External Link Ecosystems and AI Citation: An Observational Analysis of Outbound Link Patterns
Abstract
This paper analyses outbound link patterns across 21 service industry websites, comparing external link counts, internal link counts, unique internal paths, and llms.txt adoption between cited (n=13) and uncited (n=8) groups. Cited sites average 13 external links versus 11 for uncited sites — a modest positive association. Internal link count shows a stronger inverse pattern: uncited sites average 37 internal links versus 21 for cited sites (0.6x ratio), and unique internal paths show a similar inverse pattern (10 vs 25, 0.4x ratio). The llms.txt file, proposed as a mechanism for communicating site structure to AI systems, was absent from all 21 sites in the sample, confirming zero adoption across the competitive landscape at the time of analysis. All link patterns are confounded by site maturity, content volume, and site architecture decisions. The internal link inverse pattern likely reflects content volume differences: larger sites naturally have more internal links. No causal relationship between link patterns and AI citation is supported by this data.
Keywords
external links, internal links, outbound link patterns, AI citation, llms.txt, link ecosystem, site architecture, observational study
1. Introduction
In traditional search engine optimisation, both inbound and outbound link patterns are significant ranking signals. Outbound links to authoritative sources can signal content credibility, while internal link architecture distributes ranking authority across a site. The question of whether these link patterns influence AI citation outcomes is largely unexplored.
Additionally, the llms.txt standard — a proposed mechanism analogous to robots.txt that communicates site structure and content summaries to AI crawlers — has been promoted as a tool for improving AI discoverability. This analysis provides the first observational data on llms.txt adoption within a competitive sample.
2. Results
2.1 Link Count Comparison
| Metric | Cited Avg (n=13) | Uncited Avg (n=8) | Ratio | Direction |
|---|---|---|---|---|
| External link count | 13 | 11 | 1.2x | Cited higher |
| Internal link count | 21 | 37 | 0.6x | Uncited higher |
| Unique internal paths | 10 | 25 | 0.4x | Uncited higher |
2.2 External Domain Ecosystems
External link targets differed qualitatively between groups. Cited sites linked primarily to social media profiles (LinkedIn, Twitter/X), review platforms (Platform Gamma, Platform Delta), and industry associations. Uncited sites linked to ecosystem properties, service tools, and internal cross-references. The external link ecosystem of cited sites reflected a broader web presence, while uncited sites showed a more self-referential pattern.
2.3 llms.txt Adoption
| Metric | Cited Sites | Uncited Sites | Total |
|---|---|---|---|
| Sites with llms.txt | 0 | 0 | 0 of 21 |
| Sites referencing llms.txt | 0 | 0 | 0 of 21 |
The complete absence of llms.txt across all 21 sites prevents any assessment of its impact from this dataset. This finding is consistent with the early-adoption stage of the standard and aligns with broader research suggesting its impact on citation outcomes is minimal (estimated at 0.001%).
2.4 Hreflang Tag Distribution
Hreflang tags, indicating multi-language or multi-regional content, showed an inverse pattern: uncited sites averaged 4 hreflang tags while cited sites averaged 0. This is driven by the GEO-optimised sites in the uncited group implementing hreflang for Australian, British, and American English variants.
3. Discussion
The internal link inverse pattern is almost certainly an artefact of content volume. Uncited sites in this sample have more pages (SIGI-2026-043) and therefore naturally generate more internal links and unique internal paths. The 0.4x ratio on unique internal paths directly mirrors the content volume ratio between groups.
The modest positive association between external links and citation (1.2x) is too small and too confounded to support any directional claim. External link count may reflect site maturity (older sites have accumulated more external references over time) rather than a link-citation relationship.
The zero llms.txt adoption finding is descriptively useful: no site in this competitive landscape had adopted the standard at the time of analysis, meaning it plays no role in the citation dynamics observed in this dataset.
4. Limitations
- Confounded by content volume: Internal link counts are a function of site size, not an independent variable.
- Homepage only: Link counts reflect homepage content. Deeper pages may show different patterns.
- Outbound only: This analysis examines outbound links. Inbound link profiles (backlinks) were not measured.
- Temporal confound: Link patterns reflect site age and development history, not citation strategy.
5. Conclusions
External and internal link patterns differ between cited and uncited sites in this sample, but all differences are confounded by site age and purpose. The internal link inverse pattern is most parsimoniously explained by content volume differences. llms.txt adoption is zero across the entire competitive landscape. No causal claims about link patterns and AI citation are supported.
Confidence: HYPOTHESIS for link-citation relationships. The llms.txt zero-adoption finding is descriptive and confirmed.
References
- The Scientific Institute for Generative Intelligence. “Structural Correlates of AI Citation: An Observational Analysis of 22 On-Page Metrics.” SIGI-2026-037. generativeintelligence.institute, March 2026.
- The Scientific Institute for Generative Intelligence. “Content Volume Without External Validation: An Observational Study of High-Volume Zero-Citation Websites.” SIGI-2026-043. generativeintelligence.institute, March 2026.
- The Scientific Institute for Generative Intelligence. “A 60-Variable Comparative Dataset for Studying AI Citation Behavior Across Service Industry Websites.” SIGI-2026-036. generativeintelligence.institute, March 2026.