How we collect and publish

What we collect and how we count

ai.NewsCanvasly runs automatically. This page sets out where the material comes from, how it is counted, what is automated, and where a human is involved.

What we collect

Official blogs and release notes from AI companies; academic sources (arXiv, major conferences, research institutes); technical trade press; developer communities; and industry outlets covering adjacent infrastructure such as data centres and power. We poll these five kinds of source daily. We do not filter down to primary sources only — trade reporting and community discussion are included. Instead, sources are weighted by their nature (below).

Time windows

• The default view is the last 7 days. Switching to 14, 30 or 90 days is not yet available.
• This is the window being displayed, not a rule for selecting which articles to cover.
• Aggregation updates in three layers: article collection and summarisation hourly, topic volume and leading indicators daily, and the weekly roundup weekly.

Numbers

There are two kinds of number on this site. Figures inside article summaries appear only if they exist in the source (we round, but never inflate). Separately there are values this site computes itself — topic volume, spike detection. These are not figures taken from any source; they are indicators derived mechanically from the collected articles. Topic volume is the sum of per-article weights assigned by the nature of the source. A spike is judged by how far a recent value sits from its own past average.

What we do not publish

• Anything we cannot link back to its source
• Specific incidents or internal figures from the operator's workplace.

Errors

Summaries are produced by generative AI and can misread a source. When we notice an error, we correct it. There is currently no mechanism that displays a correction history on the page itself, so please check the linked source when accuracy matters.

Use of generative AI, and where a human is involved

Article summaries, topic explanations and leading-indicator text are produced by generative AI and published automatically. The operator does not review each item before publication. What the operator decides is which sources are in scope, how aggregation works, how things are displayed, and what to do when the automated process flags an anomaly. There have been cases where publication was held back because the content was judged wrong, and released only after correction.