The Content Studio: a responsible AI content pipeline

The problem

The client — a UK specialist ecommerce retailer (name withheld under client confidentiality) — needed product-education content at a pace a small team couldn't write manually, in a category where product claims carry both trust and regulatory weight. Scaled AI publishing was exactly the wrong answer.

The system

I built the Content Studio, a pipeline that treats AI as a drafting and checking tool inside a governed process:

1. Source ingestion — the first job is building the source-of-truth corpus: the client's existing website is downloaded page by page, and the supplier PDFs (specifications, certificates, safety documentation) are collected alongside it. Everything the pipeline writes later must trace back to this corpus — AI is never allowed to be the source of a fact.
2. Keyword fetch — live Google Keyword Planner data for the UK via the DataForSEO API, seeded from each product.
3. Keyword filter — an AI analyst step that discards irrelevant terms, picks one primary keyword and groups supporting and long-tail terms. It includes variant-cannibalisation logic: if the product is the organic or refined variant, the page targets the variant term rather than the highest-volume generic — so pages never fight each other.
4. Draft — structured drafting against a strict content schema, grounded in the ingested source material.
5. Claim audit — every factual claim is checked against the supplier's source documents from stage one. Unsupported claims are flagged and removed.
6. Independent quality review — a different AI vendor (GPT) reviews the draft it had no hand in writing.
7. Rewrite — revisions with a visible diff.
8. Compliance gate — a final combined AI + rules check before anything is publishable.

The client's side

The client never sees the machinery — they see the Studio: a private portal where finished pages arrive for their approval, behind a simple access code. Approval stays a human act, and the workflow's final gate belongs to the client.

Why it matters

The pipeline embodies a simple principle: AI is never the source of the facts and never the final judge of its own work. Facts come from source documents; judgement comes from a second model and a human. That's the difference between AI content that ranks and AI content that triggers a quality demotion — a line Google's helpful-content systems enforce ruthlessly.

The same method — generalised beyond this client's category — is what I build for clients as a responsible AI content system, and the keyword step is documented in detail in how I use DataForSEO to build keyword maps.