How early GEO metrics measure the amount and placement of answer content associated with a source.
Word Count Impression measures textual exposure
Aggarwal and colleagues define Word Count Impression as the share of response words contained in sentences attributed to a source. If one source receives 120 of 400 answer words and another receives 40, their observed shares are 30% and 10%.
The metric captures answer real estate, not truth, attention, or causality.
Position-adjusted share gives earlier passages more weight
Position-Adjusted Word Count applies a decaying weight so answer content near the beginning contributes more than equally long content near the end. It formalizes the intuition that users are more likely to notice early information.
The weighting assumes an attention pattern; it does not directly observe one.
Repetition creates distributed visibility
A source used for a definition, explanation, example, and comparison occupies a different position in the answer than one cited once. Reference count and paragraph coverage reveal whether visibility is isolated or spread through the response.
Human attention still needs behavioral validation
Eye tracking, reading completion, recall, clicks, and user-choice studies would be needed to validate how much early placement actually changes attention in generative interfaces.
Prominence metrics should therefore be reported as observable answer properties, not direct measurements of human impact.
Questions about this topic
What is Word Count Impression?+
It is the proportion of generated answer words associated with sentences citing a particular source.
What does Position-Adjusted Word Count add?+
It gives greater weight to attributed text that appears earlier in the answer.
Does earlier position prove more attention?+
No. It is a plausible assumption that requires validation through behavioral measures such as recall, clicks, or eye tracking.
Why measure citation repetition?+
Repeated citations reveal whether a source is visible across several answer units rather than appearing as a single isolated reference.
References
Sources are listed in APA 7 style. Preprints are identified as such and should not be treated as peer-reviewed findings unless separately published.
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., & Narasimhan, K. (2024). GEO: Generative engine optimization. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (pp. 5–16). Association for Computing Machinery. https://doi.org/10.1145/3637528.3671900
- Martinez, O. (2026). Optimizing visibility in generative engines: A critical survey of generative engine optimization (2023–2026) [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2607.14035