Argument Is Not Evidence: Technical Declarations and the Evidentiary Turn in AI Patent Eligibility

Argument is not evidence. For patent applications directed to Artificial Intelligence (AI) and machine learning inventions, this principle has taken on heightened importance. Even when claims recite a genuine technological improvement, examiners and the Patent Trial and Appeal Board (PTAB) may still reject them under Section 101 of the U.S. Patent Laws if the improvement is not sufficiently clear from the specification and claims alone.

A growing and effective response has been evidentiary rather than purely argumentative. Applicants are turning to objective evidence, notably through Subject Matter Eligibility Declarations (SMEDs), to demonstrate how the claimed invention integrates the abstract idea into a practical application, without adding new matter to the original disclosure. For additional context on the broader patent eligibility landscape for AI and machine learning innovations, see the previous article on Navigating Patent Eligibility for AI and Machine Learning Innovations.

The Post-Desjardins Landscape Places Greater Value on Evidence

Ex parte Desjardins, Appeal 2024-000567, a decision of the U.S. Patent and Trademark Office's (USPTO) Appeals Review Panel (ARP) decided September 26, 2025, and designated precedential on November 4, 2025, shifted the Section 101 conversation toward technical improvements in machine learning systems themselves. Since its designation as precedential, the PTAB has repeatedly focused on whether applicants demonstrate improvements to model architecture, training methodology, or computational efficiency. Building on the December 2025 guidance, the April 30, 2026 update to the SMED best practices memorandum incorporates feedback from the initial implementation of the program and confirms the USPTO's intent to treat the guidance as a living document, with periodic updates as additional experience and practitioner input are gathered. The updated memorandum is available here.

Attorney argument remains important, but it is not a substitute for evidence. As the Federal Circuit has put it in the context of a PTAB appeal, "Attorney argument is not evidence." Icon Health & Fitness, Inc. v. Strava, Inc., 849 F.3d 1034 (Fed. Cir. 2017); see also Manual of Patent Examining Procedure (MPEP) § 716.01(c). Objective evidence generally carries greater persuasive weight.

Characteristics of Effective SMEDs

Persuasive declarations tend to share several characteristics.

First, the declaration establishes a direct nexus between the evidence and specific claim limitations. General statements about AI performance rarely persuade. Successful declarations explain why particular claimed features improve the functioning of the technology itself, not merely what results the technology achieves.

Second, the declaration relies on objective technical evidence. Evidence like benchmark testing, computational savings, reduced storage requirements, improved model robustness, reductions in catastrophic forgetting, improved handling of data drift, or measurable increases in model accuracy can provide concrete support when tied directly to the claimed invention.

Third, the declaration includes expert explanation. A declaration can bridge the gap between a dense technical disclosure and what would be apparent to a person of ordinary skill in the art (POSITA) within the constraints of examination.

Last but not least, effective declarations consistently reference the original specification. The USPTO guidance specifies: SMEDs are designed to explain or support what already exists within the disclosure. They should not introduce new inventive concepts or supplement missing disclosure. Attempting to use a declaration to add technical details absent from the original filing risks both credibility and potential new matter issues.

Lessons from Recent PTAB Decisions

Ex parte Carmody, Appeal 2025-002843 (PTAB Dec. 30, 2025), provides a clear illustration of a favorable scenario. In that decision, the PTAB reversed a Section 101 rejection after finding that the claims recited a technological improvement to an AI-based system for marketing and sales orchestration. The panel credited the technical advantage as stated in the specification and claims, following the framework set out in Desjardins and Enfish.

This outcome underscores the value of clear specification support. It also explains why SMEDs exist as a tool when that clarity is not yet fully evident on the face of the application. When the technological improvement would benefit from objective data or expert explanation to make it apparent to an examiner, a well-crafted SMED can supply the missing bridge without adding new matter.

Analyses of the decision acknowledge the decision serves as a reminder of the importance of stating a technical advantage in the specification. When the record already articulates the technical advantage in sufficient detail, the PTAB has shown willingness to find eligibility under the current framework. When it does not, SMEDs offer a structured path to supplement the record with objective evidence.

Statistical reviews of PTAB Section 101 outcomes confirm that the reversal rate has roughly doubled under USPTO Director Squires, spiking to roughly 29% in November 2025 before settling near 20%. Mean within-judge reversal rates moved from 11.0% to 22.3%. Evidence is part of what is driving that shift.

Common Pitfalls That Undermine Otherwise Strong Cases

Well-prepared declarations are powerful advocacy tools but they are not cure-alls.

One common mistake is treating a declaration as an opportunity to repair an inadequate specification. If the original application fails to describe the underlying technological improvement, a declaration cannot create that disclosure after the fact. Under the SMED guidance, a declaration supports what is already in the specification, but cannot substitute for a deficient one.

Applicants should avoid combining multiple legal issues into a single declaration. Eligibility evidence becomes less persuasive when mixed with obviousness arguments, enablement discussions, or written description issues. The guidance recommends standalone declarations focused specifically on eligibility.

Applicants should also avoid conclusory testimony. Assertions that an invention improves AI performance carry limited weight unless accompanied by supporting data and a clear explanation connecting that evidence to the claimed limitations. And evidence must remain commensurate in scope with the claims (see MPEP § 716.02(d)). Narrow testing supporting broad claims may raise more questions than it answers.

A final caution to consider: SMEDs create witnesses. A declaration submitted during prosecution becomes part of the prosecution history, and the declarant becomes a potential witness in any subsequent litigation. The stronger and more specific the SMED, the more it helps during prosecution, but specificity cuts both ways. Before recommending that a client submit a SMED, as with any declaration, counsel should discuss who will sign, what positions the declarant is committing to, and whether the declarant can defend the positions under cross-examination.

Integrating Evidence into Portfolio Strategy

Sound SMED strategy starts long before prosecution. Applications drafted with future evidentiary needs in mind are often easier to defend. As a leading industry association's AI patenting handbook stresses, detailed technical descriptions, benchmark comparisons, alternative embodiments, architectural diagrams, flowcharts, pseudocode, and performance data provide a stronger foundation for future declarations. Applicants should also consider continuation strategies that preserve flexibility for narrower claims while maintaining broader portfolio objectives.

Evidence planning also informs the earlier question of what to patent at all. Core architectural improvements may warrant patent protection, while proprietary datasets, training methodologies, and implementation details often remain better suited for trade secret protection. Commentary on recent generative-AI trade-secret decisions illustrates that the choice of which tier of AI tools inventors use during development has direct implications for trade secret protection. Deciding early which innovations will be patented, and which will therefore require eligibility evidence, connects invention capture strategy to prosecution strategy.

Proactive Evidence Strategies for AI Portfolios

Preparing eligibility evidence should become part of a broader prosecution strategy rather than a reactive response to a rejection. Organizations developing AI technologies should identify high-value inventions early, preserve benchmark testing throughout development, coordinate closely between inventors and prosecution counsel, and maintain records documenting technological improvements as they occur.

When objective evidence exists before examination begins, responding to Section 101 rejections can become substantially more efficient. Strong prosecution strategies often begin well before the first office action. Applicants and their representatives should integrate thoughtful drafting, robust technical disclosures, and carefully prepared evidentiary support into a single, cohesive record.

As AI innovation accelerates, practitioners who view evidence as a central component of patent strategy rather than a last-minute response should be better positioned to build durable and defensible portfolios.