Drafting AI Patents to Survive Section 101: Strategies Beyond the Current Guidance Cycle

For applicants seeking to patent AI-related inventions in the United States, Section 101 of the Patent Act (35 U.S.C. § 101) remains the single most formidable barrier to allowance. At IPWatchdog LIVE 2026, a panel session titled “AI Patents Under the Microscope: Drafting and Prosecuting for the Long Game” brought together practitioners from Salesforce, Blueshift IP, Capital One, Ballard Spahr, and Ankar AI to discuss drafting strategies that can outlast any individual guidance cycle.

The structural challenge the panelists identified is what they termed the “guidance-to-practice gap” — the delay between the USPTO issuing new memos and the point at which examiner behavior actually shifts in response. Applications optimized for a particular guidance snapshot become brittle when that guidance changes.

A Rapidly Shifting Guidance Landscape

The USPTO’s approach to AI patent examination has undergone dramatic shifts in a remarkably short period. The February 2024 Inventorship Guidance, the July 2024 Subject Matter Eligibility Update, and the November 2025 memo rescinding the February 2024 guidance represent three major directional changes in under two years.

This rapid evolution creates a structural risk: applications that chase the current guidance become vulnerable when the next cycle arrives. The panelists emphasized the need for a durable prosecution philosophy rather than point-in-time compliance.

The “Analyzing” and “Determining” Trap

One of the most actionable insights from the panel concerned verb choice in claim language. Terms like “analyzing” and “determining” consistently anchor examiners in the mental step category, inviting characterization of the claimed invention as an abstract idea that could be performed by human thought.

Under the Alice/Mayo two-step framework, once a claim is characterized as “directed to” an abstract idea at Step 1, the applicant faces the uphill battle of demonstrating “significantly more” at Step 2. Claims built around “analyzing” and “determining” effectively lock in the examiner’s framing before the first office action is even issued, making subsequent arguments significantly harder.

The panelists stressed that word choice at the drafting stage determines the examination framing, and that this upstream decision is far more consequential than any response strategy deployed after rejection.

Eight Durable Drafting Strategies

The panel outlined a set of strategies designed to build applications that survive not just the current guidance cycle, but the next one as well.

1. Inverted continuation strategy. Rather than filing broad claims first and narrowing later, start with the most specific, defensible claims and broaden coverage through subsequent continuation applications.

2. Expanded specifications with layered fallback positions. Draft specifications with multiple layers of alternative embodiments, creating a foundation that allows claims to be restructured flexibly in response to guidance changes.

3. Distinguish model use from model improvement. For AI inventions employing off-the-shelf models, clearly differentiate between merely applying an existing model and genuinely improving upon it. Salesforce’s Jason Harrier emphasized the importance of pushing inventors to articulate what the system actually improves, not just what it uses.

4. Start with the invention story. Rather than reverse-engineering claims from guidance requirements, begin with a clear narrative of what the inventor actually did and why it represents a genuine improvement over the prior art, then build claims from that foundation.

5. Clinical intake process. Implement a systematic pre-filing process to identify the actual inventive contribution, particularly when AI tools were used during the innovation process. Ballard Spahr’s Jonathan Hummel highlighted the importance of early collaboration with counsel when AI tools are involved, to properly scope the inventive contribution.

6. Selective AI delegation. Use AI tools for mechanical tasks such as prior art survey organization, but keep human judgment at the core of the invention narrative.

7. Deliberate terminology selection. Avoid “analyzing” and “determining” in favor of terms that describe concrete technical processes, steering examiner framing toward “technical improvement” rather than “mental step.”

8. Filing discipline. Not every AI-related invention merits a patent application. Apply rigorous screening to focus resources on applications with a realistic prospect of overcoming the Section 101 bar.

The Rise of Chinese-Language Prior Art

The panel also flagged a significant shift in the prior art landscape: an estimated 80–90% of relevant AI prior art is now in Chinese-language filings. This development has fundamental implications for search strategy and suggests that Chinese-language search capability is becoming a prerequisite for effective AI patent prosecution.

Building Applications That Endure

The panel’s core message was clear: the goal should not be to build applications that comply with today’s guidance, but rather to build applications that maintain their defensive strength regardless of how the guidance evolves. This requires clearly articulating the technical improvement at the heart of the invention, selecting claim language that does not depend on abstract verbs, and embedding sufficient alternative embodiments in the specification to support flexible claim restructuring.

In a prosecution environment defined by rapid change, the most valuable asset is not perfect alignment with the current guidance — it is the adaptability to weather the next shift.