AI Use at the USPTO

Artificial intelligence (AI) pervades many industries. As such, it is no surprise that under U.S. Patent and Trademark Office (USPTO) director John A. Squires’ direction, AI continues to pervade the USPTO to assist in pre-examination and examination processes of patent applications.

In September 2022, the USPTO launched the Similarity Search Tool (SimSearch)[1] to help Examiners with prior art searches and integrated the AI tool within the Patents End-to-End (PE2E) search platform. SimSearch uses only the application text to provide a ranked list of prior art. This begs the question: why are the figures of the application not used? After all, everyone knows that a picture—or in this case a figure—is worth a thousand words. This is especially true in the mechanical and chemical arts, where a figure can clearly illustrate a claimed machine and composition of matter that may be challenging to describe in words. Are SimSearch’s results thus reliable? Further, use of only application text implies that SimSearch cannot be used for design applications that claim the design illustrated in the figures themselves, though Examiners could benefit from such a tool.

In July 2025, the USPTO launched DesignVision[2] to aid in the examination of design applications using image-based searching. DesignVision searches design patents, registrations, trademarks, and industrial designs to provide a list of images similar to those disclosed in the design application. However, it is unlikely that DesignVision could be expanded and used to address some of SimSearch’s limitations noted above because of the differences in utility applications and design applications, where the former focuses on function and the latter focuses on ornamental design.

In October 2025, the USPTO launched the Artificial Intelligence Search Automated Pilot (ASAP!) Program[3],[4] that provides an AI-generated list of prior art prior to issuance of the first Office Action. However, it is unclear if SimSearch was used for this program or if there was a benefit to this program as little information has been published following its closure. Because only 76 of the 169 petitions under the program were granted, it is likely that the program was closed because of lack of interest and participation.

More recently, in March 2026, the USPTO launched the Trademark Classification Agentic Codification Tool (Class ACT)[5] to aid in the pre-examination process of trademark applications, such as assigning international class codes, design search codes, and pseudo marks to unclassified trademark applications. In doing so, Examiners can better, and more quickly, perform prior art searches that may otherwise exclude these previously unclassified trademark applications.

Currently, however, the above AI tools are confined to the pre-examination process and prior art searches, though the examination process is much larger. It naturally follows that the USPTO will eventually launch AI tools to help Examiners understand the technology disclosed in an application and determine written description, enablement, double patenting, patent eligibility, anticipation, and obviousness. Of course, building such AI tools is challenging because each of these is complex and some are subjective.

Instead, the USPTO issued a satirical press release in April 2026 launching the McConaughey Agentic Tasking Technology Helping Examiner Workload (MATTHEW)[6] to aid Examiners in determining patent eligibility to make fun of Matthew McConaughey’s recent trademark challenges and the current state of patent eligibility. The press release joked that Squires said that “MATTHEW will greatly enhance our ability to make the close calls . . . as I herewith also suspend all applicable precedent, including Desjardins, Alice, and Mayo” and that “we had some concerns that we would be introducing a three-part test in place of the two-part test under Alice and Mayo.” [7],[8]. SCOUT is an agnostic platform that allows a user to select and use a generative large language model (LLM) from a menu of LLMs including Anthropic’s Claude, Google’s Gemini, Meta’s Llama, OpenAI’s GPT, and various versions thereof. This appears to be the USPTO’s first use of generative AI. As the name suggests, SCOUT aids the Examiner in summarizing an application to facilitate understanding of the technology. SCOUT also considers whether there are § 112 issues, makes recommendations based on the Manual of Patent Examining Procedure (MPEP), and analyzes software code within an application, though these features are currently only available to administrative staff and not available to Examiners yet. Though SCOUT is currently constrained to the above functionality, it is likely that SCOUT will replace both SimSearch and DesignVision and become the single tool for AI patent examination that also assesses written description, enablement, double patenting, patent eligibility, anticipation, and obviousness.

Implementing SCOUT and future developments thereof into the examination process offers several potential improvements. Using SCOUT, Examiners could issue first Office Actions on the merits faster and with stronger rejections because they better understand the technology disclosed in the application. This not only supports compact prosecution but could also reduce the time it takes to prosecute and thus reduce costs incurred by both the USPTO and Applicant. Strong Office Actions could also reduce the need for examiner interviews in which an Applicant details the invention to the Examiner because SCOUT has already provided a detailed technical explanation.

Though none of the USPTO’s AI tools are used by practitioners, practitioners should be aware of such AI tools and further developments thereof when drafting applications and reviewing and responding to Office Actions, and should keep in mind at least the following considerations.

Assuming, for now, that SimSearch remains in use, practitioners should take care when drafting utility applications to provide detailed descriptions of figures rather than relying on the figures alone since SimSearch does not consider the figures in its prior art search. Practitioners should also take care when drafting design applications as recycling and altering figures from other design applications could disadvantageously cause these other design applications to be relied on in a rejection. Practitioners should further take care to understand the technology, at both the drafting and prosecution stages. Though Examiners using SCOUT could make what initially looks like a reasonable rejection, there could be a misunderstanding between what the utility application discloses or claims and what the prior art teaches. Practitioners should also take care to draft strong applications that can withstand SCOUT’s potentially more rigorous examination that may become available to Examiners in the future. It seems unlikely that applications primarily drafted using AI, that include all the buzz words but little content, will pass muster.

Further, during prosecution, practitioners should take care to understand the Examiner’s position in an Office Action and what aspects of the prior art supports their position, especially with respect to obviousness rejections. Though anecdotal, there appears to be a surge in Examiner’s relying on five or more prior art references to support an obviousness rejection. This could be coming from Examiners over-relying on SimSearch (or, in the future, SCOUT) to mix and match teachings from among the ranked list of prior art. However, it seems implausible that a skilled person would rely on such a large number of references to arrive at a claimed invention due to a lack of “express” motivation to combine. See MPEP § 2145.X.A. Accordingly, practitioners should consider making such arguments when responding to such Office Actions. Practitioners should also take care to ensure that any § 112 rejections or interpretations are reasonable and that the MPEP is being applied properly since SCOUT, rather than Examiners, may be making such initial determinations in the future.  Practitioners should also follow SCOUT’s development, what features of SCOUT Examiners are granted access to, consider any implications associated with those features, and adjust to those features to the greatest extent possible.

While we have reviewed the use of AI tools at the USPTO, AI tools are also being used to aid the practitioner in drafting and prosecuting applications. Both practitioners and Examiners, especially newer ones, should be careful to avoid becoming overly reliant on these AI tools. This is an incredibly difficult task because both Examiners and practitioners are being asked to do more work in less time. As the situation exists today, however, the combined technical and legal knowledge of seasoned practitioners and Examiners is irreplaceable and not something AI tools will likely be able to fully replicate in the near future.

[8] https://rev-vbrick.uspto.gov/#/videos/57cb6766-ca38-4220-a321-b44da08ccbbb