Design a programmatic SEO system for 1 million landing pages

WHAT THIS TESTS: This question tests whether you view programmatic SEO as a data engineering and infrastructure problem rather than a content creation exercise. At one million pages, manual authoring is economically impossible, so the interviewer wants to see a systems mindset that separates structured data from presentation, manages crawl budget, and guarantees performance. They are looking for familiarity with ETL pipelines, rendering strategies, and search engine discovery protocols.

A GOOD ANSWER COVERS: First, data ingestion and schema design. You need an ETL pipeline that normalizes proprietary or third-party data into a strict schema where one row equals one URL. Include validation gates to prevent empty or duplicate records from generating thin pages. Second, page generation and rendering. Use deterministic templates married to structured data, rendered via SSR, SSG, or ISR depending on update frequency. Pre-generate static pages for stable data and use incremental regeneration for volatile slices. Third, URL routing and slug architecture. Implement a hierarchical slug strategy that mirrors category intent, plus a reverse index or lookup table so the router resolves any valid URL in constant time. Handle 301 redirects and 404s cleanly when data rows are deprecated. Fourth, performance and discoverability. Serve pages through a CDN with edge caching to keep TTFB under 100 milliseconds. Generate segmented XML sitemaps capped at 50,000 URLs each and expose them through a sitemap index. Add schema.org JSON-LD, canonical tags, and robots meta tags to manage indexation. Fifth, crawl budget protection. Build noindex gates for pages below a quality threshold, use log file analysis to detect bot traps, and prevent index bloat by excluding near-duplicate parameter variations.

COMMON WRONG ANSWERS: Treating the problem as an LLM bulk-writing task instead of database-driven infrastructure. Proposing on-demand database queries for every page request without caching or edge distribution. Ignoring crawl budget and suggesting all one million pages should be indexed immediately without quality thresholds. Failing to mention XML sitemaps, canonical tags, or duplicate content prevention. Suggesting a monolithic CMS to store one million pages as individual documents.

LIKELY FOLLOW-UPS: How would you handle a data update that affects half the pages without rebuilding everything? What signals would you use to decide whether a generated page deserves indexation? How do you prevent duplicate content when multiple data rows produce similar output? What is your strategy for internationalization or localization at this scale? How would you measure revenue attribution per programmatic page cluster?

ONE CONCRETE EXAMPLE: Zapier programmatically generates over ten thousand integration pages such as Connect Gmail to Slack. Each page is not hand-written; it is a database row rendered through a template that injects specific app names, descriptions, and use cases into a consistent layout. The engineering effort is fixed, but the system captures thousands of long-tail queries that a content team would need eighteen months and three hundred thousand euros to cover manually.

Source: nikoalho.fi