How to draft the claims in an algorithm patent (2022)

(If you came just for the free download, scroll down to the bottom)

Getting the claims right is critical in any patent, not just digital software patents. Because the scope of protection is not defined in the specification or the drawings, but in the claims. I’ve seen numerous patents which were found not infringed in litigation only because of a completely unnecessary claim element that allowed the infringer to easily work around the patent.

But what if the invention involves an algorithm?

Many companies I work with are experts when it comes to patenting inventions in their domain, typically in a classic field of technology. The typical claim set covers the product and the manuacturing method. But how does an algorithmic invention fit into this structure?

In mathematics and computer science, an algorithm (/ˈælɡərɪðəm/) is a finite sequence of well-defined instructions, typically used to solve a class of specific problems or to perform a computation.


Is an algorithm a product or a process?

If the invention is concerned with the dynamic behavior of the software, the most natural claim category is a method claim. The basic idea is to just list out the steps of the algorithm.

For the sake of claim scope, use the two ways of broadening your claim:

  • Broadening by omission: Leave out certain method steps and focus your independent claim only on the critical ones. The ones which are needed to solve a technical problem. Put the omitted steps in the dependent claims.
  • Broadening by generalization: Find umbrella terms for a group of steps, and break down the group in the dependent claims.

Claim 1: computer-implemented method

The European Patent Office recommends a specific claim wording for software method claims:

A computer-implemented method comprising steps A, B, ...

Keep in mind that a method claim is directly infringed by the actor who performs the steps. In software, this is often the customer. But you don’t want to sue your customer. Wouldn’t it be nice if you could sue your competitor already for selling the software?

The computer program claim

Well, you can. Just include the following additional independent claim (again, recommended wording by the EPO):

A computer program [product] comprising instructions which, when the program is executed by a computer, cause the computer to carry out [the steps of] the method of claim 1. 

The computer program claim needs to be in any European software patent claim set! I really consider it bad practice to leave it out. You get it for free at the EPO and it adds tremendous value to your software patent. This is because it covers any competitor who offers the software even without any physical carrier, e.g. for download or in an app store.

By the way, if you’re wondering why I stress the importance of the claims all the time, you should watch this short video:

Device and storage medium claims

The EPO also suggests additional product claims for the device that carries out the method and a corresponding storage medium:

A data processing apparatus/device/system comprising means for carrying out [the steps of] the method of claim 1.

I always include a device claim in my claim sets, even if my client and its current competitors sell only software. This is because you can never rule out that a competitor will start selling hardware at some point in time. Keep in mind that patents have a maximum lifetime of 20 years!. The device claim may allow you to obtain higher damages because it covers the entire physical apparatus, not just the software program.

A computer-readable [storage] medium comprising instructions which, when executed by a computer, cause the computer to carry out [the steps of] the method of claim 1.

I consider this one optional for European patents and normally only include it if I have room left in my claim set.

Additional Artificial intelligence patent claims

If the invention concerns machine-learning and artificial intelligence, it will most likely make sense to include additional claims for maximum protection.

This is because AI inventions can manifest themselves at many points along the processing pipeline. For example:

  • A unique machine-learning model
  • A unique way of training the model
  • A unique way of generating or preprocessing the training data
  • A unique structure of the training data
  • A unique use of the machine-learning method in a specific application

Depending on where the invention lies in the particular case, it may make sense to draft two independent method claims, one for training the model and one for operating it.

Claiming the machine-learning model and the method of operating it

For the operating method claim, I typically follow the classic input-processing-output model, like so:

A computer-implemented method of <overall purpose>, comprising:
   receiving an input dataset comprising <inputs>;
   <operating step(s) specific to the invention>; and
   producing an output dataset comprising <outputs>.

Sometimes the invention involves a unique machine-learning model with a specific structure. If so, I also include an independent claim directed to the data structure itself:

A machine-learning model for <overall purpose>, in particular for use in the method of claim <operation method claim>, comprising:
   <structural features of the model>. 

Claiming the training of the machine-learning model

And as I said, don’t forget the training method. An example:

A computer-implemented method of training a machine-learning model for <overall purpose>, in particular the machine-learning model of claim <machine-learning model claim>, comprising:
   receiving an input training dataset comprising <inputs>; and
   <training step(s) specific to the invention>.

Sometimes, it makes sense to have even multiple independent training method claims. This may be feasible if the invention involves some amount of pre-training by the manufacturer and a retraining by the customer, for example.

In some inventions, the way how the training data is generated or preprocessed is unique. In these cases, a method claim that produces the training data may be helpful, like so:

A computer-implemented method of generating a training dataset for a machine-learning model, in particular the machine-learning model of claim <machine-learning model claim>, comprising:
   <steps to produce the training data>.

And if the training data itself has a unique structure or reflects technical considerations, I like to include a data structure claim:

A training dataset for use in the method of training a machine-learning model of claim <training method claim>, comprising:
   <structural features that define the format of the training data>.  

Don’t forget the use cases

In many cases, the actual operation of the machine-learning invention already serves a technical purpose (that’s what we claimed above), but can then be used in various downstream use cases. For example, an image classification method may be usable both for operating a robot and for detecting bad products in a production line.

In such cases, I like to include a use claim that lists all conceivable use cases, like so:

Use of the method of claim <operating method claim> for at least one of:
   <list of all conceivable use cases>. 

How to put it into practice?

As we’ve seen, drafting a proper software patent claim set with maximum scope typically involves several independent claims that cover different aspects of the invention.

To help you with the exact wording, I prepared a free collection of templates in Microsoft Word with my preferred formulations and claim structure. Feel free to use this as is, with no warranties whatsoever, and always keep in mind that none of this is legal advice, right?


The template doesn’t cost money. The fee is that you allow Bastian to send you occasional emails with actionable tips on how to make sense of digital patents. Once you’ve signed up, you can download the template immediately.

We treat your data with respect! We will use the information you provide to be in touch with you and for regular updates. You can unsubscribe anytime with the click of a button. Privacy policy.

Foto by Erik Mclean at Pexels