Key Takeaways
- The USPTO’s AI‑patent classifier shows that AI‑related patent applications face a markedly higher rate of § 101 eligibility rejections than non‑AI filings.
- In 2025, 42 % of AI‑related first office actions received a § 101 rejection, compared with only 6.9 % of non‑AI applications.
- Obviousness‑rejection rates are virtually identical for both groups, hovering near 80 %.
- The pre‑COVID Revised Patent Subject Matter Eligibility Guidance initially cut the AI § 101 rejection rate roughly in half, but that benefit vanished by 2025.
- Non‑AI § 101 rejection rates have remained stable, suggesting the shift is specific to AI technology examination.
- The data do not clarify whether the change stems from applicants filing different types of AI inventions or from evolving USPTO examination practices.
- Practitioners should anticipate heightened § 101 scrutiny for AI claims and consider drafting strategies that emphasize concrete technical improvements and practical applications.
Introduction
In recent years, the United States Patent and Trademark Office (USPTO) has intensified its focus on artificial intelligence (AI) innovations, prompting the agency to develop specialized tools for tracking and evaluating such filings. Dennis Crouch, a noted patent law commentator, leveraged the USPTO’s newly minted AI‑patent classifier to update eligibility statistics through 2026. His analysis reveals a stark divergence in how AI‑related applications are treated under 35 U.S.C. § 101 compared with their non‑AI counterparts.
Methodology
Crouch explains that “a few years ago the USPTO created an AI‑patent classifier that identifies AI related patent applications.” By running this classifier on the USPTO’s public docket, he was able to isolate AI‑containing filings and monitor their procedural outcomes. He notes, “I used that classifier to update activity through 2026.” This approach allowed him to compare first‑office‑action outcomes for AI versus non‑AI applications across multiple years, focusing specifically on § 101 eligibility rejections and obviousness rejections under 35 U.S.C. § 103.
Findings on § 101 Rejections
The core finding is unsurprising yet striking: “No surprise, AI application receive substantially more 101 eligibility rejections than their non-AI counterparts.” To illustrate, Crouch provides a concrete example: “among applications whose first office action came in 2025, 42% of those containing AI have so far drawn a § 101 rejection, against 6.9% of those without.” This nearly six‑fold disparity underscores the heightened scrutiny AI inventions face regarding patent‑eligibility doctrine, particularly the abstract‑idea exception that has been a focal point of recent Supreme Court jurisprudence.
Comparison with Non‑AI Applications
While the § 101 gap is pronounced, the story differs for obviousness. Crouch observes that “The chart below shows no difference between the two categories for obviousness rejections, both of which hover close to 80%.” This similarity suggests that, aside from eligibility concerns, examiners treat the inventive step of AI and non‑AI inventions with comparable rigor. The parity in obviousness rates reinforces the notion that the disparity is rooted in § 101 analysis rather than a general shift toward stricter patentability standards.
Impact of Revised Guidance
Crouch highlights the effect of the USPTO’s pre‑COVID Revised Patent Subject Matter Eligibility Guidance, stating, “As you can see from the chart, the pre-COVID Revised Patent Subject Matter Eligibility Guidance cut the AI §101 rate about in half.” That guidance, issued to clarify the application of the Alice/Mayo framework, temporarily alleviated the burden on AI applicants by providing clearer examples of eligible subject matter. However, the benefit proved fleeting: “By 2025 that reduction has essentially disappeared.” The erosion of this early improvement indicates that either the guidance’s influence waned over time or that other factors began to counterbalance its effect.
Stability of Non‑AI § 101 Rates
In contrast, the eligibility rejection rate for non‑AI applications remained flat. Crouch notes, “Whereas the eligibility rejection rate in non-AI applications has not risen.” This stability suggests that the USPTO’s overall application of § 101 to traditional technologies has not undergone a comparable shift. The divergence therefore points to a phenomenon uniquely affecting AI‑related filings, whether due to the nature of the inventions themselves or to evolving examination practices specific to this technology sector.
Possible Drivers of Change
Crouch candidly admits the limits of his data: “From this data, I cannot tell the extent that the change is tied to changes by applicants in what is being filed or instead changes at the USPTO.” Two broad hypotheses emerge. On the applicant side, there may be a trend toward filing broader, more abstract AI concepts—such as generic machine‑learning algorithms or data‑processing methods—that fall closer to the abstract‑idea boundary. On the examiner side, shifts in training, workload pressures, or updated internal guidance could be leading to a more rigorous application of the § 101 test to AI claims. Disentangling these influences would require a deeper dive into filing patterns, examiner notes, and possibly interviews with patent practitioners.
Implications for Patent Practitioners
For attorneys and agents drafting AI patent applications, the data serve as a cautionary signal. Practitioners should anticipate a heightened likelihood of § 101 rejections and consider strategies that emphasize concrete technical improvements, specific hardware implementations, or tangible real‑world effects. Drawing on the USPTO’s own examples from the Revised Guidance—such as tying an AI model to a particular technological transformation or integrating it into a specific machine—may help mitigate eligibility risks. Additionally, monitoring examiner tendencies and staying abreast of any updates to the AI‑patent classifier or related guidance will be crucial for maintaining a competitive edge.
Conclusion
Dennis Crouch’s analysis of the USPTO’s AI‑patent classifier illuminates a growing chasm in how AI‑related inventions are treated under the patent‑eligibility statute. While obviousness standards remain steady across AI and non‑AI fields, the § 101 rejection rate for AI applications has surged to 42 % in 2025, dwarfing the 6.9 % rate for non‑AI filings. The temporary relief afforded by the pre‑COVID Revised Guidance has dissipated by 2025, and the underlying cause—whether shifts in applicant behavior or evolving USPTO examination practices—remains uncertain. For the patent community, the takeaway is clear: navigating AI patents now demands a vigilant focus on eligibility‑savvy drafting and an awareness of the dynamic landscape governing § 101 analysis. As AI continues to permeate every sector of innovation, staying attuned to these trends will be essential for securing robust patent protection.
Rise of AI Patents and their Corresponding Eligibility Rejections