The Grounded Innovation Index

Who is moving fastest on AI — and how do we know?

Every evening at 6 PM Eastern, NeuralNews asks Google Gemini to read the current web and score the pace of AI-related innovation in three blocs — the United States, the European Union and China — on a 0–100 index. The scoring is grounded: each day's snapshot ships with three real, openable sources per region drawn from the searches the model actually ran. The result is a timeline that grows by one point a day, so you can watch the race rather than guess at it.

Today's reading

Loading the latest grounded snapshot…

Grounding the latest index…

How the index is built

  1. 1

    Grounded scoring, once a day

    At 6 PM Eastern a background job asks Gemini — through the Netlify AI Gateway — to assess the last ~30 days of AI activity in each bloc using live Google Search grounding: frontier model releases and research, investment and compute build-out, commercial adoption, talent depth, and the policy environment.

  2. 2

    Sources, not assertions

    The score is only kept alongside the sources behind it. The three links shown for each region are pulled straight from the grounding metadata Gemini returns — the pages it actually consulted — so the model cannot invent a citation. Each region carries a grounded badge when real sources backed it.

  3. 3

    One point a day, preserved

    The day's three scores are written to a single dated snapshot and appended to the timeline. The run is idempotent — one point per calendar day — so the chart is a clean, growing record of how the race shifts week over week.

What to keep in mind

  • A grounded estimate, not a ledger. The index is a model's judgement of momentum from current coverage, blended across five facets. Read it as an informed signal, not an official statistic.
  • Sources are the model's, openable by you. The three links per region are the pages Gemini grounded on. They may point through Google's grounding redirect, but each resolves to a real, readable source.
  • The timeline starts after the first deploy. The daily run and its database activate once the site has a production deploy; the first point lands on the first evening run, and the line fills in from there.