Editorial Standards

The commitments CoreProse makes for every generated article.

Last updated: April 20, 2026

Sources and verification

Every article is built from an internal knowledge base, and factual claims are traceable to their source.

KB cross-referencing: information is checked across multiple sources before being used.
Zero invented citations: our pipelines validate that every cited source actually exists (claim-validator + orphan_source_id guard).
Source transparency: the list of sources used is shown in the transparency badge above every article.

The model that wrote and the pipeline that built are declared explicitly, not hidden.

GPT-5.x model family (OpenAI) — declared in HTML meta tags and Schema.org JSON-LD of every article.
Multi-stage pipeline: brief -> plan -> draft -> revision. Each stage is auditable via agent_run_logs.
EU AI Act Art. 50 disclosure: every article is machine-readable marked as "ai-generated" or "ai-assisted" per its mode.

Accuracy is not a marketing aspiration — it's enforced by our technical guardrails.

Schema.org validator: @types with missing required fields downgrade to Thing rather than emit invalid data (GSC clean).
RAG poisoning defense: suspicious KB sources are filtered before integration into the writer's context.
Prompt injection hardening: user inputs are neutralised before integration into LLM prompts (OWASP LLM01).

Being transparent includes admitting where AI can be wrong.

AI has no reliable knowledge of very recent events (past 24-72h) — we fall back to the web when the KB is insufficient but we flag uncertainty.
Fine-grained numerical comparisons (prices, exact benchmarks) require an explicit source — never invented.
Editorial opinions belong to the cited sources or the human author (for ai-assisted articles), not to CoreProse as an organization.