Programmatic content engine for a B2B SaaS category

The problem

The category had hundreds of long-tail comparison and integration queries the team knew converted, but writing each page through a freelance roster cost too much per page and took six weeks per batch. Pages that did get written read inconsistent. The head of marketing wanted volume, voice, and ownership — and didn't want to staff a content team to get it.

Constraints

• Voice guidelines were strict — the brand had spent two years training the market on a specific register, and AI slop would undo it.
• Every page had to be backed by real product data, not generic claims, so a hallucinated feature comparison wouldn't reach publish.
• The marketing team had to own the engine after handover — no per-page invoices, no agency retainer for upkeep.

Approach

1. Weeks 1-3: keyword and intent mapping across roughly 600 candidate pages, then a brutal cut to the 180 that actually mapped to product strength.
2. Weeks 3-6: built a typed content pipeline — structured product data in, brand-voice prompts and templates in the middle, MDX out — with a human review gate before publish.
3. Weeks 6-10: drafted, reviewed, and published the first 60 pages with the marketing lead in the loop, tuning the prompts and the review rubric each week.
4. Weeks 10-13: handed over the engine, trained two marketers to run it, and documented the prompts, the data schema, and the review rubric.

Artifacts handed over

• A programmatic content pipeline the marketing team runs from a CLI without engineering support.
• 180 published pages with consistent voice, real product data, and a documented review rubric behind every one.
• A prompt and template library versioned in the client's repo — when the product changes, the pages update with it.

Outcome

The marketing team publishes pages on their own schedule instead of waiting on a freelance batch. The cost per page is a fraction of what the roster charged, and the voice reads like the rest of the site because the prompts encode it. The two marketers who own the engine spend their time on the review rubric and net-new templates, not on writing first drafts. Rankings followed the work — but the structural change is that content is now an internal capability, not an outsourced line item.

Tech

• Next.js
• MDX
• Claude API
• TypeScript
• PostgreSQL
• Vercel