From raw occurrence records to accessible knowledge: Transforming biodiversity infrastructure through evidence-manifesting design

Biodiversity informatics has made occurrence records widely accessible, yet many databases remain optimized for human browsing rather than structured machine consumption by large language models (LLMs) and AI agents. This creates a bottleneck: analytical capacity is accelerating while biodiversity evidence remains fragmented, weakly documented, and difficult to cite reproducibly. Here we introduce evidence manifestation, a design approach that transforms raw occurrence data into versioned, citable, machine-readable knowledge artefacts with explicit provenance. Using the World of Crayfish®, we implement this concept through Species Exposure Bundles (SEBs) that package curated occurrence snapshots, spatial layers, derived range metrics (extent of occurrence, area of occupancy), geographic summaries (basins, administrative units, protected areas), and machine-readable metadata exposed via stable identifiers (JSON/GeoJSON). To validate real-world utility, we conducted controlled LLM query tests contrasting closed-book baseline (T0), web search (T1), and manifest-directed retrieval with SEB access (T2) across four AI applications. Nine endemic or regional crayfish taxa spanning heterogeneous documentation levels were compared against one high web-visibility control species ( Procambarus virginalis ). For endemic taxa, closed-book performance was low (mean 1.78/8; 47% complete failures), web search improved but remained variable (mean 4.22/8; 19% failures), and manifest-directed retrieval produced consistently high performance (mean 6.97/8; 0% failures; 50.0% perfect scores). The control species achieved 5.75/8 under closed-book conditions (T0), exceeding endemic web-search performance and approaching endemic manifest-directed performance, highlighting accessibility gaps for underrepresented taxa. These results position evidence manifestation as a practical, reproducible infrastructure pattern for AI-assisted biodiversity synthesis and conservation workflows.