Founders Use AI to Write Patents, but Cheap Filings Can Backfire

Startup founders can now use AI to produce something that looks like a patent application in minutes. Two patent attorneys say that is where the risk starts. In a guest commentary for Crunchbase News, they argue that a cheap, do-it-yourself filing can look polished while protecting very little.

Why founders are tempted

Patents can help startups gain an edge over rivals, attract investors and protect what they build. But the usual process is expensive, slow and hard to follow, especially for companies with little money. General-purpose AI tools can research existing patents, document an invention and write drafts quickly. Doing it alone can seem like an easy way to save.

Where DIY goes wrong

The authors say AI is uneven. It can do some tasks very well while missing legal and business issues that decide whether a patent is worth anything. The biggest one is the claims. Claims set the legal boundaries of a patent, meaning what you can stop competitors from doing.

Their example: a fraud detection claim tied to one algorithm may leave rivals free to offer the same feature using a different one. Claims that are too broad may overlap with existing technology. Mistakes like these can stay hidden until an investor, a buyer or a competitor looks closely.

What AI is good for

The authors say founders can safely use AI for prep work.

• Organizing technical documentation
• Summarizing existing patents
• Spotting possibly relevant earlier inventions, known as prior art
• Turning engineering notes into an organized invention disclosure

That kind of groundwork can cut costs and let a lawyer focus on strategy.

What a lawyer adds

The authors say the better model is a lawyer directing AI across the whole process, not a chatbot draft that a lawyer reviews afterward. Counsel can also decide what to patent and what to keep as a trade secret. An internal process that competitors cannot see or reverse engineer may be better kept secret. A lawyer can also help set priorities within a budget and check deadlines before a public disclosure or filing limits a startup's options.

Good claims, they say, target what competitors must do to compete. Ideally, one identifiable competitor could infringe them, and that infringement could be detected and proven.

A note on the source

The commentary comes from Michael Drapkin and Nathan Mutter of Landfall IP, which describes itself as an AI-native law firm. They have a stake in the answer. Their advice is to ask a firm how it combines AI with legal expertise, not just whether it uses AI. The piece also says the number of pending applications tells little about commercial value.