I built Ejemplode. Then I had to rebuild it.

In late 2006, I started exploring how websites could make money. I had already been building websites for years. This time, I wanted to build something of my own.

I built a keyword research tool around Yahoo Overture. It took a search, found related terms and used those to uncover more searches.

I was looking for demand. What were people trying to do? What did they need help with? Where could a small website find an audience?

I tested different niches. Hotels, insurance and mortgages looked like attractive markets. But smaller sites about letters, memorandums and language did better than those bigger bets.

People needed an example.
I could help them find one.

A recommendation letter. A memorandum. A grammar rule explained with words they recognised.

These were specific needs, repeated by many people. I could see the opportunity in the searches and in the response to the sites.

That was the start of Ejemplode.

I launched it in 2007. The idea was simple: make things easier to understand through examples.

Not every experiment deserved to become a business.

Some of my early experiments used automated or scraped content. Rankings disappeared quickly. Keeping those sites alive took more work than the model justified.

That experience pushed me towards original educational publishing. I wanted a business I could develop over time.

I built the proprietary CMS and the tools behind the publishing operation. I also handled the search strategy, editorial direction and advertising revenue.

Those responsibilities were connected. A publishing decision affected search. A technical decision affected what the team could do. Traffic affected what the business earned.

I had to understand all three.

A team that kept learning.

Over the years, I hired and trained many writers. At the busiest point, up to eight worked with me at the same time.

The team changed. People joined, learned and moved on. Training was a continuing part of running the business.

I taught web writing and SEO, reviewed the work and helped writers connect a search query with a useful answer.

The work went beyond choosing keywords. An article needed to explain something clearly, use examples well and fit into the rest of the site.

From one site to a publishing network.

The wider network grew to more than 30 monetised sites. My publishing work included Spanish, Dutch, Portuguese and an early English project.

Different topics and languages gave me more opportunities to test what worked. They also added editorial and technical work.

AdSense was how the audience became a business. Page views mattered because they supported advertising revenue. They were part of the commercial model, not just a number on a dashboard.

I was responsible for the platform, the content, the team and the result.

Recovering from Panda was not the end of the story. There were years of growth ahead.

Ejemplode went from small niches to competitive search terms. The same approach that started with specific questions grew into broader educational coverage.

I kept working across the content, the platform and search. The business grew around that combination.

On 10 November 2015, the daily AdSense report recorded 1,414,368 page views.

Over the years, Ejemplode alone passed 1.1 billion page views and 4.2 billion ad impressions.

I had built a business around finding demand, publishing useful answers and turning readership into revenue.

Those figures belong to Ejemplode alone.

The other sites were a separate part of the operation. Their traffic is not added to the figures shown here.

By 2026, I was not looking for a new homepage. I needed to rethink the content and the platform together.

There was thin content. Pages competing with each other. Weak connections between topics. A CMS built for a different era of search.

There were thousands of articles to review and years of decisions to work through.

Codex made that volume of work possible.

The scale of the work had made a full rebuild difficult to take on manually. With AI supporting the work, I could approach the whole archive and the platform as one project.

I set the direction, reviewed the changes and decided what to keep, combine or remove. Codex helped me carry out the work.

This was a different use of AI from simply generating more articles. I already had a large archive. The job was to make it stronger.

First, decide what deserves to stay.

I worked through overlapping articles and weak pages. Where several pages were trying to do the same job, I consolidated them.

The editorial inventory went from roughly 3,800 articles to 2,800.

The smaller number was a consequence of the decisions, not the objective. I wanted clearer answers, less duplication and a stronger purpose for each page.

I also worked on the explanations, examples and evidence of expertise. Publishing an answer was not enough. The content needed to give the reader a reason to trust it.

Then, make the answers work together.

I rebuilt the connections between topics, pillar pages and supporting articles.

The question was simple: once someone has read this, what would help them next?

That led to learning routes, clearer navigation and internal links based on the subject. It also changed how I thought about the archive: as connected material, not thousands of separate destinations.

Rebuild the platform around the work.

I migrated the custom CMS to WordPress. The old system had supported years of growth, but its design and limitations had become part of the problem.

The move was part of the larger rebuild: content, structure, presentation and practical features changed together.

I added tests and calculators so the experience could do more than present an article. A reader could use the material, practise or work through a calculation.

Clicks and impressions rose after the rebuild, with stronger daily peaks from late August.

The site is still far below its historic scale. The first signs are encouraging.

Visibility in a changing search experience.

The rebuild also had to account for AI-powered search. The content needed to be clear, connected and useful wherever a reader discovered it.

I see GEO as part of that work. It is another way to think about discovery, without losing sight of what happens after someone finds the answer.

The Google AI Overviews visibility report below recorded 1.41M impressions across the July–October 2026 reporting window.

A recovery signal, not the finish line.

Search visibility is moving in the right direction. That is useful evidence after a period when smaller interventions had failed to change the picture.

But visibility, visits and revenue are different things. I do not treat an impression as a visit or a rising graph as proof that the business has fully recovered.

I also changed several things together. These charts show what happened after the rebuild. They do not isolate the contribution of every individual change.

The next test is whether that attention becomes a business again.

The modernisation started with Cartas de Recomendación, then CartaDe and 10ejemplos, before Ejemplode.

These sites share a history, but they do not all need the same product. Someone preparing a letter has a different task from someone learning a grammar rule.

What does the reader need to get done?