How much do optimized tags and internal links affect SEO?

Magazine Luiza has several content clusters. In this project we worked on one cluster split into two product subgroups. We optimized title and description tags on pages whose CTR was below the site average and rebuilt internal linking through vector similarity. The question was whether the visible improvement came from these changes or merely from seasonality and normal traffic variation.

At a large e-commerce site, any change deployed at scale can be confused with noise. The challenge was to separate causation from coincidence and prove that the optimizations, rather than the calendar or market variation, caused the improvement.

We used Python and applied statistics. First, we found pages with a consistently below-average CTR and rewrote their titles and descriptions to improve relevance and appeal in search results.

In parallel, we represented pages as vectors based on their content and target queries, then created links between semantically close pages. This replaced manual or category-only linking with connections based on meaning.

To measure impact, we set up an A/B test with 10 optimized PS5 URLs as the variant and 9 comparable Xbox URLs as the control. A linear regression trained on 90 days of history generated daily forecasts, which we compared with actual performance and tested for statistical significance.

Variant pages performed 16% above the model forecast. Because the model controlled for seasonality through the relationship between both groups, the delta isolates the effect of the optimizations with statistical validation.

SEO without statistical validation is opinion. With the right method, a team can prove that an intervention worked—or confidently discard one that did not.