The point about running the same checks every few weeks and writing down what changed is the part most people skip. When you ask ChatGPT or Perplexity the customer's question and get a different answer each run, how many runs do you do before you trust the pattern?