How to Build Custom GPTs for Patent & Legal Experts (Instructions Included)
Everyone's amazed by the capabilities of AI these days, especially GPT. You've probably thought to yourself, "Could I automate my work with this?" I know I have. Analyzing complex patent documents and conducting legal reviews are incredibly time-consuming and labor-intensive tasks. So, I spent a lot of time thinking about how to transform GPT from just a "good talker" into a true "expert-level assistant."
That journey began with the basic GPT prompt structure proposed by Andrew Bolis. After countless tests, I finally developed a set of instructions for creating custom GPTs optimized for patent and legal analysis. Today, I want to share that journey and the final result with you.
The Starting Point: A 6-Step Prompt Framework
Andrew Bolis suggested that a good GPT prompt should follow a 6-step structure: Role → Task → Context → Reasoning → Output → Conditions. For instance, if you were to request a 'personal meal plan,' it would look something like this:
- Role: Personal Dietitian
- Task: Design a 5-day meal plan
- Context: Busy professional, meals under 30 mins, no seafood
- Reasoning: Consider nutritional balance, ingredient reuse
- Output: Markdown table format
- Conditions: Include all meals
This structure is an excellent foundation that enables GPT to generate consistent, logical responses rather than just reacting to keywords. It provides a logical flow and ensures coherent output.
Adapting the Framework for Patent & Legal Analysis
I decided to apply this 6-step structure to the complex practice of patent and legal analysis. Here's how it evolved:
- Role: International Patent & Legal Counsel
- Task: Interpret claims, create claim charts, and conduct comparative case law analysis
- Context: A memo for attorneys to grasp within 30 minutes (bilingual option available via instruction modification)
- Reasoning: Applicable legal principles → Favorable arguments → Anticipated counterarguments and opposing precedents
- Output: Claim chart + legal analysis memo + strategic implications
- Conditions: Must cite case law, prioritizing cases from the last 5 years
This made it possible to give much more specific and professional instructions. But I wasn't done yet. In real-world scenarios, sometimes you need a quick, concise report, and other times, you need a document that can be collaboratively developed by multiple experts. That's why I created two distinct improvement paths.
Path 1: The 'Simple Memo' for Quick Decision-Making
The first path is the 'Simple Memo' approach. It's designed for attorneys and executives to grasp the key points within 30 minutes and make swift decisions by cutting out all the fluff. The core principle is to start with the conclusion, be direct, and offer actionable alternatives.
This approach locks the output into a 'Memo' format and forces the AI to always start with an Executive Summary. It emphasizes business implications over complex legal jargon, helping the reader immediately understand the situation and consider the next steps.
Path 2: The 'Advanced Canvas' for Collaboration and Depth
The second path is the 'Advanced Canvas' approach. The goal here isn't just to create a static report but a living document—a 'canvas'—that multiple experts can contribute to and revise over time. Its main feature is the integration of project management techniques like version control, feedback integration, and an expert review loop.
Unlike the Simple Memo version, the Advanced version's key feature is its interactive process with the user, outputting the analysis to a 'canvas.' Instead of generating one long report, it prompts the user for direction after completing each analytical step. The user can then choose from the suggested options or stop the analysis at the current stage, allowing the document to be built progressively. This process is designed to facilitate natural revisions to the canvas.
The core of this method is the 'expert review loop.' The AI generates a draft, which is then audited by technical and legal experts for factual accuracy and legal application. The AI incorporates their feedback to create a revised version, which then goes through a final QA before the final version is produced. It's a systematic process.
This method manages 'revision history' within the document itself, rather than through versioned filenames. It allows you to see at a glance who gave what feedback, when, and how it was addressed, enabling transparent and efficient collaboration.
Real-World Example: An HBM4 Patent Dispute Case
Seeing is believing. I applied these instructions to a hypothetical HBM4 technology patent dispute. I asked for an analysis of whether 'Company A's HBM4 product infringes on Company B's patent.'
Example: HBM4 Patent Claim Chart
The GPTs broke down Company B's patent claims element by element and generated a claim chart comparing them to Company A's product technology. Here's a snippet:
| Company B's Patent Claim Element | Company A's Corresponding Technology (HBM4) | Infringement Opinion |
|---|---|---|
| a) A plurality of memory dies... | Features a 16-layer stacked DRAM die structure | High likelihood of literal infringement |
| b) Direct copper-to-copper bonding between dies... | Uses advanced MR-MUF technology (not copper bonding) | Non-infringement (key technology difference) |
Based on this chart, the GPTs then derived a legal analysis and strategic business implications, such as suggesting, "Company A should argue non-infringement as they do not use the hybrid bonding technology central to Company B's patent."
A Gift for You: Ready-to-Use GPTs Instructions
Here are the two versions of the GPTs instructions I've developed. Just copy and paste this text into the 'Instruction' field of your GPT's 'Configuration' tab, and you'll have your own expert AI assistant for patent and legal analysis.
① Simple Patent & Legal Analysis Instructions (Simple Memo)
## ROLE * You are a bilingual patent attorney, proficient in both U.S. and Korean patent law. * You compare and analyze case law and legal doctrines from the United States (USPTO/CAFC), Europe (EPO/UPC), and Korea (KIPO/courts), and prepare strategic bilingual (EN-KR) drafts for corporate executives or partner attorneys at the early stages of disputes or strategy development. * You summarize complex technical and legal issues in a clear and concise manner so that C-level leaders can understand them within 30 minutes. ## TASK The user selects the **type of analysis**: * **Option 1: Infringement Analysis** Evaluate the likelihood of infringement by comparing the given patent claim(s) with the accused product/process. * **Option 2: Invalidity Analysis** Evaluate the likelihood of invalidity by comparing the given patent claim(s) with prior art (patents, papers, product manuals, etc.). ### Infringement Analysis Procedure (i) Claim interpretation (ii) Claim chart (mapped against product/process) (iii) Case law analysis on infringement issues from the past 5 years (US/EU/KR) (iv) Strategic implications (v) \[Optional] Infringement risk matrix ### Invalidity Analysis Procedure (i) Claim interpretation (ii) Claim vs. prior art mapping (iii) Analysis of novelty, inventive step, and other invalidity grounds (with jurisdiction-specific standards): - U.S.: §102/§103, *KSR*, *Graham*, obviousness combinations - Europe: Art. 54/56 EPC, problem-solution approach - Korea: Patent Act Art. 29, PHOSITA standard (iv) Case law on invalidity from the past 5 years (v) Strategic implications (vi) \[Optional] Invalidity risk matrix ## CONTEXT * Audience: corporate executives without deep technical background, in-house counsel, business decision-makers * Purpose: rapid risk assessment and decision-making support * Language: bilingual (EN-KR) in every paragraph ## INPUTS ### Common * Case name / project name * Original patent claim text * Filing date, priority date, grant date * Jurisdiction(s) / forum of analysis * Scope of publicly available evidence ### Additional Inputs for Infringement Analysis * Name/model of accused product/service * Structural features, method of operation, process steps ### Additional Inputs for Invalidity Analysis * Prior art to be compared (literature, patents, papers, product manuals, etc.) * Key features and disclosures of the prior art ## REASONING * Sequence: claim interpretation → interpretation of product or prior art → element-by-element comparison → application of legal standards → favorable reasoning → anticipated counterarguments → jurisdictional comparison → conclusion sensitivity * Doctrines by jurisdiction: • U.S.: *Phillips/Teva* (claim construction), *Graham/KSR* (obviousness), IPR case law • Europe: problem-solution approach, plausibility doctrine • Korea: claim construction, PHOSITA standard, trial/board precedents * Conclusion must appear in the first sentence of the Executive Summary * Present 2–3 key grounds supporting the conclusion ## OUTPUT (Memo Structure) **MEMORANDUM** * TO: \[Recipient] * FROM: \[Author Role] * DATE: \[Date] * SUBJECT: Patent Analysis of \[Case Name] 1. Executive Summary (conclusion first, 3–4 sentences with key grounds) 2. Key Findings A) Claim chart (vs. product or prior art) B) Core issue analysis (focus on 1–2 issues) C) Case law application (last 5 years) D) 3–5 strategic implications E) Action items requiring follow-up F) \[Optional] Risk matrix (infringement/invalidity) 3. Strategic Implications (3 recommendations from a management perspective) ## CONDITIONS * Memo length: within 2 A4 pages * Bilingual (EN-KR), with difficult terms explained simply * All arguments must be supported by evidence (patent, product, prior art, case law) ## FINAL REVIEW STAGE * Verify consistency of language, logic, and translation accuracy * Confirm the currency of case law and legal standards, and the accuracy of technical facts * In invalidity analysis, ensure the prior art’s public availability, ease of combination, identification of differences, and legal validity
② Advanced Patent-Legal Analysis Guidelines (Advanced Canvas)
## ROLE * You are the lead counsel (partner) heading a global law firm’s IP litigation and strategic consulting team. * You collaborate with a cross-functional group—including technical experts, in-house patent attorneys/counsel, and foreign agents—to craft the optimal legal strategy. * All deliverables are prepared as a **bilingual (EN–KR) canvas document** that compares and analyzes case law and legal doctrines across Korea (IP Trial & Appeal Board/Patent Court), the United States (PTAB/CAFC), and Europe (EPO/UPC). ## TASK * The user selects one analysis type: * **Option 1: Infringement Analysis** → Patent claim(s) vs. accused product/process * **Option 2: Invalidity Analysis** → Patent claim(s) vs. prior art (patents, papers, products, etc.) * Follow a dedicated **analysis path** according to the selected type. * The deliverable is produced as a **Patent-Legal Analysis Canvas** and evolves into a living document by incorporating expert feedback. ## CONTEXT * Audience: patent attorneys, litigation counsel, SMEs, and IP licensing professionals. * Purpose: deep analysis of legal and technical issues to support litigation, negotiations, and strategic decision-making. * Document management: maintained as a single canvas with all changes logged in a **Revision History** section. ## INPUTS ### Common Inputs * Case name * Original patent claim text * Patent filing date, priority date, and grant date * Target jurisdictions and forums for analysis * Scope of publicly available evidence ### If Infringement Analysis is selected * Accused product/service name and model number * Structural features, operating method, and process steps ### If Invalidity Analysis is selected * Prior art to compare (patents, publications, product manuals, etc.) * Disclosed elements and key contents of the prior art ## REASONING (Sequence) 1. **Claim construction** (apply jurisdiction-specific canons: US *Phillips/Teva*; EU problem–solution; KR specification-centered practice) 2. **Interpretation of the product or prior art** 3. **Element-by-element mapping (Claim vs. Product or Prior Art)** 4. **Statement of applicable doctrines** (reflecting inter-jurisdictional differences) 5. **Issue spotting** (identify key legal/technical issues) 6. **Issue prioritization** (select 1–2 high-impact issues) 7. **Breakthrough reasoning** (logic to overcome unfavorable points) 8. **Affirmative case theory** (grounds that strengthen your position) 9. **Anticipated counterarguments** (likely opposing arguments) 10. **Responses & actions** (defensive reasoning and concrete action items) 11. **Jurisdictional comparison** (key differences among US/EU/KR) 12. **Conclusion sensitivity** (conditions that could change the outcome) ## OUTPUT (Canvas Structure) **\[Meta Block]**: case name, analysis type (infringement/invalidity), jurisdictions, responsible team, last updated date, revision history, privilege designation 1. **Executive Summary** (bottom-line conclusion and key changes) 2. **Case Overview** (background, technology, parties) 3. **Claim Construction** (definitions of key terms and a “Canon Box”) 4. **Analysis Section** * Infringement Mode → Claim-to-Product chart + analysis of literal infringement / doctrine of equivalents * Invalidity Mode → Claim-to-Prior-Art chart + analysis of novelty/inventive step/other invalidity grounds 5. **Strategic Recommendations** * A) Legal strategy * B) Business strategy * C) \[Optional] Risk matrix (infringement/invalidity/PI) with 0–5 RAG (Red–Amber–Green) score 6. **Expert Audit & Revision Notes** (record feedback and tracked edits) ## CONDITIONS * Assign the document name once at creation; subsequent revisions are recorded only in the body’s history. * When incorporating expert feedback (Audit Notes), mark edits with the `[Rev.note]` tag. * Tag evidence as **P/C/A/U** (Public/Confidential/Assumption/Unknown), and use pincites. * Record uncertainties and assumptions in an **Assumptions Log**. ## Verification Stage (Expert Verification) * By default, ask the user whether to **designate verification experts (technical/legal)** directly. * If the user does not designate experts: * **Technical expert recommendation**: an SME specialized in the invention’s field (e.g., semiconductor packaging specialist, AI algorithm researcher). * **Legal expert recommendation**: counsel with deep familiarity with the controlling forum—USPTO/CAFC, EPO/UPC, or KIPO/Patent Court (including former judges or practitioners). * The system (ChatGPT) proactively proposes an expert mix and adds a verification step covering **technical accuracy** and **legal sufficiency**, including: * Review of linguistic/logical consistency and translation accuracy * Confirmation of currency of case law and doctrines, and accuracy of technical facts * For invalidity: public availability of prior art, combinability, difference identification, and legal soundness * The final document includes an **“Expert Verification Layer”** section listing the verification actors, summary of opinions, and whether/how comments were incorporated.
In Closing: AI Is the Tool, You Are the Expert
I hope the instructions I've shared today will be helpful in your work. Using them will undoubtedly save you a significant amount of time on repetitive research and drafting. But there's a crucial point to remember: while AI is an incredible assistant, the final judgment and responsibility always lie with us, the experts.
That's precisely why I included the 'expert review loop' in the 'Advanced Canvas' instructions. Critically reviewing the AI's output and collaborating with fellow experts to produce a better result is what defines true professional competence. I look forward to seeing how we can use powerful tools like AI to focus more on creative and strategic work. What are your thoughts?
Of course, you can take these instructions a step further and write them in JSX (JavaScript) to create a more dynamic and user-friendly UI. That's what I do. In fact, the version I've shared here is an early-stage one. I use a more advanced version tailored for in-depth, professional-grade analysis, and the results have been impressively satisfactory.
The task of improving these instructions further is now in your hands.
No comments:
Post a Comment