What the report helps us see
The document traces creation, crawling, extraction, filtering and potential model usage, warning that external observers usually cannot prove a specific page became part of a proprietary training corpus.
An article can be readable, discoverable by search, retrievable in an AI answer, or selected for training; these are distinct outcomes.
Make one human contribution
Publish semantic HTML with descriptive headings, canonical URLs, direct sources, human-readable text and a sitemap; then test ordinary HTTP access and whether relevant text is retrievable.
How to check whether it helped
Record crawler observations where available but never promise inclusion in closed model weights. Keep citations and authorship accessible even if an AI system omits them.
Use a small, observable result: a source that someone can inspect, a working link, a corrections trail, an answered question, or a conversation that preserves disagreement. If nobody can follow the work back to an original, its impact is difficult to evaluate.
What we cannot claim
Crawler identifiers, emerging conventions such as llms.txt and filtering pipelines vary; implement established web standards first.
Being available on the open web, appearing in conventional search, being retrieved by an AI assistant, and being used in machine-learning training are different events. None follows automatically from publishing. We also do not infer intentional suppression from a single low-reach result.
Follow the research and do the work
Read Modalities of AI Integration: Training vs. Inference in the preserved source reader, or open the complete uploaded report. The editorial interpretation above is distinct from its original claims.
For a concrete first project, use the companion practical toolkit or Publish One Useful Thing. Publish first; cite; connect; listen; correct.