Case studiesMarketplace
A programmatic SEO engine that scaled a marketplace from 1,200 to 38,000 indexed pages in 90 days
Programmatic SEO + content engine for a B2B marketplace.
“We were three weeks from hiring four contract writers. Instead we shipped an engine that publishes the work of forty.”
The challenge
A B2B marketplace with strong supply (3,400 listings across 80+ subcategories) and almost no organic traffic. Their SEO surface was a small editorial blog updated twice a month. Competitors with similar inventory were ranking for category × geography long-tail queries the marketplace was not even targeting.
The team had budget for one full-time content lead but no realistic path to manually creating the 30,000+ category pages, location pages, and comparison pages they needed to compete on long-tail.
How we approached it
Sprint (2 weeks) — engine architecture. Built a content engine on top of their existing Next.js + Postgres stack. Templates for four page types: category pages (e.g., "[service] in [city]"), comparison pages ("[provider A] vs [provider B]"), educational guides ("how to choose a [service]"), and FAQ aggregator pages. Each template combines structured data from their listings with model-generated narrative.
Pipeline: topic generator (pulls keyword opportunities from Serper API + their internal listing data) → outline drafter (Claude Sonnet 4) → section writer (GPT-4.1, parallel calls per section) → fact-checker pass against the listings table → editorial review queue.
90-day retainer — content reviewer + scaling. A content strategist on our side worked with their content lead to tune voice, build the editorial guidelines, and triage the daily review queue. We added internal-link suggestion (every new page links to 3–8 related existing pages with anchor-text variation), schema generation, and ISR-driven publishing. Backend cron orchestrated the whole thing — generation runs overnight, review queue is empty by lunch.
The outcome
38,000 indexed pages live within 90 days. Organic sessions grew 612% year over year in the same window. Cost per published page settled at $0.41 all-in (model usage + infra + the editorial review time amortized across the volume).
The bigger result was the inbound shift: leads coming through long-tail organic now outpaced paid by 4× and convert at a rate 2.3× higher because the searcher arrives with high purchase intent.
Stack
- OpenAI GPT-4.1
- Anthropic Claude Sonnet 4
- Next.js ISR
- Postgres
- Serper API
- Vercel
Team
Two engineers + content strategist
Handoff
Full code in their Vercel org. Editorial guidelines doc. Runbook for adding a new page template in under a day. Continued at a light retainer for new template builds and quarterly model upgrades.
Your engagement, next on this list