DEJAN’s Content Optimizer evaluates snippets by establishing baseline positions against competing search results and generating testable hypotheses via an automated seeder. In each round, the engine applies Thompson sampling across structured Rank Factor classes to pick a hypothesis, prompting an ideator model to rewrite or patch the snippet. An LLM ranker then scores multiple output samples to measure rank improvements, continually updating the Beta distribution posterior for each tested claim until the snippet achieves top ranking or exhausts its round budget.