# Positional Bias in AI Citations: 74.9% of Cited Sentences Are in the First Half of the Page **URL:** https://organikpi.com/blog/content-strategy/top-of-page-positional-bias-ai-citations/ **Published:** 2026-04-28 **Modified:** 2026-07-02 **Author:** Daniel Shashko > Our May 2026 study of 153,425 citations finds the mean cited sentence position is 37% through the document. 74.9% of cited sentences appear in the first half of the page. The bottom quarter of a page receives only 7.4% of all citations. Our March 2026 study of 42,971 citations found a consistent mean position of 34.9%. Position is calculated as block_index / total_blocks across all p, li, h1-h6, and td elements. Two mechanisms drive the bias: author convention placing key claims in the lead, and RAG chunking that scores top-of-page content higher on retrieval. The fix is restructured pages: lead every section with the citable claim, add context after. Combine positional restructuring with register and sentence-length audits for maximum citation lift. --- > Our May 2026 study of 153,425 citations finds the mean cited sentence position is 37% through the document. 74.9% of cited sentences appear in the first half of the page. The bottom quarter of a page receives only 7.4% of all citations. Our March 2026 study of 42,971 citations found a consistent mean position of 34.9%. Position is calculated as block_index / total_blocks across all p, li, h1-h6, and td elements. Two mechanisms drive the bias: author convention placing key claims in the lead, and RAG chunking that scores top-of-page content higher on retrieval. The fix is restructured pages: lead every section with the citable claim, add context after. Combine positional restructuring with register and sentence-length audits for maximum citation lift. In our May 2026 study of 153,425 citations, the mean cited sentence position was 37% through the document, and 74.9% of cited sentences appeared in the first half of the page. The bottom quarter of a page receives only 7.4% of citations. Where your claim sits on the page is one of the strongest predictors of whether AI search will surface it. ## The May 2026 positional data In our [May 2026 study of 153,425 citations](https://organikpi.com/blog/seo-strategy/ai-mode-text-fragments-dead-153425-citations/) across Google AI Mode, Gemini, Perplexity, ChatGPT, Copilot, and Grok, we matched every cited sentence to its source page and calculated relative position as block_index / total_blocks. Blocks include all p, li, h1-h6, and td elements. The results: MetricMay 2026 (153,425 citations)March 2026 (42,971 citations) Mean position37% through document34.9% through document Citations in first half74.9%~75% (estimated from March data) Citations in bottom quarter7.4%not separately reported The March 2026 figure of 34.9% mean position, from our [42,971-citation study](https://organikpi.com/blog/geo-ai-search/decoded-42971-ai-citations-google-research/), is consistent with the May result. The difference between 34.9% and 37% reflects the larger, more diverse May dataset rather than a directional change in behavior. Both studies confirm the same structural reality: AI engines overwhelmingly cite the opening of a document. 74.9% of all cited sentences in the first half of the page is a strong structural signal. If your most citable claim sits in the second half of a 2,000-word post, it is competing for less than 25% of the citation budget. The bottom quarter, below the 75% mark, draws only 7.4% of citations total. The three structural checks every page needs before it is ready for AI citation: position, sentence length, and readability register. ## Why the top of the page wins Two mechanisms combine to produce the positional effect. Understanding both helps you write pages that work with the bias rather than against it. - **Author bias toward the lead.** Skilled writers front-load their most direct, declarative claims because readers expect the answer in the opening. The [BLUF (Bottom Line Up Front) writing structure](https://organikpi.com/blog/content-strategy/bluf-writing-format-ai-content/) mirrors how AI retrieval pipelines work: the highest-value claim first, supporting evidence after. AI extraction reflects the same convention human readers prefer. - **Retrieval scoring on chunked content.** RAG-based pipelines split pages into chunks before scoring relevance. Top-of-page chunks tend to contain the query terms in tight proximity because introductions are written to answer the core question directly. These chunks score higher on [semantic retrieval](https://organikpi.com/blog/technical-seo/content-chunking-rag-seo/), which translates directly to citation probability. Either mechanism alone would produce a positional bias. Together, they create a citation zone that is real, small, and anchored to the top 37% of your document. ## What the positional bias does NOT mean Do not delete the bottom half of your content. The data shows where citations come from, not that the bottom of the page has no value. Three reasons to keep the depth: - **Retrieval confidence depends on supporting depth.** AI engines assess topical authority partly by the breadth of coverage on a page. A thorough page that covers sub-questions earns higher domain trust even if sub-sections are rarely cited directly. - **Organic ranking still matters.** In our May 2026 study, 76.95% of cited URLs are not in the organic top-10. But the 23.05% that are in the top-10 still represent a large citation volume. Organic ranking amplifies citation probability. Long, thorough content tends to rank better. - **Multi-page reader journeys start from cited sentences.** A user who clicks through from an AI citation lands on your page. Supporting depth keeps them there, which feeds [AI search conversion rates](https://organikpi.com/blog/seo-strategy/geo-roi-ai-traffic-revenue/). Adobe Analytics reported that during the 2025 holiday season AI-referred shoppers converted 31% more than other traffic sources. The correct action is to restructure pages so that citable claims appear early. Depth at the bottom is fine. Hiding your best claim at the bottom is the problem. ## How to restructure pages for positional bias In our client work we use a five-point structural checklist before publishing any pillar or cluster post. The [BLUF format](https://organikpi.com/blog/content-strategy/bluf-writing-format-ai-content/) is the starting framework, adapted for AI search. - **Open with a direct answer, not context.** Most content writers start with history, definitions, or “why this topic matters.” That setup pushes the citable claim past the 37% mark before the reader gets there. Start with the answer. Add context after. - **Write a TLDR or summary block in the first 150 words.** Pack it with 6-12 [atomic sentences](https://organikpi.com/blog/content-strategy/atomic-sentence-seo-ai-citations/), each carrying one independently citable fact. The 6-10 word range accounts for 45.2% of all citations in our May 2026 study. - **Lead every H2 section with a declarative sentence.** The first sentence under each heading should answer the implied question of that section. Supporting evidence follows. This ensures every section has at least one citable sentence near the top of that content block. - **Move case studies and narrative examples to the back half.** Stories and anecdotes are rarely cited directly. They belong after the citable claims, not before. Restructuring a post to lead each section with the claim and close with the example is typically a one-hour edit. - **Avoid context-dependent sentences.** Sentences that reference “as discussed above” or “in the previous section” break atomic fact extraction. Every cited sentence must stand alone as a complete, falsifiable claim. See our guide to [atomic sentence structure](https://organikpi.com/blog/content-strategy/atomic-sentence-seo-ai-citations/) for the rewrite framework. ## The pages where positional bias matters most The positional effect is most pronounced on long-form content above 2,000 words. On pages under 800 words, the entire page sits close enough to the citation zone that structure matters less. Audit priority: Page TypePositional Audit PriorityReason Pillar posts (2,000+ words)HighestKey claims can easily fall past 40% mark Category landing pagesHighOften written with intro-heavy structure Product comparison pagesHighComparison queries trigger AI Mode frequently Cluster posts (800-1,500 words)MediumShorter pages; less likely to bury claims Short FAQ or glossary entriesLowEntire page is within citation zone Product pages are a counterintuitive priority. Comparison and “best X for Y” queries trigger AI Mode at high rates. If your product page buries the core differentiation claim in section 4 behind feature lists and company history, AI Mode will rarely surface it. The [Google AI Mode optimization playbook](https://organikpi.com/blog/seo-strategy/google-ai-mode-optimization-playbook/) we use starts every product page audit by locating the first claim of differentiation and measuring its block position. ## A practical restructuring checklist Run this audit on your top 10 highest-traffic pillar posts: - Open the post and identify your five most important claims, the facts you would most want an AI engine to surface. - Measure the block position of each claim: count the total number of block elements (p, li, h1-h6, td) in the post, then count which block number each claim falls on. Divide to get its percentage position. - Flag any claim with a position above 37%. These are the restructuring targets. - Rewrite those sections to lead with the claim. Move the supporting context after the claim, not before. - Track citation rates in the [GEO/AEO Tracker](https://organikpi.com/tools/geo-aeo-tracker/) (github.com/danishashko/geo-aeo-tracker) weekly after publishing the restructured version. Most citation changes are visible within 3-4 weeks. The most common structural mistake we find in client audits: the introduction sets context for three paragraphs, section one covers background, section two covers definitions, and the actual differentiated claim appears in section three or four. That pattern pushes the citable content past 40% in almost every case. Restructuring those posts to lead with the claim is usually a 60-90 minute edit per article that produces measurable citation lift. ## Combine positional and readability signals Positional bias and [bimodal readability](https://organikpi.com/blog/content-strategy/bimodal-readability-ai-search/) are not independent. A claim in the top 37% that is written in the Flesch 50-59 dead zone still underperforms. A technically excellent Very Easy sentence buried at 70% through the document rarely gets cited. Both signals must be right. In our [GEO audit](https://organikpi.com/services/geo-audit/) process, we run three checks simultaneously: position audit, register audit (Flesch scoring), and sentence length audit (targeting the 6-10 word range that accounts for 45.2% of citations). All three are required for a sentence to enter the high-probability citation zone. A page that passes only two of three leaves significant citation share on the table. For a full walkthrough of how we apply this in practice, see the [BLUF format guide](https://organikpi.com/blog/content-strategy/bluf-writing-format-ai-content/), the [atomic sentence guide](https://organikpi.com/blog/content-strategy/atomic-sentence-seo-ai-citations/), and the [bimodal readability post](https://organikpi.com/blog/content-strategy/bimodal-readability-ai-search/). Together they cover the three dimensions of sentence-level citation optimization that our research identified as the highest-leverage content structure decisions in 2026. If you want us to map the positional profile of your current pillar pages alongside the full citation audit, the [content strategy service](https://organikpi.com/services/content-strategy/) includes this analysis as a standard deliverable. We report block position, Flesch band, and sentence length for every paragraph in your top 20 posts, along with a prioritized rewrite roadmap. ## Frequently Asked Questions ### What percentage of AI citations come from the first half of a page? In our May 2026 study of 153,425 citations, 74.9% of cited sentences appeared in the first half of the page. The mean cited position was 37% through the document. The bottom quarter of a page receives only 7.4% of all citations. ### What was the mean citation position in the March 2026 study? Our March 2026 study of 42,971 citations found a mean cited position of 34.9% through the document. This is consistent with the May 2026 figure of 37%. Both studies used block_index / total_blocks as the position metric across all p, li, h1-h6, and td elements. ### Why do AI engines prefer content near the top of the page? Two mechanisms combine to produce this effect. First, skilled writers naturally front-load their most direct claims in the opening, following the BLUF convention that both readers and AI pipelines prefer. Second, RAG-based retrieval systems chunk pages and score top-of-page chunks higher because introductions tend to place query terms in close proximity. ### Should I delete the bottom half of my content to improve AI citations? No. The bottom half still provides supporting depth that builds topical authority and improves organic ranking, which in turn amplifies citation probability. The correct action is to restructure pages so citable claims appear in the top 37%, while keeping supporting evidence and examples in the lower sections. ### Which pages should I prioritize for a positional audit? Pillar posts over 2,000 words are the highest priority because key claims can easily fall past the 40% mark in long-form content. Category landing pages and product comparison pages are next. Pages under 800 words are lower priority because the entire page sits within the citation zone.