Methodology
Citead computes the GEO score with a published formula.
This page details the score's eleven sub-measures, the weight of each criterion and the studies behind them.
We publish the formula so that every score can be checked.
To fix a page, you need to know which criterion is pulling it down and by how much.
Each sub-score is shown with the evidence found in the text and the fix to apply.
Eleven sub-measures and freshness make up the score.
The GEO score is not the average of the five criterion scores. It starts from eleven sub-measures of the text, each passed through a square root to limit the effect of a measure pushed to the extreme, then weighted: content counts for 92% and freshness for 8%. The bars below give the share of each criterion, that is the sum of the weights of its sub-measures. Naturalness is scored and shown, but does not count in the score. The weights are described in the paper arXiv:2609.07559.
Citead scores five criteria.
Each agent scores one criterion. Four criteria count in the GEO score; naturalness is shown separately.
Semantic richness
52.8%Analyzes linguistic quality through information theory metrics. Higher scores indicate more substantive, well-structured content.
What it analyzes
- Information density (Shannon entropy)
- Lexical diversity (TTR, MTLD)
- Semantic coherence
- Entity flow
- Topic consistency
AI citability
31.4%Measures what makes a passage quotable as is by an AI engine: direct answers, self-contained phrasing, attributed data. This criterion does not predict that content will be cited, it scores what is under your control on the page.
What it analyzes
- Answer-first structure
- Statistical data density
- Source attributions
- Self-contained paragraphs
- Quotable statements
- Entity density
- Author credentials
- Readability level
Content organization
6.0%Evaluates information architecture and how well content is structured for both human readers and AI parsing.
What it analyzes
- Section order
- Relevance-diversity balance
- Heading hierarchy
- Logical content flow
Authority & citations
1.8%Measures whether sources are cited and relevant, and how dense the verifiable facts are.
What it analyzes
- Citation presence
- Domain authority
- Citation relevance
- Q&A optimization
- Factual density
Naturalness
not in scoreFlags wording, vocabulary and constructions over-represented in AI-generated text. This score is shown separately and does not count in the GEO score.
What it analyzes
- Vocabulary patterns
- Structural patterns
- Contrast constructions
- Linguistic variety
- Natural flow
Research
Published studies have measured what generative engines cite.
The foundational GEO paper
10,000 queries tested on Perplexity and Bing Chat in 2023: verifiable statistics, expert quotes and answer-first structure came out as the strongest levers. On current engines our measurements show markedly weaker effects. Citead reuses this paper's evaluation method and sets aside its gain figure.
Read the paperQuotation addition
Adding quotable statements experts can reference
Statistics addition
"8% of visits with an AI summary end in a click, versus 15% without (Pew Research Center, 2025)" instead of "few users click"
Cite sources
Links to studies, official documentation
AI engines favor third-party sources
Chen et al. (2025) compared the sources cited by ChatGPT, Perplexity and Gemini with Google's, across several verticals and languages: AI engines show a systematic bias toward earned media (third-party sources) over brand-owned content. Citing recognized sources in your pages does not replace that external coverage, but it is the part of authority you control.
Read the paperWhat this means for you
The score weights are dated and versioned
AI engines do not all cite the same sources, and their models change regularly. The score therefore measures what depends on the page: its sources, its structure and its wording.
The current weights (version v5) have been in production since June 2026; this page describes that version.
Frequently asked questions
Does the GEO score predict that a page will be cited by ChatGPT or Perplexity?
No. The score measures page properties you control: sources, structure, wording. Being cited also depends on signals outside the page, especially third-party media coverage, which Chen et al. (2025) identify as a dominant factor for AI engines.
Why does the Authority criterion check every factual claim?
Because generative engines attribute their sources poorly. Liu, Zhang and Liang (2023) audited four engines (Bing Chat, NeevaAI, perplexity.ai, YouChat): on average, 51.5% of generated sentences were fully supported by their citations, and 74.5% of citations supported their associated sentence. A claim already tied to its source in your page leaves less room for error.
Is there a specific optimization for Google AI Overviews?
According to Google Search Central, there are no additional requirements to appear in AI Overviews or AI Mode: the page must be indexed and eligible to be shown with a snippet. The score criteria therefore apply to Google as to other engines.
Does the Naturalness criterion rely on an AI detector?
No. It flags phrasing and vocabulary that are over-represented in generated text, one by one. A binary detector verdict would be unreliable: Liang et al. (2023) show that common detectors misclassify writing by non-native English authors as AI-generated.
References
- Aggarwal et al., « GEO: Generative Engine Optimization », KDD 2024
- Liu, Zhang, Liang, « Evaluating Verifiability in Generative Search Engines », 2023
- Chen, Wang, Chen, Koudas, « Generative Engine Optimization: How to Dominate AI Search », 2025
- Google, « Search Quality Rater Guidelines » (E-E-A-T)
- Google Search Central, « Creating helpful, reliable, people-first content »
- Liang et al., « GPT detectors are biased against non-native English writers », 2023
- Shannon, « A Mathematical Theory of Communication », 1948
- Reimers, Gurevych, « Sentence-BERT », EMNLP 2019
- Liu et al., « Lost in the Middle: How Language Models Use Long Contexts », 2023
- Google Search Central, « AI features and your website »
- Pew Research Center, « Google users are less likely to click on links when an AI summary appears in the results », 2025