What Makes AI Cite a Page? A 96,418-URL Study
·
AEO for LatAm Team
SISTRIX compared 96,418 URLs and found that pages cited by ChatGPT, AI Mode and AI Overviews more often had a table of contents, Article schema and at least five H2 headings than matched pages that were not cited [1]. The company published the study on October 22, 2026, using data from the second quarter of 2026 for the DACH region (Germany, Austria and Switzerland) [1]. The authors state that the results are correlations, not proven causes [1].
How did SISTRIX run the study?
SISTRIX paired every cited page with a non-cited page from the same domain, so brand strength and domain authority do not explain the differences [1]. The sample holds 96,418 URLs from 17,380 domains, with 43 page features checked on each URL [1].
Results are reported as factors. A factor of 1.0 means cited and non-cited pages look the same on that feature, and 1.7 means cited pages are 1.7 times as likely to show it [1].
The study covers ChatGPT, AI Mode (Google’s conversational search mode) and AI Overviews (the AI summaries above Google results), with about 16,000 cited URLs for each engine [1]. Perplexity, Claude and Gemini chat were not included [1].
Which page features lined up with citations?
Structure features led the list. Each factor below compares a cited page with a non-cited page from the same domain [1]:
A table of contents with jump links: 1.7, and only 14% of pages have one [1].
Article schema, which is structured data that labels a page as an article: 1.6 [1].
At least five H2 headings: 1.6, although the authors caution that longer texts naturally contain more subheadings [1].
FAQPage schema: 1.5 [1].
At least one question written as an H2: 1.4 [1].
A visible last-updated date: 1.3, and only 5% of pages show one [1].
SISTRIX also combined eleven features into one score. Pages that met many of them were cited 2.1 times as often as barely optimised pages, and only 10% of cited pages met many of them [1].
Does llms.txt help pages get cited?
In this study, llms.txt showed no benefit. llms.txt is a plain-text file that lists a site’s main pages for AI systems. SISTRIX writes: “As things stand, there is no measurable effect on AI citations.” [1]
Of all cited URLs, 11.7% came from domains with an llms.txt file, compared with 14.0% across all domains [1]. The comparison was made at domain level, so other differences between sites may affect it [1].
What are the limits of this study?
The authors list several limits, and each one changes how far you can apply the numbers [1]:
Only the DACH region was studied, and the page does not explicitly state the language of the pages [1].
The results show correlations. They do not show that adding a feature causes a citation [1].
Title and H1 findings were analyzed retrospectively and are described as initial indications [1].
The year-in-title factor of 2.7 may reflect content that is kept fresh rather than the year itself [1].
What should a marketing team do with these findings?
AEO for LatAm’s recommendation: use the list as a test plan on your ten most important pages, not as a guarantee. Content for Extraction shapes your content so AI engines can quote it, with answer-first pages, FAQs and comparisons. It is one of the four AEO for LatAm pillars, and the SISTRIX pattern points in the same direction [1].
AEO for LatAm’s recommendation: add a table of contents with jump links to guides longer than 1,000 words, since only 14% of pages in the sample have one [1]. Then check the same prompts in ChatGPT, Claude, Gemini and Perplexity before and after the change.
AEO for LatAm’s recommendation: keep llms.txt as low-cost housekeeping and do not plan a citation lift around it, because this study found none [1].
Frequently asked questions
What is the SISTRIX citation study? It is a matched-pair analysis of 96,418 URLs from 17,380 domains, published October 22, 2026. SISTRIX compared pages cited by ChatGPT, AI Mode and AI Overviews with non-cited pages from the same domain across 43 features, using data from the second quarter of 2026 [1].
Does FAQPage schema help AI citations? FAQPage schema had a factor of 1.5, meaning cited pages carried it more often than matched non-cited pages [1]. Breadcrumb, HowTo and QAPage schema showed 1.0, which means no measurable difference [1]. The study shows correlation only, so adding the markup alone does not guarantee a citation.
Does the study apply to websites outside Europe? The data covers the DACH region only, and the page does not state the language [1]. Results for the United States or Latin America may differ. AEO for LatAm’s recommendation: treat the factors as hypotheses and test them on your own pages and prompts.
Does a longer article get cited more often? Pages with at least 1,000 words had a factor of 1.5, and the authors say this mainly reflects better structure rather than length itself [1]. Readability scores showed no significant effect, with a factor of 1.03 [1]. Clear headings and a table of contents matter more than word count.
Should I put the current year in my titles? Titles with the current year showed a factor of 2.7, but SISTRIX analyzed titles retrospectively and calls the result an initial indication [1]. The effect may reflect freshly maintained content, and the data cannot show whether changing the year alone is enough [1].
Run a Quick-Start Blueprint on your domain.
References
Johannes Beus, Julia Weissbach and the SISTRIX Content Team. SISTRIX. “Citation Worthiness: Which GEO Tactics Hold Up Across 96,418 URLs.” Published October 22, 2026, last modified October 23, 2026. Data collected in the second quarter of 2026, DACH region. https://www.sistrix.com/blog/citation-worthiness-study/ (opened October 26, 2026).