AI Wrote My Patent Application

The benefits, risks and potential pitfalls of relying on artificial intelligence to write patent specifications

Using an AI tool to write a patent application, saving the cost of a patent expert, may feel like a great idea, but the consequences could be significant for the ultimate protection of your invention, Andy Flaxman writes.

I used AI to help me write a patent application, so I just need you to check it over and file it

This is something I’ve heard from many inventors and, worryingly, something I’m hearing more and more of.

Since large language models (LLMs) like OpenAI’s ChatGPT, Google’s Gemini and Anthropic’s Claude became readily available, people have started to use them to assist them with everyday tasks, such as writing emails to work colleagues, writing emoji-laden LinkedIn posts (don’t get me started) and, yes, writing patent specifications describing their inventions.

First things first, with many publicly available free-to-use AI tools, there is no guarantee that the information you provide as your prompt (i.e., the text that you provide to the AI model to ask it to perform the desired task) will be kept confidential.

Patents can only be granted for inventions that are ‘new’; that is to say the invention has not been disclosed by anyone, anywhere in the world before a patent application for that invention has been filed. Providing an invention disclosure to an AI tool will almost certainly constitute a public disclosure, meaning any patent granted for the invention could be invalid.

Putting the likely lack of confidentiality aside, AI tools are capable of producing something that, on the face of it, looks like a patent specification, in less time than it takes to find the phone number of your local friendly patent attorney firm.

AI is great at writing in fluent technical English and it can explain your invention clearly. However, the quality of a patent application isn’t measured by how well it reads. Its true value is tested years later, when a third party tries to invalidate it or design around it or when you try to enforce it.

So, what’s wrong with an AI-drafted patent application?

AI doesn’t know your commercial objectives; as patent attorneys, we consider who your competitors are what they do, which features are commercially valuable, how your product and business plans might change over the coming years, and where in the world you need protection for future expansion. AI doesn’t care about these things.

An AI-drafted specification is unlikely to take into account what already exists – the so-called “prior art”.  Your patent application should distinguish over what has been done before and should demonstrate a non-obvious development over it. Human patent attorneys ensure the patent specification is full of useful fallback positions – optional features of the invention that are not known from the prior art.

A human attorney will also use broad terminology to cover not just the embodiment that the inventor has described, but alternative embodiments that competitors might use. AI has a tendency to re-use the wording provided in the prompt but without coming up with synonymous terms that might give broader protection.

A qualified patent attorney doesn’t hallucinate like AI does. Modern AI tools have improved, but they can still generate text that – while appearing convincing – is simply untrue. Including inaccurate information or data in a patent application can lead to problems down the line.

AI tools love to use functional language, like ‘a module configured to optimize…’ without explaining how or why the module does that, or what performs the optimization. This could lead to clarity issues or – worse still – enablement issues, where a patent application is found to be insufficiently disclosed.

Patent attorneys are often experienced in drafting patents for filing in multiple jurisdictions where laws differ. Ensuring from the outset that a patent specification is likely to stand up to scrutiny from examiners from different national Patent Offices is essential if the invention is to have commercial relevance overseas. Once the first patent application has been filed, the scope for amendment is limited and so it is vital to get it right at the outset.

Many patent attorneys agree that AI is an excellent assistant, but a poor substitute for specialist patent drafting expertise. A patent specification drafted by AI is likely to look good and can be included as part of a patent application. It is later on that the lack of experience that has gone into its drafting may come to light and limit options for claim amendments as the application is examined by a Patent Office. If the granted patent is enforced or challenged, then an opposing party will do their best to find irreparable issues with the specification. A human drafted specification is likely to stand up to such scrutiny better than an AI drafted one.

Using AI to draft a patent specification is like asking a very sophisticated DIY website to design the foundations of your house. It can produce beautiful plans, explain every component and give you something that looks remarkably professional. But it doesn’t know the ground beneath your particular plot, it doesn’t know how the building will be used in 20 years, and it won’t be there when the foundations start to crack. By the time you discover the problem, the house may already be built, and fixing the foundations can be impossible.

SH&P use AI, as all leading firms do, but we use it sparingly and only as a tool to assist us. Our professionals have many years of experience and that gets woven into each and every patent specification that gets written.

If you would like to discuss your invention before discussing it with AI then get in touch.

 

Posted: 24th September 2026

 

Written by

  • Andrew Flaxman, Patent Attorney, Partner

    Andrew Flaxman is a partner and patent attorney at Stevens Hewlett & Perkins.