For Software Founders
How to Patent Software
Yes, you can patent software — the "you can't patent software" line is a myth. But the Alice decision made it genuinely hard, and getting it right means understanding exactly what survives a Section 101 challenge and how to draft for it.
Educational guide, not legal advice. Software patent prosecution is specialized — work with a patent attorney experienced in your art.
Yes, you can patent software — the Alice caveat
The myth that "you cannot patent software" is wrong. Software is patentable in the US. What changed is the difficulty: the Supreme Court's Alice Corp. v. CLS Bank (2014) decision made software far harder to patent by targeting one specific move — taking an abstract idea and patenting it just by saying "do it on a computer."
Alice did not ban software patents. It established a two-step test: (1) Is the claim directed to an abstract idea? (2) If so, does it add an "inventive concept" — something significantly more than the abstract idea implemented on generic hardware? Software that clears step 2 by reciting a concrete technical improvement is patentable, and thousands issue every year.
What kinds of software ARE patentable
Technical performance improvements
A method that makes a computer process faster, use less memory, or consume less power — a concrete improvement to how the machine works. Courts treat improvements to computer functionality itself as patent-eligible.
Novel data structures and architectures
A new way of organizing or indexing data, a specific database architecture, or a particular system design that solves a technical problem in a non-obvious way.
Compression, encoding, and signal processing
Methods that transform data more efficiently (audio/video codecs, compression algorithms, error correction) are classic patentable software because they are concrete technical methods.
Specific machine-learning methods
A novel model architecture, a specific training methodology, or a concrete application of ML to a technical problem with measurable improvements — not just 'use AI to do X.' (See the dedicated AI patents guide.)
Networking and security techniques
Specific protocols, routing methods, encryption schemes, or security mechanisms rooted in how computers and networks operate.
What is NOT patentable
Business methods on a computer. "A method of [doing some business process] using a computer" is the textbook Alice failure — an abstract idea (the business process) plus generic computing. Escrow, hedging, advertising exchanges, and loyalty programs have all been struck down.
Abstract math and algorithms in the abstract. A mathematical formula or algorithm claimed without a specific technical application is an abstract idea. The same math tied to a concrete technical improvement may be eligible — the application is everything.
Results without a mechanism. "A system that generates realistic images" or "a method of detecting fraud" — claimed as a desired outcome without specifying the novel technical way it is achieved — fails. You must claim the how, not just the what.
Anything a person could do mentally or with pen and paper. If the claimed steps are just organizing information or making a decision that a human could perform, putting them on a computer does not make them patentable.
How to draft software claims that survive
Claim the technical "how," not the abstract "what." Instead of "a method of recommending products to users," claim the specific technical mechanism: the particular data structures, the specific steps, the concrete architecture that produces the result in a novel way.
Frame the specification around a technical problem. Open by identifying a concrete technical problem ("conventional systems require X memory and Y latency") and explain how your specific approach solves it. This framing directly supports the Alice step-2 analysis.
Include measurable improvements. "Reduces query latency by 40% while using half the memory" is evidence of a concrete technical improvement — far stronger than describing a business benefit.
Recite specific implementation in dependent claims. Layer in detail — particular algorithms, thresholds, architectures — so even if a broad claim is challenged, narrower claims with concrete technical limitations survive.
The realistic odds and timeline
Be clear-eyed. Software has the longest USPTO backlog of any field (often 30 to 48 months to grant) and the highest rate of Section 101 rejections. Expect multiple office actions and a real fight over eligibility. The art unit your application lands in significantly affects your odds.
Track One prioritized examination can compress the timeline to roughly a year for an extra fee — worth considering when speed matters in a fast-moving software market where a 4-year-old patent may cover obsolete technology.
When NOT to patent software (and what to do instead)
For many software startups, patents are the wrong tool. If your moat is execution speed, network effects, a data advantage, or brand, a patent may cost more (in money, time, and disclosure) than it is worth.
Copyright protects your code automatically — the specific source and object code, no filing required. Trade secrets protect backend algorithms and methods that never ship to customers and can be kept genuinely secret. And often the strongest "moat" for early software is simply shipping faster than anyone can copy you.
Patent software when the underlying technical innovation is genuinely novel, central to your moat, hard to design around, and worth disclosing publicly in exchange for the monopoly. Otherwise, save the capital.