Turkevich method is one of the most employed techniques to synthesize gold nanoparticles. Despite its simplicity, the mechanism has been an issue of debate over the past years. The general belief is that particles are formed by a classical nucleation and growth theory, originally described by LaMer's model. In the present work, we provide new experimental evidences that supports either LaMer's theory and their detractors. In the former model, it is proposed that particles are generated by a burst nucleation form the initial 'seeds', from which their growth in a second and quasi-independent step. Instead, our experiments (DLS, UV/VIS and TEM measurements) support the idea that nanoparticles 'seeds' tend to form large intermediate clusters at the beginning of the synthesis, that afterwards disassemble to yield the final nanoparticles. However, unlike other reports, we propose that during the cluster formation the particles do not coalesce, instead they come close to each other without losing their identity. As the synthesis continues, these clusters are progressively separated into the final particles. As a consequence, a path to synthesize ultra-narrow size nanoparticles is provided, along with their stability against salt aggregation, and shelf-time. We found that these ultra-homogeneous nanoparticles are stable for several months, making them suitable for many applications in the biomedical and analytical research.