My 10-Step Claim Drafting Process

“The name of the game is the claim.”

It’s a cliché, but it’s true. That’s why I start every new patent application with the claims. But I’ve always dreaded that awkward “blank page” moment.

I’m sure you know the feeling.

You’ve digested the invention disclosure. You’ve mapped out the protection hierarchy. Yet, there you are, staring at a blinking cursor.

Do I start with a method or a system claim? What on earth should go into the generic term?

Drafting great claims requires a weird, beautiful mix of rigid structure and creative intuition. Over my 18-year career as a patent attorney, I’ve developed a workflow that kickstarts my creative brain, eliminates writer’s block, and ensures I never miss an angle.

Here it is:

(Prefer to read? Here is the step-by-step breakdown of what happens in that video…)

Project Setup

I always dedicate about an hour to fully understand the invention documentation supplied by the client. If the input is chaotic, I run this prompt to organize it before I even begin. I never proceed until every technical ambiguity and gap is resolved with the client.

A Note on Privacy: I don’t use cloud-based patent tools because they don’t support my granular workflow. Instead, I use a general-purpose LLM. To ensure absolute client confidentiality, I recently invested in a private AI computer (NVIDIA DGX Spark) to run large models locally on my desk. No data ever leaves my office.

I initialize the chat by defining a strict persona, objective, and stylistic guardrails:

# Role and Objective
You are a claim drafting assistant with the expertise of a European patent attorney, acting as a critical thinking facilitator, not a user-pleasing servant. Your goal is to guide the user in drafting a high-quality claim set through a structured, stepwise process.

# Conversation Style
- Use only supplied information. Do not invent anything. Clarify missing points. Mark assumptions.
- Communicate in clear, precise, technical language. Avoid marketing, praise, or absolutes.
- Always ask clear, open, or multiple-choice questions where appropriate. Use critical thinking techniques.
- Ask only one question at a time.
- All output must be plain text, formatted for human reading.

Analysis & Prior Art

Step 1: Understand the Invention

I load the invention documentation into the chat. But I don’t ask it to write yet. The trick is to let the AI interview me about:

• The strategic protection goals.
• The target infringing actors.
• The system boundaries.
• The conceptual unique principles of the invention, and their hierarchy.
• Any missing, unclear, or inconsistent technical details.

Once clarified, the AI generates a structured, one-paragraph invention summary to ensure we are perfectly aligned.

(Side note: I created a Custom GPT to demonstrate this specific step. I invite you to try it with a sample invention. Most attorneys are shocked that using AI feels like debating a colleague, not just prompting a text generator.)

Step 2: Prior Art

Next, I build a prior art reference list containing client-supplied documents, known approaches mentioned in the invention documentation, and the AI’s internal knowledge base.

I often ask the AI to suggest keywords, CPC classes, and competitors so I can perform a manual novelty search. Crucially, I do not allow the AI to browse the web. It is unclear which keywords might be leaked to third-party search logs, and I take no risks with security.

Finally, the AI identifies the closest prior art (CPA) using the selection standards codified in the EPO Guidelines.

The Core of the Invention

Step 3: Distinguishing Features & Technical Effects

I ask the AI to list every feature that distinguishes the invention from the CPA.

• Positive features: New components, steps, or functions.
• Negative features: Technical simplifications or the removal of complex elements (provided they yield an advantage).

I do not filter yet; quantity breeds quality here. The AI then suggests technical effects for each feature in EPO-examiner language. We then discuss and fine-tune these effects together.

Step 4: Objective Technical Problem

I run a prompt to formulate the objective technical problem (OTP) based on the classic problem-solution approach. I’ve fine-tuned this prompt to favor broad formulations over specific ones. Frankly, I rarely have to correct the output.

Step 5: Claim Strategy

Before writing a single word of the actual claim, I devise the “Claim Strategy.” This saves me from endless revisions later. We define:

• The claim category: “product” or “process”?
• The claim scope: Which distinguishing features should go in the main claim?
• The claim actor: Who executes the claim?
• The claim boundaries: Which components are required vs. merely present?

Step 6: The Main Claim

Only once the strategy is solid do I let the AI generate the main claim. Because the guardrails are set, the first draft is usually quite good. The “blank page” paralysis is gone.

I then run the draft through a rigorous battery of tests:

• Broadening by omission: Can we remove elements without breaking the inventive step?
• Broadening by generalization: The AI suggests a generalization and a concretization for each key term for me to select from.
• Clarity Check: The AI puts on the EPO examiner hat and raises clarity objections.
• Single-actor infringeability.
• Detectability.
• Possible Workarounds.

I iterate until I have a main claim I am fully comfortable with.

The Finish

Step 7: Additional Main Claims

If the invention involves distinct phases (training vs. inference), distinct actors (server vs. client), or alternative solutions, I generate independent claims for those scopes and run them through the same testing gauntlet as Step 6.

Step 8: Other-Category Claims

For computer-implemented inventions, I generate the standard trifecta (method, data processing system, computer program) using wording compliant with the EPO Guidelines.

Step 9: Fallback Positions

I generate an expanded set of dependent claims using the “unused” distinguishing features from Step 3. I revise these until satisfied.

Step 10: Compiling the Result

Remember, this entire process happens inside a single chat thread. Once finalized, I run one last prompt to generate the deliverables:

1. The complete Claim Set.
2. A “Claim Strategy Summary” (listing the CPA selection, key technical effects, OTP, and key decisions).
3. “Notes for the Description” (aspects required for the detailed description that aren’t obvious from the claims).

Deliverables #2 and #3 are helpful for the subsequent drafting of the patent description (more on that soon).

Copy My Process

You are welcome to copy this workflow. However, if you don’t have the time to build and fine-tune a prompt library from scratch, you can grab my Patent Drafting Toolbox.

• The “Sovereign” Tier: Full access to my raw prompts for local use.
• The “Custom GPT” Tier: If you don’t want to fumble with local AI hardware, this gives you instant access to the process logic via a Custom GPT or Gemini Gem.

[Get the Patent Drafting Toolbox Here]

Hope this helps you stay sharp,
Bastian