How the AI analyst works
A rules-based engine understands the question, runs a real query, and reports only what it computed.
The analyst is deliberately not a free-writing chatbot.
Pipeline
- Understand. It reads your sentence and extracts the intent (area, change, ranking, comparison, carbon, concept…), any country, years and numbers (for example "10%" or "5 km").
- Calculate. It runs the matching analysis on Atlas' own database, switching to the country you named.
- Answer. It writes the result with the numbers, a table or chart, the sources and a confidence label.
- Suggest. It offers follow-up questions.
It never invents numbers. If Atlas lacks the data it says so. For concept questions it quotes the Knowledge Base.
Optional wording model. If configured, a language model may rephrase an answer from the computed facts only; it never sees the database and cannot add numbers.
Records. Every question is logged (who, when, what intent) so answers can be improved. Institutional admins can ask 120 questions an hour, other users 30.
Related reading
More in Methodologies
- Confidence labels: calculated, illustrative, assumed
- The official yearly series (1985–2025)
- Spatial statistics: Moran's I and Getis-Ord Gi*
- Why Atlas totals are 2.3% higher than the official figures
- How the data-quality score is calculated
- The rules behind recommendations
- Repairing invalid geometry
- Net change versus gross loss and gain
Still have a question?
The AI analyst answers from these articles and from the data.
Ask Atlas