Transparent methodology

Purpose and unit of analysis

The dataset measures AI Understanding’s own verified news archive. One record equals one stable canonical story URL. When a later development is substantively the same story, editors may update that URL and append a visible update-history entry.

Inclusion and evidence

Records enter the dataset only after publication. The news workflow requires AI to be the direct subject, checks the source material, rejects rumors and minor or duplicate items, and records a primary evidence URL. Vendor claims are described as claims unless independently established.

Calculations

• 7- and 30-day counts use each story’s UTC publication timestamp and rolling windows calculated when requested.
• Distinct source domains normalize the hostname of each primary evidence URL by removing “www”.
• Updated canonical stories have one or more recorded update-history entries.
• Image coverage is the share of records with a recorded story image.
• Topics use published, inspectable rules based on title, excerpt, category, source name, and URL. A story may match more than one topic.

Reproducibility

The public JSON dataset includes summary values, definitions, topic tables, source-domain tables, daily counts, canonical URLs, source records, and update histories. The CSV provides a flat record-level export. Both are regenerated from the same archive query.

Limitations

This is not a census of the whole internet. Feed availability, publisher access controls, verification requirements, editorial thresholds, and source timing affect coverage. Counts do not measure importance, quality, investment, adoption, or public impact. Overlapping topic rows must not be summed.

Corrections and history

Corrections to published stories follow the public corrections log and editorial standards. Dataset output changes automatically when a canonical record is corrected, updated, or removed. Material methodology changes increment the version and are listed here.

Citation

Suggested citation: “AI Understanding. AI News Observatory, version 1.0.0, accessed [date], https://aiunderstanding.org/research/ai-news-observatory.” Data is available under CC BY 4.0.