By: Rob Hendry

Over the past year, clients have increasingly sent me documents – invention disclosures, patent prior art analyses (reviews of third-party patents), and even full patent applications – that have been drafted using artificial intelligence. Some have benefited from inventor revisions, but most are passed along as is. So far I’ve seen some very good invention disclosures, while the prior art analyses are hit-and-miss. These analyses use lots of pseudo-legal lingo, but they often get things just plain wrong. In one recent analysis I received, the AI said one old patent was the closest to the invention while another old patent was irrelevant…when it was actually the other way around once we reviewed the prior art ourselves. Clients will often admit up front that the document was AI-generated and we might not find it helpful at all – a fairly healthy view, I think. Caution is a virtue.

In the case of patent applications, though, the message (sometimes implied, sometimes explicitly stated) is often that this is some solid stuff, and it should only need a soft editing touch before filing in the patent office. The AI even drafts full claim sets, which look impressive even to clients with considerable patent experience. As patent agents and patent attorneys with ethical obligations to our clients, though, we need to look for the errors and inadequacies which are often glaringly obvious (that’s why you pay us, right?). Inventors are normally receptive to our comments flagging such issues, but they’re still surprised that there are any errors at all. The widespread view is that AI should easily generate such documents, based on untold numbers of old patents that have been used to train it.

Clients ask me why the AI isn’t nailing the patent applications, even as an initial draft. For example, how can AI sometimes get the claims – which define the monopoly you’re seeking – so horrendously wrong? The starting point for me, and it should really be for all of us, is that artificial intelligence is not “intelligent”. It is not a sci-fi computer that has achieved sentience. You, the inventor, are smarter than the AI. I know, I know, it’s blasphemy to even suggest such a thing in 2026 as we’re all repeatedly assured that AI will be running everything soon and we can fall back to a life of leisure (talk about sci-fi…). You, the inventor, are smarter. Hold, and keep, the reins.

AI generates documents based on character/word/sentence patterns that it calculates are correct based on billions of other documents it was trained on. That doesn’t mean that the AI is intelligent, just that it can access lots of content and generate output according to the algorithm and the prompts it was given. As we all know, AI can get things wrong, badly wrong at times. It can word things so as to flatter us (never good in legal documents) or even make stuff up entirely. I hate to use the word “hallucinate”, as it propagates the anthropomorphizing of AI, but these models frankly do generate a lot of nonsense that looks technically correct until you review it. The same is true with patent applications, where AI is again generating a word collection that superficially looks right, but it leaves patent professionals shaking their heads.

The standard answer for bad AI output is “blame the prompts”, the old garbage-in garbage-out argument that hides more than it reveals. “I just need to tighten up my prompts, the result will be better next time.” Sure, maybe, that sometimes is the case. Often, though, the AI just does not have the ability to generate exactly what you’re looking for. In terms of patent materials generated by AI, this is where inventors and their advisors need to step in. AI can save you time, and generate a highly polished document, but the inventor is the subject-matter expert on the invention.

Inventors, you need to believe in yourselves. I’ve been doing this patent gig for over 25 years now, and some of the most brilliant, creative individuals I’ve ever met have been clients. By all means, use AI to gather and present your invention, but in my experience it will take your patent agent unnecessary time (and therefore money) to review and revise an AI-generated application, or to review it and ultimately advise that it’s beyond saving. We do have a duty of loyalty to you, and we’ll tell you when the cause is lost. If you do decide to generate your own patent application using AI, though, go for it, but make sure to review the document and revise it based on your own understanding of the invention. Inventors can draft patent applications themselves; the tools are available, have at it. But don’t expect AI to generate something that is ready for filing. Remember where the real intelligence resides: in the mind that conceived of the invention. Use all the tools at your fingertips to explain your invention and try to define what you think is inventive, but it’s your invention, your story to tell, and we must hear your voice.

I am no Luddite. I’ve dedicated my professional life to supporting technology development and the inventors who make it happen. I believe in technology and what it can do for us. However, I do have what I think is a healthy skepticism, and AI is one area where the online and offline chatter is so intense and so insistent that a wise person should pause and ponder. The technology may hit a ceiling and be a flash in the pan, or it may develop towards intelligence or even self-awareness. Who knows? This is exciting stuff, and we don’t know what the future will hold for AI technologies and their ability to help humanity. There are also AI products specifically developed to help patent agents draft applications, and the learnings from those may filter out to public AI programs.

Consider using AI for what it does best: taking your random mad-scientist thoughts and cobbling them together into a coherent narrative. It can help you explain the invention, and that’s definitely helpful for your patent agent. But then review the output before sending it on. I suspect you’ll catch a few things that look wonky. You’re the intelligence behind the invention, after all. Use AI for what it does best, and that requires figuring out what it is and isn’t, what it can and can’t do, and play around with it. It’s a tool, and maybe it will help, but understand if you’ve got a hammer or a toaster before you engage.

I don’t believe that AI can replace the inventor, but what inventor ever shied away from a potentially useful tool? Just remember that you are the mind behind the invention, and you can and should review and revise AI output relating to your invention. Doing that can save lawyer time and your money, which should be important for every inventor looking to introduce their innovation to the world.

Rob Hendry is Senior Counsel with Procido LLP, a lawyer, patent agent, and trademark agent. If you have questions or comments arising from this article, please reach out to Rob or our other IP lawyers, Craig Zawada and Dana Lucas.

Disclaimer

This publication is provided as an information service only and may include items reported from other sources. We do not warrant its accuracy. This information is not meant as legal opinion or advice. Contact Procido LLP (www.procido.com) if you require legal advice on the topics discussed in this article.