The Two-Clock Model: Structural Presence and AI Perception of Technology Entities

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Working paper introducing the two-clock model, which separates two things AEO practice usually conflates: structural presence S(t) — how machine-readable an entity's website is — and AI perception P(t) — what AI platforms actually know and cite about that entity. The two move at very different speeds, and their difference, the perception gap G(t) = S(t) − P(t), is largest when an entity is new and closes only as third-party citation accumulates. The paper assembles five mechanisms (knowledge cutoffs, temporal misalignment, fact decay, retrieval staleness, and long-tail knowledge) to argue that a wide early gap is a structural feature of how language models work, not a sign of failed optimization. It then tests the predictions on fifty technology entities launched January 2023–March 2025, reconstructing AI perception from dated OpenAI model snapshots (a model-cutoff natural experiment), measuring structure from archived homepage snapshots, and tracking citation via weekly GDELT news mentions. Key findings: (1) the birth gap is universal — structure averages 64.2% of maximum at birth versus 0.5% for perception, a 63.6-point mean gap with zero counter-examples; (2) entity-mention precision is bimodal — distinctive names exceed 70% precision while common-word names collapse; (3) citation ramps precede perception onsets in 28 of 33 entities (85%; median lead 83 days; two-sided exact sign test p = 6.6 × 10⁻⁵), supporting a citation-transmission mechanism (reported as supporting evidence, not proven causation). For AEO practice, the paper offers a way to distinguish expected lag from genuine failure, a diagnostic threshold (perception should measurably narrow by month six), and three levers that close the gap: citation building, external mentions, and answer-format content. Companion dataset and analysis code: https://doi.org/10.5281/zenodo.21532575 (CC-BY-4.0). Working paper; not peer reviewed.