Part of what Gydmation AI does beyond generating a workflow is explain it — what each node does, why it's connected the way it is, and how to extend it later. That explanation is only useful if it's actually correct, and a language model's unassisted recall of node behavior is exactly the kind of detail it can state fluently and get wrong.
Every explanation of a node or workflow pattern is generated alongside a citation back to the specific documentation source it drew from, not produced first and justified after. If a claim about how a node behaves can't be traced to an actual source in our knowledge base, it doesn't get stated as fact — the same discipline research assistants like Perplexity built their credibility on, applied to workflow guidance instead of web search.
A wrong citation in a casual chat answer is an inconvenience. A wrong explanation of what a node does, acted on by someone editing a live automation, breaks something that was working. Guidance about infrastructure changes carries a higher cost of being subtly wrong, which is why we treat "traceable to a real source" as a requirement here, not a nice-to-have.
Citation grounding only helps as much as the underlying knowledge base is current — a correct citation to outdated documentation is still a wrong answer. Keeping that source material in sync with n8n's own node updates is an ongoing part of this work, not a problem we consider solved.
