SEO is a data problem: lessons from building AffRank.ai

For years I managed content sites the way many people do: spreadsheets of keywords, a rank tracker in one tab, the CMS in another, and affiliate dashboards somewhere else. Every update was manual, repetitive and easy to get wrong.

AffRank.ai is my answer to that: an affiliate and SEO management platform that brings ranking and keyword data, metadata workflows, structured FAQ data, affiliate-link management and website integrations into one place. Building it taught me that SEO is mostly a data problem wearing a marketing hat.

Before building anything, I lived in the standard tools: SE Ranking and Semrush for keyword research and rank tracking, and NeuronWriter for drafting content around a keyword. They're good at what they do. The problem was everything between them.

The sources of truth are scattered

A single page's performance depends on data from several systems:

• Search rankings and impressions for its target keywords.
• Its own metadata: title, description, headings, schema markup.
• Its content and FAQ sections.
• The affiliate links on it, and whether they still work and still pay.

Each lives somewhere different and changes on its own schedule. You can't make good decisions until they're joined together.

Model it like data engineering

The core idea behind AffRank.ai is an explicit data model, built on PostgreSQL behind a FastAPI service. Conceptually:

• Sites have pages.
• Pages target keywords, which have ranking history.
• Pages have metadata and FAQ entries.
• Pages contain links, which belong to affiliate programs.

Once the relationships are explicit, useful questions become simple queries: which pages lost rankings this month and haven't had their metadata touched in a year? Which affiliate links point at a program that changed its terms?

Structured FAQs pay twice

FAQ content is useful to readers, and when it's marked up with structured data it's easier for search engines to understand. Treating FAQs as structured records — question, answer, page, last reviewed — instead of loose HTML makes them reusable, auditable and easy to keep fresh.

Automation with a review step

It's tempting to let automation (and AI) rewrite metadata across a whole site. That's how you break a site in an afternoon. The design rule is: automation proposes, humans approve. In practice that means:

• A suggested metadata change should be visible as a diff against the current version.
• Bulk updates should be previewable and applied in small batches.
• Every applied change should be recorded with what changed and why, so it can be rolled back.

That's the same change-control discipline I used in enterprise infrastructure, applied to content.

Measure the effect

Each change carries a date. Rankings and clicks before and after can be compared, which turns "I think the new titles helped" into an actual answer. Not every change will help; knowing which ones didn't is just as valuable.

What I'd tell site owners

1. Put your SEO and affiliate data in one model, even if it's just a well-designed spreadsheet to start.
2. Treat metadata and FAQs as structured data, not decoration.
3. Automate the repetitive parts, but keep a review step and a change log.
4. Measure before and after every meaningful change.

The sites that win long-term aren't the ones with the cleverest hacks. They're the ones that are managed consistently, with good data, over years.