How does this work?
When you ask a conservation question, Conservation Evidence Copilot reads it carefully, routes it to the right evidence sources, filters out irrelevant results, and synthesises everything into a structured, cited answer.
1
Question Analysis
Copilot parses your question to extract the species, habitat, intervention type, and outcome.
It flags ambiguous queries so you can refine them, and looks up scientific names to make sure
it's working with the right organism.
2
Selecting Agents
A supervisor AI decides which specialist tools to call in. Most questions go to the Conservation
Evidence database. Questions about specific invasive species also trigger a statistical
meta-analysis. You only get the evidence that's relevant to your question.
3
Searching the Evidence Base
The system searches thousands of conservation interventions using both keyword matching and
semantic meaning-matching simultaneously, then ranks results by relevance.
4
Statistical Meta-analysis
For invasive species questions with sufficient data, Copilot runs a pooled statistical model
across all available studies to calculate an overall effect size.
5
Relevance Filtering
An LLM judge reviews every search result against the specific criteria of your question before
any answer is written. Results are sorted into relevant, marginal, or off-topic, so nothing
irrelevant makes it into your answer.
6
Synthesis and Quality Checking
A synthesis AI writes a structured answer with a summary, detailed findings, recommendations,
and limitations. Every citation is verified against the database, and every factual claim is
checked to ensure it has a reference behind it.