RESEARCH PAPER — SIGI-2026-040 — March 2026

Lexical Signatures of AI-Cited Versus Non-Cited Service Websites: A 306-Word Corpus Analysis

Q: What are the most distinctive words in AI-cited service websites?

'Brand' is the most distinctive word, appearing 300 times on cited sites versus 37 on uncited sites. Other cited-dominant words include 'designers' (67 vs 0), 'hire' (37 vs 6), 'product' (32 vs 13), and 'animation' (31 vs 8). These reflect industry-specific service vocabulary rather than optimisation language.

Q: Why does GEO-specific terminology appear only on uncited sites?

Words like 'citation' (31 occurrences), 'ecosystem,' 'multiproperty,' and 'generative' appear exclusively or predominantly on uncited sites. These are the meta-language of GEO optimisation itself, used by sites explicitly designed to be cited by AI. The paradox is that the vocabulary of optimisation for AI citation is anti-correlated with actually being cited.

V. Tavitian, J. Tavitian — The Scientific Institute for Generative Intelligence

Category C — Competitive Intelligence • Evidence Level 2-3 • Observational Corpus Analysis

This paper analyses 306 unique words extracted from the top-frequency vocabulary of 21 anonymised service websites, classified by AI citation outcome. Cited and uncited sites display distinct lexical signatures. "Brand" is the dominant word in cited content with 300 occurrences versus 37 on uncited sites. Agency and identity vocabulary dominates cited content, while GEO-specific meta-language — including "citation" itself (31 occurrences, exclusively uncited) — appears only on uncited sites. Social proof language dominates cited content; operational/structural language dominates uncited content. These vocabulary differences are associated with citation outcomes but are confounded by site purpose, industry vertical, and audience.

What Are the Most Distinctive Words in Cited Websites?

The word frequency analysis reveals a clear vocabulary profile for cited service websites. "Designers" appears 67 times on cited sites and 0 times on uncited sites — the most exclusively cited-dominant word in the corpus. "Hire" (37 vs 6), "artists" (20 vs 0), "players" (16 vs 0), "platforms" (16 vs 0), and "professionals" (10 vs 0) are all exclusively or predominantly cited-site vocabulary.

These words reflect the core business vocabulary of established service providers — game development studios, design agencies, and creative platforms. They describe what the businesses do and who works there. The vocabulary is service-oriented, not optimisation-oriented.

Word	Cited Frequency	Uncited Frequency	Dominant In
designers	67	0	CITED
hire	37	6	CITED
product	32	13	CITED
animation	31	8	CITED
citation	0	31	UNCITED
artists	20	0	CITED
ecosystem	0	19	UNCITED
developers	20	7	CITED
unreal	18	8	CITED
generative	0	13	UNCITED
players	16	0	CITED
optimization	0	13	UNCITED
professionals	10	0	CITED
traditional	0	13	UNCITED

Table 1. Selected words from the 306-word corpus showing frequency distribution between cited and uncited sites. All site names removed from source data.

Why Does GEO Meta-Language Appear Exclusively on Uncited Sites?

Perhaps the most striking finding is that the vocabulary of AI citation optimisation itself is anti-correlated with being cited. "Citation" appears 31 times across uncited sites and 0 times on cited sites. "Ecosystem" (19 occurrences), "generative" (13), "optimization" (13), and "traditional" (13, used in the context of "traditional SEO versus GEO") are all exclusively uncited vocabulary.

The explanation is structural, not causal. The uncited sites in this sample include GEO-focused properties whose core business is citation optimisation — naturally, their content discusses citations, ecosystems, and optimisation. The cited sites are service providers (game studios, design agencies) whose content discusses their services. The vocabulary difference reflects what these businesses do, not what language choices cause or prevent citation.

However, the pattern does raise a methodological concern: sites that are explicitly optimised for AI citation using current best practices are not achieving citation in this sample, while sites following no deliberate GEO strategy are highly cited. This may reflect the confound of site age and authority, or it may reflect a genuine disconnect between current GEO recommendations and actual citation mechanisms.

"Citation" appears 31 times exclusively on uncited sites. The meta-language of GEO optimisation is anti-correlated with being cited in this observational sample. This association is confounded by site purpose and age.

What Social Proof Language Characterises Cited Content?

Social proof language — words associated with testimonials, reviews, and client relationships — dominates cited site vocabulary. Words in this cluster include established brand names (anonymised in our analysis), words like "professionalism," "exceptional," "success," and "recommend." These reflect the accumulated social proof of established businesses with years of client relationships and platform reviews.

Uncited sites show minimal social proof vocabulary, which may reflect their newer status (fewer clients to reference) rather than a deliberate content choice. The social proof vocabulary difference is therefore another dimension of the age/authority confound that pervades this dataset.

What Are the Limitations of Lexical Signature Analysis?

Vocabulary differences in this analysis are maximally confounded by at least four factors: site purpose (service provider versus optimisation consultancy), industry vertical (game development vocabulary versus GEO vocabulary), target audience (prospective clients versus marketing professionals), and site age (established versus new). No individual word-frequency difference can be attributed to a causal effect on citation probability.

The analysis is also limited to the top-frequency words extracted from the initial site audit. Lower-frequency words that might reveal more nuanced patterns are not captured. The 306-word corpus represents the most prominent vocabulary differences but does not constitute a comprehensive linguistic analysis.

Suggested Citation

Tavitian, V. & Tavitian, J. (2026). Lexical Signatures of AI-Cited Versus Non-Cited Service Websites: A 306-Word Corpus Analysis. The Scientific Institute for Generative Intelligence, SIGI-2026-040. https://generativeintelligence.institute/publications/SIGI-2026-040/

Frequently Asked Questions

What are the most distinctive words in AI-cited service websites?+

"Designers" (67 vs 0), "hire" (37 vs 6), "product" (32 vs 13), and "animation" (31 vs 8) are the most cited-dominant words. These reflect service vocabulary rather than optimisation language.

Why does GEO-specific terminology appear only on uncited sites?+

"Citation" (31 occurrences), "ecosystem," "generative," and "optimization" appear exclusively on uncited sites. These sites are GEO consultancies whose content naturally discusses optimisation. The vocabulary reflects business purpose, not a causal factor.

Do these vocabulary differences prove what content gets cited?+

No. Vocabulary differences are confounded by site purpose, industry vertical, audience, and age. The vocabulary reflects what these businesses do, not necessarily what causes citation.

APPLIED RESEARCH

Apply This Research

For applied GEO services based on this research → TDS GEO Agency

TDS GEO Agency Contact SIGI