What models help with
Models help search for relevant material, read long announcements and documentation, compare claims, pull together timelines, identify unanswered questions and draft dossier copy. They are particularly useful when one development is spread across product pages, release notes, pricing terms, research and independent tests.
What AI does not get to decide
AI does not decide that a draft is ready for the public site, and it cannot publish autonomously. New dossiers arrive as private drafts. Updates to a published dossier arrive as proposed versions while the current page stays live. Humans review, edit and explicitly publish the result.
Sources remain visible
The public dossier links to the useful material behind it. Sources are not displayed as a score or a decorative proof wall; they let a reader inspect the announcement, documentation, benchmark or reporting directly. Keeping the source register also makes it easier to challenge an interpretation and correct a mistake.
AI can be wrong
A model can misread a table, merge two product tiers, miss a date or produce a confident sentence that the source does not support. Human review reduces that risk but does not make the system infallible. Readers can report a problem through the contact page, and meaningful corrections are handled visibly.
Models will change
BTN may use different models or providers as their capabilities, pricing and data terms change. The editorial standard should not depend on a model brand. The durable parts are the workflow: research beyond the announcement, retained sources, clear uncertainty, review before publication and versioned updates.
Understanding first, control around it
AI is used for understanding and software is used for control. That division is deliberate. Models are good at reading and synthesis; deterministic application rules are better for keeping drafts private, requiring explicit update targets, retaining previous versions and ensuring that only an authenticated human action publishes.