Artificial intelligence and machine learning are now being used across nearly every industry. Inventors, startups, software companies, and established businesses are using AI to optimize schedules, classify images, detect fraud, generate recommendations, automate workflows, analyze risk, process medical data, and improve countless other business and technical operations.
But using AI does not automatically make an invention patentable.
The Federal Circuit’s decision in Recentive Analytics, Inc. v. Fox Corp. has quickly become one of the most important patent eligibility decisions for AI and machine-learning inventions. The case does not hold that AI inventions are categorically unpatentable. Instead, it reinforces a critical distinction: using generic machine learning to achieve a useful result in a new field is different from claiming a specific technological improvement to machine learning, software, computer functionality, or another technical system.
That distinction matters for inventors because many AI-related ideas are described in terms of results. For example, an inventor may say that the system “uses AI to optimize scheduling,” “uses machine learning to predict risk,” “uses a neural network to classify data,” or “updates outputs in real time.” Those descriptions may be commercially valuable, but after Recentive, they may not be enough for patent eligibility under 35 U.S.C. § 101.
The Recentive Decision
Recentive Analytics involved four patents directed to using machine learning to generate network maps and schedules for television broadcasts and live events. Recentive sued Fox and related entities for patent infringement. The district court dismissed the case, finding that the asserted patents claimed patent-ineligible subject matter under § 101. The Federal Circuit affirmed.
The court held that the claims were directed to the abstract idea of using generic machine-learning techniques in a particular environment. The patents did not claim an improvement to machine learning itself. They used machine learning as a tool to generate schedules and network maps.
That is the key point. The court was not saying that machine learning is always abstract or that AI cannot be patented. Rather, the court concluded that applying established machine-learning techniques to a new data environment, without disclosing an improvement to the machine-learning model itself, does not make a claim patent eligible.
The Federal Circuit also rejected the argument that iterative training or dynamic real-time adjustment supplied the required technological improvement. The court treated those features as inherent to machine learning, not as inventive improvements by themselves.
In other words, saying that a machine-learning model is trained, retrained, updated, adjusted, or applied to live data may not be enough. The patent application must explain what is technically improved and how that improvement is achieved.
Why Recentive Matters
The significance of Recentive is not limited to television scheduling or network mapping. The case has already become a major § 101 authority. The Shepard’s report shows no negative subsequent appellate history, notes that the Supreme Court denied certiorari on December 8, 2025, and identifies 176 citing decisions.
That matters because courts and the Patent Trial and Appeal Board are using Recentive as a broader rule for AI, machine learning, neural networks, predictive models, optimization systems, and software claims that rely on data-driven models without claiming a specific technical implementation.
The emerging lesson is clear: an AI-related patent application should not merely describe the use of AI to reach a result. It should describe the technical contribution.
Courts Are Using Recentive to Reject “New Field of Use” Arguments
One of the most important themes from Recentive is that applying machine learning to a new industry or field of use is not enough by itself.
A patent claim does not become non-abstract merely because it is the first to use machine learning in a particular business area. An “AI for X” invention still needs to be claimed and described as a technological invention, not merely as a field-specific use of generic AI.
The Shepard’s report shows courts citing Recentive for this point in several cases. For example, in All Terminal Services, LLC v. Roboflow, Inc., the court followed Recentive and rejected the argument that applying machine learning to a new field of use made the claims non-abstract. The claims involved OCR and machine-learning models, including CNN and RNN models, but the court still focused on whether the claims did more than use generic machine learning in a new environment.
The Federal Circuit also followed Recentive in Rensselaer Polytechnic Institute v. Amazon.com, Inc., explaining that claims directed to applying machine learning to a “new field of use” remain problematic where the claims disclose only that machine learning is used in a new environment.
For inventors, this means that the following types of descriptions may be vulnerable if not supported by technical detail:
- AI for scheduling;
- AI for advertising;
- AI for fintech;
- AI for healthcare administration;
- AI for legal workflows;
- AI for real estate;
- AI for manufacturing logistics;
- AI for image classification;
- AI for business risk scoring.
The issue is not whether those ideas are useful. They may be extremely useful. The issue is whether the patent claims describe a specific technological improvement, rather than the use of known AI tools to perform a result-oriented task.
Courts Are Looking for the “How,” Not Just the Result
Another major lesson from Recentive is that courts are looking for the mechanism behind the alleged improvement.
It is not enough to claim that a system “optimizes,” “predicts,” “classifies,” “updates,” or “generates” an output. Patent eligibility often depends on whether the claims and specification explain how the technology achieves the improvement.
In Partner One Acquisitions Inc. v. WhatFix Private Ltd., the District of Delaware cited Recentive for the principle that claims invoking computer technology without identifying how that technology is applied in a specific or inventive manner are typically abstract. The court contrasted those types of claims with claims that solve particular technical problems, such as by making specific improvements to computer, software, or network functioning.
Similarly, in GoTV Streaming, LLC v. Netflix, Inc., the Federal Circuit cited Recentive in connection with the “how” requirement and the insufficiency of result-focused functional language.
This is one of the most practical drafting lessons from the case.
A weak AI patent disclosure might say:
The system uses machine learning to analyze data and generate an optimized recommendation.
A stronger disclosure would explain:
- what data is used;
- how the data is transformed;
- what model architecture or training process is used;
- what technical constraint is solved;
- how the model operates differently from conventional models;
- how computer performance, model accuracy, resource usage, latency, convergence, or another technical characteristic is improved; and
- how those improvements are reflected in the claims.
The patent application should not merely claim the desired output. It should explain the technical path used to reach that output.
Generic Hardware and Generic Computer Implementation Are Usually Not Enough
Recentive is also being used at Alice Step Two, where courts ask whether the claim contains an inventive concept sufficient to transform an abstract idea into a patent-eligible application.
In B.E. Tech., L.L.C. v. Google LLC, the court cited Recentive at Alice Step Two in finding that certain alleged improvements were insufficient to provide an inventive concept.
The Federal Circuit’s United Services Automobile Association v. PNC Bank decisions also cited Recentive for the principle that generic implementation does not supply the inventive concept. In one decision, the court held that including a handheld mobile device did not add an inventive concept because the mobile device was generic hardware. In another, the court reiterated that the abstract idea itself cannot supply the inventive concept.
For AI inventions, this means that merely reciting standard components may not be enough. A claim is not necessarily patent eligible simply because it includes:
- a processor;
- a server;
- a database;
- a mobile device;
- a neural network;
- a trained model;
- a training dataset;
- real-time data;
- cloud computing;
- generic model execution;
- generic data analysis.
Those elements may be part of a patent-eligible invention, but they usually need to be tied to a specific technical improvement.
Recentive Is Not a Death Sentence for AI Patents
The most important nuance is that Recentive does not mean AI inventions cannot be patented.
Courts are also distinguishing Recentive where the claims include specific technical detail.
In Aon Re, Inc. v. Zesty.Ai, Inc., the court distinguished Recentive. The court explained that Recentive involved claims that broadly referenced iterative training and real-time updating without specifying the steps through which machine-learning technology achieved an improvement. By contrast, the claim before the court did not simply recite the result or broadly direct the use of a computer and machine learning to perform risk analysis. The court emphasized the specificity of the claim.
Similarly, in Nielsen Co. (US), LLC v. Hyphametrics, Inc., the court rejected the argument that the asserted claims were like Recentive. The court noted that the claims allegedly provided an improvement to image processing rather than merely using generic machine learning to generate an output.
These cases show the other side of the rule. Recentive hurts claims that use AI as a black box. But it may leave room for claims that describe a concrete technical improvement.
The distinction is critical.
A claim that says “use machine learning to produce a better result” may be vulnerable.
A claim that explains a specific improvement to model training, model architecture, data representation, image processing, computer performance, resource allocation, or another technical process may be in a stronger position.
The PTAB Is Applying Recentive Broadly
The Shepard’s report also shows extensive use of Recentive by the Patent Trial and Appeal Board. The Board has cited the case in appeals involving neural networks, predictive models, AI engines, fraud detection, scheduling, sales prediction, medical analysis, anomaly detection, image analysis, and other data-processing uses.
Several PTAB decisions use the same basic distinction: Recentive applies where claims merely use generic machine learning in a new data environment, while claims may be stronger where they improve how the machine-learning model itself operates.
That distinction has become especially important in view of cases discussing Ex parte Desjardins, where the Board recognized that certain improvements to machine-learning model operation may support eligibility. The practical takeaway is that AI claims should be drafted to show whether the invention improves the model, the computer, the data processing pipeline, or another technical system — not merely that AI is being used.
Practical Lessons for Inventors
For inventors and startups, Recentive provides several important lessons.
First, do not assume that using AI makes an invention patentable. AI is a tool. Like any other tool, it must be used as part of a patent-eligible technological invention.
Second, avoid describing the invention only in terms of results. Phrases like “optimize,” “predict,” “recommend,” “classify,” “score,” “detect,” and “automate” should be supported by technical detail.
Third, explain the technological problem. Is the invention improving model accuracy? Reducing computational load? Improving latency? Changing the way training data is selected? Improving convergence? Reducing memory usage? Improving image processing? Improving operation of a device or system? The application should identify the technical problem and the technical solution.
Fourth, make sure the claims reflect the improvement. It is not enough for the specification to mention technical benefits if the claims only recite a broad result.
Fifth, avoid treating machine learning as a black box. The more the application treats AI as a generic tool that receives inputs and produces outputs, the more it may look like Recentive.
Conclusion
Recentive Analytics v. Fox is a major warning for AI and machine-learning patent applications. The case does not eliminate patent protection for AI inventions, but it makes clear that patent eligibility requires more than applying generic machine learning to a new business problem or data environment.
The best AI patent applications should be drafted as technical documents. They should explain what the invention improves, how the improvement is achieved, and why the claimed approach is more than using existing AI tools to perform a known task faster or in a new context.AI can still be part of a patentable invention. But after Recentive, the application needs to do more than say “use machine learning.” It needs to disclose and claim the technical contribution.
