Can I patent software, an app, or an AI model? A European answer

Introduction

"Can I patent software?" is the most Googled question in European startup IP, and the answers online are almost uniformly bad. Most sources give the US answer (broadly yes), a vague "it depends", or a flat "no" that's simply wrong. The European answer is more nuanced than any of those, and understanding it accurately is the difference between filing a strong software patent and dismissing the option entirely.

This article explains the European rule, how it applies to apps and AI, and how to tell whether your specific software is likely patentable. No prior knowledge assumed.

‍

The European rule on software patents

The European Patent Convention excludes "programs for computers as such" from patentability. Those two words, "as such", do a lot of work. The EPO's interpretation, consistently applied across thousands of decisions, is this: software is not patentable when it is nothing more than a sequence of instructions for a computer to follow. But the moment that software causes a further technical effect beyond the normal operation of the computer itself, it becomes patentable.

The EPO grants software patents routinely. Thousands every year. What it doesn't grant patents for is pure business methods, pure mathematical methods, or pure presentation of information, when those things happen to be implemented in software.

‍

What counts as a "technical effect"

This is where founders often get stuck. "Technical effect" is not defined by the novelty of the idea. It is defined by how that idea interacts with the physical or computational world.

Software that has been granted European patents:

• Image-compression algorithms that reduce file size while preserving visual fidelity.
• Error-correction codes used in wireless transmission.
• Database-indexing methods that improve query performance measurably.
• Computer-vision algorithms that detect specific features in an image.
• Control software for physical systems.
• Encryption and security methods.
• Specific machine-learning architectures applied to a technical problem.

Software that has repeatedly been refused:

• Business method for auctioning items, implemented in software.
• User interface arrangement where the novelty lies only in the visual design.
• Mathematical method for calculating a financial risk score.
• Method for organising information in a database where the novelty is only the organisation itself.

The pattern is clear once you see it. Granted patents solve a technical problem with a technical solution. Refused applications solve a business problem with code, and the code is incidental to the innovation.

‍

A useful test: would this still be novel if done manually?

Not a legal test, but it catches most cases correctly. Imagine doing your invention without a computer. If a human with a pen and paper could do the same thing, slowly, tediously, but in principle, you are probably dealing with a business method or an abstract method, and the EPO will be sceptical.

If your invention depends on specifically technical characteristics (processing speed, memory architecture, sensor input, network behaviour, cryptographic properties) that a human could not replicate in principle, you are firmly in patent-eligible territory.

‍

Can I patent a mobile app?

Apps are software on a mobile device, and the same rules apply. The question is always: what does the app technically do?

An app that provides a novel user interface for booking restaurants: almost certainly not patentable. An app that uses the phone's accelerometer and GPS to detect and classify a specific physical activity with a new algorithm: likely patentable. An app that schedules push notifications according to a new business rule: not patentable. An app that optimises push-notification delivery to minimise battery drain through a novel technical method: potentially patentable.

The mistake founders make is equating "my app is novel" with "my app is patentable". Novelty in the market isn't the same as novelty in the technical sense the EPO cares about. An app can be a unicorn without having anything patentable about it.

‍

Can I patent an AI model in Europe?

AI patenting is one of the most actively evolving areas in European patent law. The EPO publishes guidelines that are updated regularly, and the position is clearer than most people assume.

Training methods are patentable when they solve a specific technical problem: a novel loss function for a particular task, a new architecture for a specific type of data, a training technique that reduces compute by a measurable factor. Applications of AI are patentable when the AI is applied to a technical problem: AI-assisted medical diagnosis, AI-controlled manufacturing, AI-based signal processing. Mathematical methods underlying AI are not patentable in isolation. You cannot patent "a new way of doing matrix multiplication"; you can patent "a matrix multiplication method that reduces latency in neural-network inference on specific hardware architectures". Pure AI-generated business outputs are not patentable. A model that recommends marketing copy, scores leads, or writes emails is solving a business problem, not a technical one.

For AI startups, the most valuable patents are usually around specific architectures, training-efficiency techniques, novel inference methods, and technical applications of models to engineering problems.

‍

How to tell if your software is likely patentable

Three questions, in order.

1. What technical problem does it solve? If you cannot articulate this in engineering terms, the patent office will also struggle. "It helps users find restaurants faster" is not a technical problem. "It reduces query latency on a distributed database through a novel partitioning scheme" is.
2. How does it solve that problem? The solution needs technical character: algorithms interacting with data structures, sensors, networks, hardware, or cryptographic properties. Not a business rule implemented in code.
3. Is it new and non-obvious? Even a clearly technical invention still has to pass the patentability criteria. The patentability test on ptntpwr covers this.

‍

The most common mistake founders make

Assuming, based on US advice, forum threads, or a conversation with a lawyer who does not specialise in patents, that software cannot be patented in Europe. It can. The EPO grants software patents every week. The ones that fail tend to fail because the founder could not articulate the technical effect clearly, not because software is unpatentable.

A good patent attorney spends much of the first drafting session helping the founder re-express the invention in technical terms. This is not about stretching the truth. It is about finding the technical characterisation that is already present in the invention but hidden under business language.

‍

Conclusion

Software is patentable in Europe when it produces a technical effect. Most software worth patenting does. The test isn't whether your idea is innovative in a business sense. It's whether the software solves a technical problem in a technical way. Apps follow the same rule. AI models follow the same rule, with additional nuance around training methods and applications.

If you have built software with any non-trivial technical mechanism at its core, a custom algorithm, a novel architecture, a specific optimisation, a distinctive data-processing technique, do not dismiss patentability on the basis of general advice. Test it specifically.

Find out if your software is patentable.

‍

Frequently asked questions

‍

Can you patent software in Europe?

Yes. The European Patent Office grants software patents routinely, thousands every year. The condition is that the software must produce a technical effect beyond the normal operation of a computer. Software that solves a technical problem, such as improving processing efficiency, reducing energy use, or enabling a specific hardware function, is patentable. Pure business methods implemented in software are not.

‍

What does "technical effect" mean in European patent law?

A technical effect is an outcome that goes beyond simply running a program. Examples include improved data-compression rates, reduced error rates in a communication system, faster processing of a specific data type, or improved control of a physical device. Business outcomes such as increased revenue or better user engagement do not qualify as technical effects under EPO examination standards.

‍

Can I patent an AI model or machine learning system in Europe?

Yes, in many cases. AI inventions are patentable in Europe when they solve a specific technical problem or produce a measurable technical effect. Novel training methods, specific model architectures applied to technical tasks, and AI systems that control physical processes regularly receive EPO grants. General language models performing business tasks are harder to patent without a clearly defined technical application.

‍

How do I know if my software invention is patentable?

Ask three questions. First: what technical problem does it solve, in engineering terms? Second: does the solution involve algorithms, data structures, or hardware interactions with a specific technical character? Third: is it new and non-obvious given existing prior art? If you can answer all three clearly, the invention likely has a patentable core. The ptntpwr patentability test covers this in under two minutes.