How to document inventorship when Cursor, Copilot, or Claude helped
Short answer: Record what you conceived — prompts, rejections, design tradeoffs, and benchmarks — not just the final AI-generated file.
USPTO guidance treats AI as a tool; patents require a natural-person inventor who meaningfully contributed to the conception of the claimed invention. When you build with Cursor, GitHub Copilot, or Claude, the risk is not “AI bad” — it is thin documentation that cannot show your human role if challenged.
What to log (weekly is enough)
- Problem statement you wrote before any model output.
- Prompts and edits where you steered architecture, data structures, or APIs.
- Rejected suggestions — AI proposals you discarded and why.
- Benchmarks — latency, accuracy, cost, error rate tied to your design choices.
- Commit messages that explain non-obvious tradeoffs (not “fix bug”).
Evidence that helps in a coaching session
Patent PreCheck’s paid review accepts supporting uploads (screenshots, sketches, logs) processed for your coaching report — not stored as raw evidence files long-term. Use them to show conception moments: whiteboard photos, before/after metrics, email threads debating approaches.
What not to rely on
- “The AI wrote most of the code” without your directional input.
- Generic prompts (“make this faster”) with no recorded human insight.
- Claiming inventorship on boilerplate the model produced unchanged.
Score human conception strength
Our free tier includes a human conception signal alongside patentability pillars. Paste your module at the free analyzer, then strengthen weak sections in the Interactive Code Review. Deep dive: USPTO AI inventorship rules and is AI-generated code patentable?