DYNAMICS-WEB: A Pre-Registered Measurement of Web Content Character over the Link Graph, and a Powered Proof that its Sub-Perceptual Half Changes Conversion

Read full text
Authors: Jason Duke
Cited by: 2

Abstract

Abstract \noindent Search reduces a web page to two properties, its authority and its topic. We show the same link graph carries a third: the character of a page, how it reads across eight axes with subject and standing held fixed, and that the deepest part of this signal changes what a reader does without ever being consciously seen. We present DYNAMICS-WEB, a measurement of web content character, and establish five results in support of one claim. Character is distinct from authority, which reconstructs it at a coefficient of determination of 0.030, and from a full surface battery at 0.256. Its axes transfer directly to Arabic, Chinese and Swahili, and the same engine recovers geographic relevance for forty six countries, so the method is general. Three foundation models from three developers recover five to six of the eight axes in common, so the construct is not an artefact of one teacher. Character propagates through the link graph at about 0.4 per axis against a null of zero. And in a pre registered, powered choice experiment (n = 117), matching the sub perceptual axes of a page to a reader raised conversion from nineteen to thirty one per cent (one sided p = 0.0295), while the loud commercial axes carried none of the lift. We are explicit about the bounds: the conversion effect is modest, a secondary alignment did not replicate, and two axes do not transfer cleanly. The measurement is released in full; a single downstream personalisation function is proprietary and is not needed to reproduce any result here.

Fields of Study & Subjects

Information Retrieval and Search Behavior Digital Communication and Language Reading and Literacy Development

Publication details

Journal / source
Research Square
Publisher
Springer Science and Business Media LLC
Publication date
8 September 2026
Language
EN
Document type
Preprint