My Exact 10-Step Process for Drafting Patent Claims

00:00:00: I am going to show you the exact process how I draft claims as a patent attorney with eighteen years experience.

00:00:06: Back in the days I always started with a blank word document, but this has two major problems.

00:00:11: Number one is writer's block.

00:00:13: So when the cursor was blinking at me, it always took me a lot of time, you know, to figure out, should I start with a method claim?

00:00:22: Should it be a system claim?

00:00:23: What should the generic term be?

00:00:25: And it took me quite some time to get into the flow.

00:00:28: Number two is lack of structure.

00:00:30: So when I started from a blank document, I found that, you know, the whole claim drafting session more or less turned into an artistic exercise at the risk of forgetting to ask, you know, a number of questions, which are, really critical, which you have to ask yourself before you even write the first word.

00:00:51: So if you don't want to make my mistakes, copy the process, here it is.

00:00:54: First of all, the project setup.

00:00:56: This is step zero, if you will.

00:00:58: So... Typically, I would take like an hour or so to really go through the invention description through the materials I received from my client and really understand the invention, right?

00:01:11: Sometimes the documentation might be a bit chaotic or there may be certain gaps in it, right?

00:01:18: In this case, I like to run the documentation through a special prompt I designed, which then, you know, improves the structure and also consistency.

00:01:28: I'm going to put a link to this prompt in the description for you.

00:01:31: So at this point, we are now assuming that we have an invention description.

00:01:35: on the table in which all technical facts are in there and which is consistent and complete.

00:01:41: Next thing I do is I start up my AI chatbot and I'm putting in a first prompt which just has the purpose, you know, to set the AI up on the wrong track to give it a role and a proper task description and some general guard rails.

00:01:59: You can see that it now introduces itself and it asks me to start by providing the source material, which is the invention and some prior art optionally and also optionally some focus notes.

00:02:13: And this is actually step one.

00:02:14: already understanding the invention, right?

00:02:17: So I'm going to input the invention description here.

00:02:21: And now I'm having a structured conversation with my AI about the invention, right?

00:02:27: You can see here that the AI has first asked about the protection goal.

00:02:31: So what is the primary protection goal for this invention?

00:02:34: Broad protection, narrow technical, a hybrid.

00:02:37: So I can now.

00:02:38: chat with the AI and describe what, you know, the strategic goal behind this application is.

00:02:44: The AI will also ask me about the target infringing actors, the system boundaries, and, you know, the conceptual unique principles of the invention, as well as the hierarchy, the relationship between the building blocks, so to say.

00:02:58: And also it will ask me about any, you know, missing, unclear or inconsistent technical detail.

00:03:03: I'm going to skip ahead in this demo.

00:03:05: And now after all these questions have been clarified, my chatbot outputs structured invention summary.

00:03:13: This makes sure that, you know, I can check whether the AI has understood everything and whether we have a common understanding of the invention.

00:03:20: By the way, if you want to try this out for yourself, I have a demo of this except first step of my workflow as a free custom GPT on my website.

00:03:28: I will put a link into the description.

00:03:30: Step two of the process is the prior art.

00:03:33: So we can see that my chatbot now looks which prior art it has on hand.

00:03:39: There is no documents uploaded, but there is some prior art mentioned in the description.

00:03:45: So my chatbot even offers me to ask me first to upload any documents.

00:03:52: It also offers me to do a kind of a top-up search in its own knowledge base, and it can suggest, you know, search strings, keywords, CPC classes, things like that, so that I can do a web search.

00:04:05: Once we have all the prior art collected, so to say, right, the chatbot will go on and select the closest prior art based on the typical, you know, EPO framework.

00:04:15: So we have that here.

00:04:17: Step three is about the distinguishing features and their technical effects.

00:04:20: So next prompt.

00:04:22: And now my chatbot will help me, you know, figure out, yeah, what are the features that really set the invention apart from the closest prior art.

00:04:33: right and then we will work on their technical effects.

00:04:36: So what I do is I'm also using a special prompt for that so that I have get a first suggestion by the AI.

00:04:45: So we can see here that it is now doing a table of all the key features listed in the invention description and it checks whether those are disclosed already in the prior art.

00:05:02: And then it puts out a list of distinguishing features, which make the invention differ over the prior art.

00:05:09: So we have a nice table here, right?

00:05:12: The AI identified like.

00:05:14: eight distinguishing features describes their function and also gives some idea of the technical effect of the respective feature.

00:05:25: The technical effects are formulated in a way that you know they resonate with an EPO examiner and in a real case I would now have a discussion with my AI you know tweaking the distinguishing features maybe changing their their ranking and so on until I'm fully satisfied.

00:05:41: Now that we've identified the closest prior art and all the distinguishing features with their technical effects, we can formulate the objective technical problem, right?

00:05:49: We're talking about EPO style problem solution approach here.

00:05:53: Here you have it.

00:05:54: So this is kind of the EPO textbook style formulation of the objective technical problem in this case.

00:06:04: I finetuned my prompt so that it's, you know, formulates the objective technical problem rather broadly than, you know, overly specific.

00:06:14: By the way, in case you're wondering what strange kind of chatbot I'm using here, it's a local solution.

00:06:20: So I recently bought a small AI supercomputer, which is called the DGX Spark from NVIDIA, which allows me to run quite heavy open weight, large language models on my desk without sending data to any third party outside my computer network.

00:06:39: And if you want to copy my prompts, I also have good news for you.

00:06:42: in the sovereign tier of my patent drafting toolbox, you'll get the exact, you know, raw source code.

00:06:48: So to say of all the prompts I'm using in this example, so check it out.

00:06:51: I'll put a link into the description.

00:06:53: Okay, step five is the claim strategy.

00:06:56: So what I'm doing is, Before I actually write a single word for my independent claim one, I always devise what I call a claim strategy.

00:07:08: And this claim strategy includes more strategic decisions like the claim category.

00:07:13: Is it a product or is it a process?

00:07:15: The claim scope, which is basically, you know, the the selection of those distinguishing features from the step before which goes into the main claim.

00:07:24: Then the claim actor, so you know from which perspective should the claim.

00:07:29: be written and also the claim boundaries, so which components and functions are required in the sense that they are really part of the invention and which merely need to be present for the invention to work.

00:07:40: You can see that the AI has some suggestions here.

00:07:44: And again, I would now have a discussion with my chatbot to really, you know, find you and revise and finalize my claim strategy.

00:07:51: I found that getting this right saves me endless revisions of, you know, the actual claim wording later on.

00:07:57: You can see that my AI even asks me questions, for example, about the claim category and things like that.

00:08:05: So this is meant to, you know, spark a discussion trigger my own critical thinking here.

00:08:10: For the sake of this example, let's say this strategy is all okay.

00:08:13: And then the AI will suggest the main claim.

00:08:19: So we could now have a discussion about this.

00:08:21: I will just fine.

00:08:23: for the sake of this example.

00:08:25: And now we are entering a structured, you know, thinking process how this main claim can be improved.

00:08:33: So the first step is, can it be broadened by omission?

00:08:37: So is it possible to remove certain elements or steps from the claim without making it overly broad?

00:08:44: Then the AI suggests for all the key terms, both a more generic term and a more specific term.

00:08:51: So I can dial the abstraction layer for each key term of the invention.

00:08:56: We also talk about clarity.

00:08:58: Here I let the AI take on the pattern examiner head, if you will, and raise some clarity objections.

00:09:05: then we check whether there is a single actor infringibility, whether the claim can be detected from the outside and if there are possible workarounds.

00:09:14: And this step typically takes, you know, anything between ten to maybe one hour, right, until the main claim is really spot on and passes all those tests.

00:09:26: Once the main claim is set, we can talk about possible additional main claims.

00:09:31: Sometimes it makes sense, you know, to have multiple independent claims with different scopes.

00:09:35: So for example, first case when the invention involves multiple distinct phases, you know, like in AI, a training phase and a configuration phase and then an inference phase.

00:09:49: Second case group when the invention involves distinct actors, like in a client server architecture, base station and user device provider customer, things like that.

00:10:00: And thirdly, when the invention involves distinct alternatives for you know, the same objective technical problem.

00:10:06: As you can see, I can run prompt to suggest these and then again, have a discussion with my AI about the exact selection of these additional main claims.

00:10:17: In addition, the AI also suggests other category claims, which in my cases, I'm working only on digital innovations, right?

00:10:26: Is typically, you know, the EPO trifecta of a method claim, a data processing system claim and a computer program claim.

00:10:33: Once we have the main claims all set up, let's talk about fallback positions.

00:10:38: And here I run another prompt, which then goes through the list of distinguishing features.

00:10:45: It turns.

00:10:46: every distinguishing feature which we haven't included in the main claim into one fallback position, right?

00:10:52: So it will write a group of dependent claims for that supplemented with, you know, interesting implementation details from the description.

00:11:01: We can see that here.

00:11:02: So the AI first suggests again, some kind of claim strategy on a more abstract level, right with certain groups of features.

00:11:12: And then I can now have a discussion with my AI.

00:11:15: you know, to rearrange the groups or change them completely and once that's fine.

00:11:21: the AI will write out the dependent claims.

00:11:24: So you can see it's now writing out the dependent claims and now we're almost finished.

00:11:28: So last thing I do is I let the AI, you know, clean up everything and compile all the intermediate results.

00:11:35: And here we have the final results.

00:11:37: We have first of all the full claim set, you know, copy and paste ready.

00:11:41: Then the AI gives me a claim strategy summary, which, you know, summarizes the main findings of the thinking process, like the invention summary.

00:11:50: the closest prior art selection, the list of distinguishing features and the technical effects, the OTP, the claim strategy and so on.

00:11:59: This is helpful later when I'm actually drafting the patent description.

00:12:03: And thirdly, the AI gives me so-called notes for the description, which, you know, random facts, which need to be included into the patent description, like examples or definitions, which are not directly apparent from the claim.

00:12:19: alone.

00:12:20: So that's my claim drafting workflow.

00:12:21: Hope this was helpful.

00:12:24: Feel free to copy my process.

00:12:26: If you don't have the time to set up your own prompt library from scratch, you can check my patent drafting toolbox.

00:12:33: I'll put a link into the description.

00:12:35: There is the sovereign tier, which comes with full access to all my raw prompts.

00:12:40: And if you don't feel like fumbling around with local AI, there's also a custom GPT tier where you can run this whole process in inside a custom GPT.

00:12:50: Hope this helps you.

00:12:51: stay sharp.

00:12:52: See you next time.