Your business publishes information for customers. AI systems can also access files describing an organization, its rules and its sources. Across three sites in one ecosystem, Dual Web Reference documents recorded accesses to these resources, with figures and an inspectable method. This study helps distinguish technical accessibility from accurate understanding. You can examine the results and identify what needs documenting in your own digital presence.
The public DWR repository retains the tables, figures and methodology. This Pagup presentation offers an operational reading. It is neither a promise of visibility in AI answers nor a measurement of a service’s return on investment.
A bounded observation
The window covers September 15 through October 14, 2026 inclusive, in UTC. It includes gautierdorval.com, pagup.com and better-robots.com, three sites within one ecosystem. This is a descriptive observation, not a controlled experiment or a representative market sample.
The main selection contains 245 read requests for which the server returned the success code 200, known as “GET 200” accesses. They concern 174 combinations of site and file path within the selected documentary scope. The same path on two sites therefore counts as two combinations.
The selection includes training, search and on-demand fetch bots. The producer recorded their provider origin as verified; the public tables cannot independently repeat that verification for every request. Each of the 30 days has partial coverage, with no guarantee that every access was recorded.
Within this selection:
- 6 accesses to
/llms.txt. - 239 accesses to other main documentary resources.
- 245 accesses in total.
- 230 automated accesses, excluding on-demand fetches.
In this study, /llms.txt is the text file intended to guide AI systems toward site content. It is one of the observed resources; its presence does not guarantee that a system follows it.
The 239 accesses outside llms.txt do not establish a preference for other files or greater effectiveness. Available content, exposure and demand are not controlled. The 15 on-demand fetches do not establish outside initiation either; initiation provenance remains unknown.
What an organization can reuse
The reusable contribution is a discipline of evidence. Defining included resources first avoids mixing governance documents, control files and addresses used to test site accessibility. Distinguishing how firmly a request’s origin has been established avoids presenting every request with a bot name as established provider origin. Keeping periods, denominators and coverage states explicit makes comparisons conditional on compatible definitions.
A machine-first architecture can then make entry points, canonical sources, policies, entities and documentary relationships explicit. Their presence and their recorded accesses remain separate facts. A declared graph relationship does not establish that an agent followed it; an HTTP request does not establish that a rule was understood or applied.
For a Pagup engagement, this distinction shapes the deliverables: a resource inventory, selection rules, reproducible checks and interpretation limits. It avoids turning a bot access into a promise of citation, recommendation, conversion or causal impact.
The canonical Dual Web definition describes the relationship between human and machine surfaces. DWR documents accesses to part of that documentary layer; this observation does not validate the framework as a whole.
A canonical source with explicit review states
Edition r2 is a privacy revision of the same observations as r1. The original r1 bytes remain accessible. r2 withholds protected daily rows together with the coverage totals and differences that could reconstruct them; a withheld value is not zero. These revisions do not measure change over time.
The editions are publicly accessible, with no recorded reference approval and no documented independent scientific review. Recalculating sums checks a calculation; it does not repeat collection or independently validate every actor’s identity. The method and reproduction procedure state what a reader can actually check.
Pagup points to the repository for the data and their versions. This site does not maintain a second dataset. The observation supports a digital-readability and evidence methodology, subject to the dataset’s explicit limitations.
To examine your own situation, the digital readability diagnostic identifies gaps between your offering, content and accessible signals. This study provides an example of evidence to request and define, rather than a result to promise.
Finding our insights useful?
Add to my Google sources