It seems that the latest GPT models and other LLMs are getting better at deeper patent analysis or, at a minimum, are becoming more willing to take on such analytical tasks. It’s worth trying out the new capabilities, so the 2025.3 release of ClaimMaster enhances our +Drafting patent drafting tools with the ability to supplement GPT/LLM prompts with the attached Word or PDF documents to have AI perform patent content analysis. We’ve also added examples of GPT/LLM prompt templates that are specifically tailored for patent claims and specification analysis. These templates support many section-specific replacement fields to extract and populate prompts with claims, detailed description, and other sections from the attached documents.
Here are some of the AI patent content analysis/drafting tasks you can now perform with the new GPT/LLM prompts in ClaimMaster:
Generate draft applications from invention disclosures
You can now ask the configured GPT/LLM service to turn the attached invention disclosure forms (or other similar documents) into quick patent application drafts. Notably, you can pre-configure your prompt with a desired patent application outline. Even better, you can further link the prompt to a pre-configured document workspace that provides examples of previously drafted applications to provide your GPT/LLM with as much context as possible about the desired output content and style.
Find unclaimed subject matter in applications based on the existing claims
You can now pass one or more claims to the configured GPT/LLM service and ask it to compare those claims to the existing disclosure (e.g., Detailed Description section). In particular, the LLM may help you identify any unclaimed subject that still remains in the disclosure for Continuation applications. You can either attach the original application for analysis and have ClaimMaster pull the detailed description or other sections from the currently open Word document.
Analyze differences or similarities between claims and prior art
Likewise, you can ask the configured AI service to perform deeper patent content analysis and compare the provided claim(s) to a particular prior art disclosure or a text section. If the reference section is prior art, then the LLM will attempt to determine whether the prior art reads on your claims. If the reference section is your own application, then the LLM will determine whether you have support for your claims in the original application. You can further modify the prompt depending on whether you need to distinguish or find similarities with the disclosure text.
Identify key concepts in claims and generate prior art search strings
You can ask the configured GPT/LLM service to identify core concepts in analyzed claims and also turn those concepts into keywords search strings for public patent sites, such as Google Patents and the USPTO’s Patent Public Search. You can further adjust prompts to generate search strings based on formatting and Boolean/proximity operators specific to those sites.
Summarize the attached documents
Finally, as you are now able to attach various documents to GPT/LLM prompts, you can ask the configured LLM to summarize those documents for you. For example, you can attach a prior art reference and ask LLM to generate a 3-4 paragraph summary of that reference or its claims.
In conclusion
So how reliable are the latest LLMs (e.g., GPT-4o or LLama3) for deep patent claim and specification analysis or end-to-end application drafting? They appeared to do fairly well for some tasks in our testing, but you should always double-check the results (same goes for all AI generated content). We don’t expect these tools to completely replace attorneys, but they may be helpful for brainstorming or getting a quick second opinion when you are stuck. For example, if you are having a hard time finding additional subject matter in the application to claim for a continuation, it’s worth running the document and claims through GPT/LLM prompt to see if AI can find something that you’ve missed. Similarly, GenAI could save you time and do an acceptable job in turning attached invention disclosures into initial application drafts, especially when AI has access to a document workspace filled with examples from previous applications.
Here’s more information on the specific features discussed in this article:
We’ve also integrated with USPTO’s ODP system in this release to quickly pull various application biblio and IFW data. To learn more, click here.