OpenTelemetry GenAI Semantic Conventions: A Practical Guide

OpenTelemetry’s GenAI semantic conventions provide a standardized `gen_ai.*` vocabulary for observing LLM and agent workloads across providers, SDKs, and frameworks, replacing inconsistent custom fields with shared attributes for operations, providers, requested and served models, token usage, sampling parameters, response completion reasons, errors, and conversation IDs. The conventions support spans for individual AI operations, histograms for token consumption and latency, and separately enabled content events for prompts and responses, whose capture is intentionally opt-in because of privacy and PII risks. Core chat and embedding attributes are considered sufficiently mature for production dashboards, alerts, cost analysis, and latency SLOs, while tool-calling, agent, and multi-agent attributes remain in development and may change. The guidance recommends using automatic instrumentation where available, mapping older `gen_ai.system`, `llm.*`, and provider-specific fields to current standard names, treating finish reasons as arrays, grouping multi-turn activity through `gen_ai.conversation.id`, and redacting captured message content through an OpenTelemetry Collector before storage.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.