USPTO pushes AI to cut patent backlog, but examiners say it adds work

The U.S. Patent and Trademark Office is pressing staff to use internal AI tools to shrink its patent backlog. But some examiners say the technology adds to their workload rather than lightening it, Bloomberg Law reported November 13—a challenge to the agency’s effort to speed decisions with an already strained workforce.

The agency says AI can help examiners check whether applications overlap with existing patents and summarize accompanying documents. Some examiners, however, say the tools fall short of being useful and cannot replace their methodical searches. The disagreement concerns whether AI removes work from the examination process or adds another task.

An AI search does not finish the search

A November 5 Government Accountability Office report gives more detail on that friction. Stakeholders told the watchdog that the agency’s AI tools do not yet reliably return useful patent search results. Its account identifies two tools with different jobs:

• Similarity Search supplies potentially similar U.S. and foreign patent documents for examiners to review.
• SCOUT is a broader AI platform for general questions, problem-solving, coding and analysis.

The Patent Office Professional Association, the examiners’ union, told GAO that mandatory use of Similarity Search increased search time. Its results were incomplete, the union said, so examiners still had to conduct full searches. GAO also noted that verifying AI search results takes time. Those complaints describe an additional review step, not a replacement for the underlying investigation.

That investigation is central to patent examination. Examiners must assess whether an invention meets legal requirements, including novelty and nonobviousness. They search prior art—information about the invention’s subject matter available before the application was submitted—to help make that judgment.

About 756,000Applications awaiting an initial decision

The backlog stood at about 756,000 applications in August 2026, down from about 838,000 in January 2025.

20.6 monthsAverage wait for an initial decision

Applicants waited an average of 20.6 months for an initial decision in August 2026, according to USPTO data cited by GAO.

More output, less training

AI is one part of a wider backlog-reduction effort. The USPTO is also requiring examiners to review more applications and has reduced training, GAO found. Stakeholders warned that higher workloads and the elimination of time for subject-matter training could hurt examination quality. The union also said administrative duties had been added without time to complete them.

Staff retention is another constraint. The examiner attrition rate rose from 2.9% in fiscal 2020 to 7.6% in fiscal 2025. GAO’s model found that, even with examiners doing more work, the agency was unlikely to bring the backlog below 599,000 applications before 2031 without reducing departures. The USPTO set that backlog target in September 2025 without a deadline.

GAO based its review on agency data and 36 interviews with examiners, former officials, patent attorneys, companies, trade associations and researchers. The examiners interviewed were union members. The USPTO declined requests for information during the review, though it commented on a preliminary statement of facts; GAO incorporated those comments as appropriate.

For applicants, a smaller queue has not yet meant faster final answers. Average waits for a final patentability decision increased 11% between January 2025 and August 2026, GAO found. The watchdog recommended that the USPTO assess and address the risks its backlog strategy poses to examiner retention and examination quality.