Training is how a model learns; inference is when it works — taking your question (and any retrieved content) and generating the answer. Everything an agent does for a user happens at inference time.
This distinction matters for findability: a model's training is fixed and dated, but at inference it can pull in live content. That live step is your opportunity to be included, even for a model trained before your business existed.
A model trained last year knows nothing about your new service — but at inference it retrieves and reads your current page, so it can still recommend you today.
Why this matters for AI findability
Inference is where retrieval, grounding, and citation happen — the live window in which your content can shape an answer. Optimising for that window is the practical core of agentic findability.