We benchmarked Claude Opus 4.8, GPT-5, and Gemini 2.5 on AEO writing tasks to measure hallucination rates. See the data, our methodology, and how to run your