Programmatic SEO is not new, but AI has made it indistinguishable from human-written content at scale. In this case study, we walk through exactly how we built a system that generated 100,000 unique, indexed landing pages for a SaaS client — resulting in 340% organic traffic growth and £1.2M in attributed pipeline over 12 months.
The Problem: Competing in a Saturated Market
Our client was a B2B SaaS tool competing in the project management space — arguably the most contested category in software. Their domain authority was 34, their competitors were publishing 50+ articles per week, and they had a content team of two people. Traditional SEO advice said: "create better content than your competitors." With their resources, that was a fantasy.
The insight that changed everything: there were over 180,000 unique combinations of their core product features, user roles, and use case contexts — each representing a genuinely different search query cluster. No one was targeting the long tail. Most tools had templates for location-based landing pages or comparison pages, but nothing for the hyperspecific feature × role × use-case combinations our audience was searching for.
The Architecture: n8n + GPT-5 + Next.js
We built the system in three layers. The data layer was a structured spreadsheet of 800 seed variables: feature names, job titles, industries, use cases, and pain points. n8n combined these into unique permutations and generated structured content briefs. GPT-5 then consumed each brief to produce the page content — not just filler text but genuinely useful, specific content that addressed the exact combination of variables.
The front-end layer was a Next.js dynamic routes setup with generateStaticParams generating all 100,000 pages at build time. Each page had a unique URL structure (/for/[role]/using/[feature]/in/[industry]) and pulled its content from our database of AI-generated copy. The critical quality gate was a GPT-4-based evaluation step that scored each piece of content for specificity, accuracy, and usefulness before it entered the production database.
The Viral Loop Component
Here's where it gets interesting. Each landing page included a prominent "Share this workflow" section that generated a personalised referral URL — and a "Generate a page for your use case" interactive widget. Users who converted from organic search could customise and share their own version of the page with colleagues. This created a viral coefficient of approximately 0.4 — meaning every 10 organic visitors generated 4 additional referred visitors.
Over 12 months, this viral component added approximately 28,000 additional visitors per month on top of organic search growth — at zero additional cost.
What Made It Work (And What Nearly Killed It)
The system worked because of three design decisions: content quality gates, semantic deduplication, and structured internal linking. The quality gates prevented thin or duplicated content from being indexed. Semantic deduplication used embedding similarity to catch near-duplicate page variants before they were generated. And structured internal linking — automatically generated based on semantic relationships — created the site architecture that Google's crawlers needed to understand and index the content efficiently.
What nearly killed it: we initially indexed all pages simultaneously. Google's crawl budget was overwhelmed and most pages sat unindexed for months. The fix was a phased indexing strategy — submitting 2,000 pages per week in priority order based on search volume estimates. This paced approach allowed the pages to accumulate engagement signals before we scaled up, dramatically improving the final indexed/submitted ratio.
Twelve months in, 94,000 of the 100,000 pages are indexed. They rank for approximately 340,000 unique keyword variations. The client's organic traffic has grown 340% year-over-year. And the system is still running — generating new pages as they add product features, and continuously updating existing pages with fresh data from their product changelog.