How much do optimized tags and internal links affect SEO?
E-commerce·2025

How much do optimized tags and internal links affect SEO?

+16%uplift above forecast
75 daystest period
―――

context

Magazine Luiza has several content clusters, but in this project we worked on a single one, split into two smaller subgroups by product. We applied two optimizations: improving title and description tags on pages with a CTR below the site’s own average, and restructuring internal linking through vector similarity. The question was whether these changes actually worked, or whether what we were seeing was just seasonality or natural traffic variation.

―――

challenge

Any change made at scale on a large e-commerce site can be mistaken for noise. The challenge was to tell causation from coincidence: to prove, with statistical rigor, that the optimizations were responsible for the observed improvement, and not the calendar, the market or any other external factor.

―――

approach

Data analysis and processing were done with Python and applied statistics. First, we identified pages with a CTR consistently below the site average and rewrote their title and description tags to increase relevance and appeal in the search results.

In parallel, we used vector similarity to structure internal linking. The technique represents each page as a vector of meaning, generated from its content and keywords, and creates links between semantically close pages. Instead of linking by category or by hand, links follow the closeness of meaning between pieces of content, which tends to reinforce the cluster’s topical authority for Google. Dozens of pages pointed links to the pages eligible for the test, all with real context between source and destination.

To measure the impact, we set up an A/B test with two groups: 10 PS5 URLs as the variant (the optimized ones) and 9 Xbox URLs as the control: the same page type, the same search intent and an equivalent product in the market. We used 90 days of history to train a linear regression, which generated daily traffic forecasts for the variant group. Actual performance was compared with the forecast, and the difference was evaluated statistically to confirm whether there was causation.

―――

outcome

-5%0%+5%+10%+15%+20%15/abr22/abr29/abr06/mai13/mai20/mai27/mai03/jun10/jun17/jun24/jun01/julrealprevistouplift positivouplift negativo

The variant pages performed 16% above what the model forecast for the period. The green areas in the chart represent the days when the optimized group beat expectations; the red areas, the days when it fell short. The model controls for seasonality through the ratio between the two groups, so what shows up in the delta is the effect of the optimizations, statistically validated.

―――

learnings

SEO without statistical validation is opinion. With the right method, you can prove that an intervention worked, or rule out what didn’t. In the end, everything becomes a result that feeds the strategy in some way.

In collaboration with

Mardem Reifison

SEO Manager at Magalu

―――author
Lucas Cassapula

Lucas Cassapula

Partner & Head of SEO at Wesearch and Co-founder of Mentionflow

I am a partner at Wesearch and co-founder of Mentionflow. I have worked with SEO for almost 10 years. I am a data-driven geek who is always testing hypotheses, looking for patterns and turning ideas into products. I share studies, experiments and automations focused on SEO and GEO.