AI Content Pipeline With Named Human Sign-Off

Six to eight weeks and your team publishes AI-assisted pages that carry a name against every factual claim in them. The unit of review is the claim, not the article: a 900-word explainer with eleven checkable statements gets eleven sign-offs, each stamped with a reviewer and a source URL in the CMS record itself, where an answer engine and your own legal team can both find it. Why it is built this way: Answer engines cite what they can attribute. A paragraph carrying a figure, a date and a named source gets pulled into a generated answer; a paragraph of adjectives never does. Everything below follows from that, and so does what I refuse to build. What the build covers: Grounding first. Your product documentation, rate cards, regulator pages, returns policy, the support tickets your team actually answers and the pages that already earn citations go into a pgvector store with source URLs and expiry dates attached. The drafting step retrieves from that before it writes. A model that cannot find a grounded source for a sentence is instructed to leave a gap for the reviewer rather than fill it, and the gaps are the point. Then routing. Drafts land in your CMS, with a status field, a reviewer assigned by topic, and a Slack thread carrying the claim list. Reviewers approve or reject claims individually. A rejection sends back the specific claim, not the whole draft. Then provenance. Each published URL gets a machine-readable record: which model and version drafted it, which sources were retrieved, who approved which claim and when, and an author entity in Schema.org markup that resolves to a real person with a real byline history. This is also what makes serving an llms.txt file worth doing; without it you are advertising pages nobody can verify. Then observability. Prompt traces, retrieval hits and reviewer turnaround go into Langfuse so you can see where the pipeline stalls. In every build so far the bottleneck has been reviewer capacity, not generation. What is not included: I do not write your content, now or later. There is no drafting retainer waiting behind this. I do not run your editorial calendar, commission freelancers, or take an SLA on publishing volume. Images, video and social distribution are out of scope. I will not build a step that invents citations or presents a model draft as an expert's authored opinion. Who this is not for: If nobody in your company can put their name to a statement about your own product, pricing or compliance position, there is no pipeline to build here, and buying one only makes that visible faster. Appoint or hire that person first. Equally, if the goal is several hundred pages a month with a light editorial touch, we will disagree in week one, and it is cheaper to disagree now. How we work: Kickoff is a two-hour session with the person who will own the pipeline after I leave, plus whoever owns your CMS. I work in your repository and your CMS from the start rather than building elsewhere and migrating in. Handover is a walkthrough with the owning team, the repository, a runbook of failure modes, and a documented path back if you decide to change models. English or Arabic; the Arabic path is treated as its own review chain rather than a translation afterthought. I keep no standing access after handover.