J·Index The Journalism AI Index

How the Data Is Gathered

The database that underlies J·Index started in January 2026. The methodology is combined and includes:

1. Manual research by the author.
2. Human-guided research by Claude Code to given parameters (for example, a certain type of sources, or a scarcely covered country or region). This data is verified against a live source.
3. Some research is manually outsourced to other AI models where they have an edge — for example, Gemini to search YouTube podcasts and videocasts.
4. In June 2026 an automated layer was added that polls about 20 trade-press, accelerator and industry sources at various cadences. It surfaces candidate stories for human review.
5. Candidate entries are checked by a person and deduplicated automatically before being added to the database.

How the Dashboard Is Built

Behind the dashboard is a single master table: one verified case per row, 18 structured fields each — organization, country, use case, tool, domain, function, human-involvement level, adoption stage, funder, source link and more — all drawn from a fixed controlled vocabulary so the data stays consistent as it scales.

Outlets are classified by their audience and home country — where their journalism actually lands — not by a headquarters chosen for tax, exile or diaspora reasons. An exile outlet reporting for a home-country audience from abroad is filed under that home country, not its host.

A deterministic build step reads that table and generates this dashboard.

When automated, human-reviewed sweeps add cases, the same build re-runs and the charts update. You can see the date of the last update on the masthead and in the footnotes.