Keyword research has a chicken-and-egg blind spot: tools report what people already type, which reflects what content already exists. The phrasings your market speaks — in questions to creators, in how tutorials title the pain, in the vocabulary of comment sections — run ahead of the typed record. By the time a phrasing shows volume in a keyword tool, someone else owns the SERP.

The spoken layer is the early signal: how people describe the problem before the search engine teaches them the "correct" query. SEO teams that mine it write for phrasings six months before the tools confirm them — which, in content SEO, is the entire game.

deep_search reads the spoken layer: how creators title and phrase the problem space, what vocabulary the how-to content in your category actually uses, which question-formulations recur. Transcript-level search means you're mining real usage, not guessing at it.

chat turns the mined language into keyword strategy — the no-retrieval half of the workflow: cluster the phrasings, generate long-tail variants, map them against your existing coverage. The brainstorm runs grounded in observed language instead of starting from your team's internal jargon — which is where most keyword lists quietly go wrong on day one.