Your Personalization Is Working. Your Deployment Isn't.
For many publishers, the story goes like this: years of investment in a recommendation engine, a sleek "Recommended for You" block on the homepage — and then the results come in. The block is doing great. But total site engagement hasn't moved. Is the technology actually working?
Yes. But you've fallen into the Dilution Trap.
The Math of the Small Block
Imagine a homepage where the average reader visits 2 pages per session. You replace a curated editorial block — one that drives 2% of all homepage clicks — with a personalised selection. Your engine is genuinely excellent: it doubles the clicks on that block. A 100% improvement.
But 100% of 2% is still 2%. Your overall session length moves from 2.00 to 2.02. In a real A/B test, that shift is invisible — lost in the natural noise of daily traffic. To every stakeholder reading the results, personalization looks like it did nothing.
Why Surface Area Is Everything
If a recommendation system could predict the future perfectly, you'd only need to show each reader one article. But readers are unpredictable. You show them multiple recommendations to increase the probability of discovery — and that math only works in your favour if the technology has enough room to operate.
One block isn't enough. Put personalization where readers are actually clicking: article feeds, section fronts, most-read blocks, the top bar. Coverage determines whether you can measure impact at all.
There's also a compounding effect at play: every click creates another chance to click again. That's why a 30% CTR improvement doesn't produce a 30% jump in session length — the two numbers measure different things, but they're driven by the same reading behaviour.
The Motorsport.com Proof
This is exactly what we validated with Motorsport.com. Instead of one block, they deployed personalization across a large part of their homepage and article pages — covering up to 30% of all site clicks.
With enough room to influence the reader journey, session length didn't move by a fraction. It increased by 20%. And for registered subscribers, the lift was even stronger.
The difference between 20% and statistical noise wasn't a better algorithm. It was coverage.
What This Means for You
Personalization is a force multiplier. A multiplier applied to a small number stays small.
The question isn't "does our recommendation block have good CTR?" It's: what share of our readers' journey are we actually influencing?