibility tooling, with about 1,200 observations and no statistical test Han · MaxAEO · 2026C.

The observational data does not show the same recovery at the foot of the page. In Indig's second study, the bottom 10% of a page earned only 2.4% to 4.4% of citations Goodwin · Third Door Media / Search Engine Land · 2026B. In the first study, the bottom band's 0.82x against the middle's 0.78x is too close to call Goodwin · Third Door Media / Search Engine Land · 2026B.

Prompt-content alignment is the strongest page-level predictor of AI citations

Placement is one page variable among several, and other page variables carry measurable weight. Across about 2 million citations from ChatGPT, Claude, Google AI and Gemini, prompt-content alignment was the dominant page-level predictor, with a beta of +0.37 Moore · Discovered Labs · 2026B. Domain-level AI authority outweighed page features roughly sixfold in the same data. The study is an agency-authored preprint, not peer-reviewed, and covers B2B SaaS only.

Length on its own is close to neutral. Across 174,048 pages cited in Google AI Overviews, word count correlated with citation at 0.04 (Spearman), and 53.4% of cited pages ran under 1,000 words Gavoyannis · Ahrefs · 2025B. The Ahrefs data gives no reason to add length for its own sake.

Answer-first blocks: Google's main-content definition and one provisional test

Google's October 2026 revision of its helpful-content guide calls main-content quality one of the most critical factors in page quality Google · 2026A. It defines main content as anything that directly helps a page achieve its purpose, including tabbed or expanded sections. The guide does not address where on a page that content should sit.

One team-run test points toward answer-first blocks, with a caveat in the same data. OnBoard added a Key Takeaways section with schema markup to one page, and its citation rate rose from 1.45 to 2.33 over six weeks Debski · Peec AI · 2026C. A control page moved from 1.61 to 1.72, while the test page's visibility fell from 22% to 9%. This is Tier C evidence: one page pair, with schema and copy changed together, so it is provisional.