Pronunciation Dictionary

Context

Once Letras Academy existed, an idea came up in the company’s idea bank: tap into the SEO potential of dictionaries and the behavior of people who were already learning a language through Letras translations and arrived wanting a specific word, its translation and, above all, its pronunciation. That’s how the Pronunciation Dictionary was born, in 2021. With the team buried in other projects, I designed the feature on my own.

Challenge

Deliver something that felt like a dictionary, focused on a word and its translations, but that went beyond the ordinary dictionary: actually teach, and teach through music, which is what makes Letras unique. All of it on a page that had to exist at scale, thousands of words, to capture search, without each page turning into empty content.

Process & decisions

Music as the differentiator, backed by our own data. What sets this page apart from any dictionary is seeing the word used in song excerpts and being able to hear each one. That came from an asset we already had: the subtitle database, which stores the exact moment each word appears in a song. Whenever we could use our own data, we did, because it’s more reliable and it’s our brand. The page also offers pronunciation in different English accents, variations, synonyms, and the word used in example sentences, with audio.

Mass generation with AI, quality first. This was 2021, the early days of generating content with AI. For the content that didn’t come from our own data, I used generative AI, testing different clients and models and running a few rounds of curation to refine the prompt before generating at scale. The rule was simple: there’s no quantity without real quality. In a pronunciation dictionary, one wrong piece of information breaks trust, so curation came before scale, never after.

Outcome

At its peak, the dictionary passed 30,000 visits a day. And it wasn’t idle traffic: more than half of that audience went on to Letras Academy’s offers, fulfilling the entry-point role that motivated the project. It’s still live. It was also an early, concrete lesson in generating content at scale with AI without giving up quality, at a moment when that was still new.