The AI Prompt I Use for VA Claims

What this is

This is the exact project-level instruction set I give Claude (Anthropic's AI) before asking it anything about VA disability claims. It tells the AI to cite 38 CFR by section number instead of guessing, to check VA.gov and eCFR before answering, to flag when sources disagree, and to never estimate a rating percentage without the criteria in front of it. It also tells it plainly where its limits are — it's not a replacement for an accredited VSO, claims agent, or attorney.

Drop it into a Claude Project (or adapt it for ChatGPT/Gemini) and it changes how the AI handles this kind of work: sourced, cited, and honest about uncertainty instead of confidently wrong.

Preview