The U.S. Patent and Trademark Office is seeking information on an AI-driven image search tool to assist patent examiners, according to a contract opportunity document posted Wednesday.
The agency said in its application that the tool will be used to “improve search completeness and examiner efficiency,” particularly in evaluating existing global patents and references, known as “prior art.”
“As the volume of global intellectual property applications increases, relying solely on text-based or traditional classification searches is no longer sufficient, especially for visual design and high-utility applications,” the RFI said. “This AI-driven tool is intended to enhance, not replace, the existing USPTO enterprise search system.”
Data shows that the USPTO receives thousands of patent applications each year, with a total of about 1.26 million pending as of last month. According to an April press release, the agency has approximately 774,000 unexamined applications, and fiscal year 2026 will be the first time in nearly 10 years that the agency’s number of applications exceeded the number of applications in a fiscal year.
“Although it may not seem like a big deal, this milestone is important because we have reached a turning point in momentum in applicant’s favor,” USPTO Director John A. Squires said in a release. “We are steadily releasing robust AI tools to support our proctors, and we are halfway through our aggressive hiring target for proctors. That’s why I’m bullish.”
According to USPTO data, it takes nearly two and a half years for a patent application to reach a final decision, and the office has nearly 8,800 patent examiners, not including design patent examiners.
Aaron Capron, a partner and co-leader of the AI technology tools committee at intellectual property law firm Finnegan and a former patent examiner, said he is not surprised that the U.S. Patent and Trademark Office is looking for a new vendor for this. The main reason for this is that some large-scale language models have problems crunching numbers, and the agency will need something more bespoke.
“Each application has different numbers, and the way people create the numbers is different, so there’s probably more variation,” Capron said. He said this is a starting point and that it would help if the agency “could be as transparent as possible about the process.”
The request for information states that the tool must be highly accurate, scalable, and approved by FedRAMP. It takes images directly as queries and should be “highly configurable” to “significantly improve” patent quality.
Responses to requests are due by noon on June 19th.
