What I'm saying is, it's faster to skim something and recognize errors (edit) than it is to produce it (author). Assuming that there's even a vaguely categorical difference between personal and public, you can probably have an LLM separate these things. Each time you see something wrong, explain why it's wrong, and run it again. If nothing else, it's a valuable experiment in the capabilities of current systems. I've found Claude far exceeds my expectations in tasks like these.