Real Software Patent Examples — What Gets Protected (and What Doesn't)

Software patents are the most contested category in US patent law. The Supreme Court's 2014 decision in Alice Corp. v. CLS Bank created a two-step test that invalidated thousands of existing patents and raised the bar for new ones. Yet software patents are still granted — tens of thousands per year. The question is: what makes a software patent survive?

The answer, post-Alice, is specificity. A software patent must describe a specific technical solution to a specific technical problem — not an abstract idea implemented generically on a computer.

5 real software patents that were granted

1. PageRank — US6285999

Larry Page's foundational search algorithm patent, assigned to Stanford and licensed to Google. The claims describe a specific method for ranking hyperlinked documents by calculating an importance score based on the number and weight of inbound links — where the weight of each link depends recursively on the importance of the linking page.

This patent survived scrutiny because it claimed a specific computational method that improved the functioning of a search engine, not just "rank documents on a computer." The claims describe a concrete mathematical relationship that produces a specific, measurable output.

Status: Expired 2011. Public domain. View at /patent/us/6285999.

2. Amazon One-Click Ordering — US5960411

One of the most famous (and controversial) software patents ever granted. US5960411 claims a method for ordering items online using a single action — storing payment and shipping information so a customer can complete a purchase in one click without re-entering data.

Critics argued this was trivially obvious. Amazon didn't invent storing customer information — they combined existing concepts in a commercially clever way. The patent was granted in 1999, drew enormous criticism, and expired in 2017 after being subjected to failed re-examination attempts and an international prior art challenge by Amazon's competitor.

Status: Expired 2017. Public domain.

3. Google Maps Routing — US8762056

Covers methods for generating turn-by-turn navigation routes that account for real-time traffic data, incorporating historical traffic patterns to predict future traffic conditions during the route. The claims specify particular algorithmic steps: how historical data is weighted, how predictions propagate across a network, how the route is updated as new data arrives.

The technical specificity — a particular approach to a known computational problem — distinguishes this from claiming "do navigation on a computer."

Status: Active, in force.

4. Predictive Text Input — US8898576

Apple's patent covering predictive text input that identifies statistically likely completions based on a character-level language model, position in the text stream, and context of surrounding words. The claims describe specific data structures, model update methods, and candidate ranking approaches.

Post-Alice analysis: this solves a specific technical problem (reducing keystrokes on a touch keyboard) via a specific technical mechanism (a constrained statistical model with defined inputs and outputs). Not abstract.

Status: Active, in force.

5. Dropbox File Sync — US8065397 (and related)

Covers methods for synchronizing files across multiple devices by tracking block-level changes (rather than re-transmitting entire files), maintaining a content-addressable store, and resolving conflicts using vector clocks. The technical claims are specific to the architecture of a distributed file synchronization system.

The specificity — block-level diffing, content-addressable storage, vector clock conflict resolution — separates this from a generic "sync files on a computer" claim that would fail Alice.

Status: In force.

3 examples of software claims that fail

1. "Perform financial transaction on the internet"

The claim at the heart of Alice Corp. v. CLS Bank — intermediated settlement of financial transactions via a computer system. The Court found that "computer implementation" of an abstract idea doesn't add anything that makes the claim patent-eligible. Implementing a known financial practice on a generic computer adds no inventive concept.

Why it fails: Abstract idea + generic computer implementation = patent-ineligible.

2. "Apply machine learning to [industry-specific data]"

A common pattern in rejected patent applications: claim a machine learning model applied to, say, medical image classification, without specifying any novel architecture, training method, or mathematical approach. Generic claims like "a neural network trained to identify tumors in radiology images" are abstract — they claim the result, not a specific method of achieving it.

Why it fails: No specific technical solution to a specific technical problem. Any neural network architecture would satisfy the claim.

3. "Send a notification when a threshold is met"

Monitoring data and triggering an alert — without specifying anything novel about how the monitoring works, how the threshold is calculated, or how the notification is transmitted — is an abstract concept that has been rejected consistently post-Alice.

Why it fails: The "abstract idea" is event-triggered notification. The computer implementation adds nothing inventive. This describes what every alarm system does.

What software patent examiners look for post-Alice

The USPTO's guidance requires examiners to ask: does the claim (1) recite an abstract idea, mathematical concept, or certain mental process, and (2) if so, does it include additional elements that integrate the abstract idea into a practical application or add a specific inventive concept?

Practical application means something concrete: improving the functioning of a computer itself, using a computer in a specific way that produces a useful, concrete result, or solving a technical problem in a particular technical way.

The key insight for inventors: write claims that describe what your software actually does technically — not what it achieves commercially. "A method for recommending products" fails. "A method for generating product recommendations using a collaborative filtering algorithm that weights co-purchase frequency by recency-adjusted decay functions applied to user session vectors" has a better chance.

Use the PatentBrief idea checker to survey the existing patent landscape before investing in a software patent application.

PatentBrief is not a law firm. Nothing here is legal advice. Consult a qualified IP attorney before making filing decisions.