Four things you stop having to check by hand.
Content Brain isn’t faster because it writes more. It’s faster because the things that normally come back from review — wrong voice, invented claims, a service area the client doesn’t cover — are refused at write time instead of found at read time.
Every client, one login
Each client is its own workspace with its own brand records, ideas backlog and drafts. You switch between them from the same account. Nothing crosses over, because the separation is structural rather than a filter someone remembered to apply.
The voice holds after you stop watching
A client's Voice Profile records how they sound in writing and what they never sound like. Every brief carries it, and every draft is generated against the exact version in force when the brief was frozen. A new writer inherits the voice instead of learning it.
You can answer “where did this come from?”
Every number in a research read-out names the API call that produced it. Every finding names the page it was observed on. Nothing is stored with a source it can't produce, so the answer to a client's question is a record, not a recollection.
Half-built clients can't burn budget
A run reads the brand record first and refuses to start unless the profile, the voice, the offers and the customers are all in place. The readiness screen names which record is blocking. You find out before the model spends anything, not after.
Eight screens, in the order the work happens.
No blank canvas anywhere. Each screen hands the next one something structured enough to be checked.
Brand Foundations
The hub over the records every run draws from: Brand Profile, Voice Profile, Offers, Ideal Customers, Locations. It shows a readiness percentage and names the specific condition that isn't met yet.
Topical Map
An entity-driven cluster-and-node hierarchy for the client's subject, with coverage tracked against it. You can see what the site owns, what it half-owns, and what it has never touched.
Gap Analysis
Page-level differentiation across the pages retrieved and cited for a topic. It records three kinds of finding — a missing subtopic, a glossed-over nuance, an unhandled edge case — each naming the pages it's absent from. No priority score, because nobody could re-derive one.
Content Ideas
Scored, evidence-backed ideas from a run against a target site and its rivals, under a credit ceiling you set. Each idea lands in one of three bands — build now, queue, backlog — and carries a lifecycle: accepted, rejected, promoted.
Topic Audit
Sixteen layers of research on one topic: keyword ideas, named entities, co-occurring entities, knowledge triples, aliases and contextual vocabulary, prompt modifiers, community signals, follow-up intent, answer blocks, queries. It exists so no downstream step re-runs research.
Content Engine
One screen that carries a piece from approved idea through audit, Blueprint, outline sign-off and draft. Approval isn't a status flip: it asks which framework writes the piece, which offer it sells, and which location it's about, because the system won't guess any of the three.
Drafts
Typed blocks are what's stored — the editor is just an editor. Revisions, comments, per-section regeneration, approve and publish. A human edit is recorded as a first-class origin, not as an override of the machine.
Tasks & Calendar
Who owes what on the content already in the plan, and the same work laid out on dates. A task points at the object it's about and never holds a copy, so the calendar can't drift from the work.
Seven stages. Each one can stop the run.
A stage that repairs its own input quietly becomes the stage that decided what the input was. So nothing here repairs. It halts, names what’s missing, and waits for a person.
Brand DNA
Record who the business is and the claims it may and may not make, how it sounds and what it never sounds like, what it sells at the granularity it publishes, who it sells to in their own words, where it operates, and which sources it trusts.
HALTS IF ·a required record is missing or incompleteMap, rivals, gaps
Build the topical map, register the competitors that actually rank for the subject, and read the pages retrieved for each topic to find what nobody covers well.
HALTS IF ·a finding can't name the pages it's absent fromContent Ideas
Run ideas against the client's site and its rivals under a credit ceiling. Every idea comes back scored, banded, tagged to one of the four pillars, and carrying the evidence it was scored on.
HALTS IF ·a metric can't name the API call it came fromTopic Audit
Sixteen layers of research on the approved idea, collected once. Re-running writes a new record and supersedes the old one rather than overwriting it, so last week's brief stays auditable against the signals that existed when it was written.
HALTS IF ·a keyword volume has no connector call behind itBlueprint
The brief freezes. It records the exact version of every record it read — profile, voice, offer, customers, location — so the brief stays readable against the truth it was written from, even after that truth changes.
HALTS IF ·a required input for this piece's pillar is absentOutline sign-off
A person approves the structure before a word is generated. That approval might come four days later, and the pipeline is built to wait rather than hold a run open over inputs the client can edit underneath it.
HALTS IF ·the outline doesn't satisfy the frozen briefDraft
One generation call per section, into typed blocks. Then revisions, comments, per-section regeneration, approval and publishing. Your edits are recorded as yours.
HALTS IF ·a brand slot in the copy doesn't resolve to a real fieldA model handed {{brand.nmae}} will answer it with invention. So an unrecognised slot doesn’t render — it stops the run.