We create an ad auction mechanism for LLM-generated content that ensures truthful advertiser behavior, aligns utility with social welfare, and provably converges to optimal LLM outputs.