Mastering Patent Claim Construction with LLMs (3): Injecting Guidelines for Claim Chart Drafting

Practical Framework Part 3

Mastering Patent Claim Construction with LLMs (3)

Injecting Claim Chart Drafting Guidelines Drawn from Practical Experience

The next step is a research phase aimed at establishing the format of the final deliverable.

Rather than organizing claim construction results into a simple narrative report, I prefer to structure them in the form of a Claim Chart specifically designed for claim interpretation. This is not merely to make the output look polished, but to secure a practical tool that can be used under a consistent standard in all downstream tasks, including infringement analysis, invalidity review, and opinion drafting.

1. Why Claim Chart Standards Must Be Established First

There is one important point here. A Claim Chart is not just a “format for organizing results,” but a core framework that determines the quality of the interpretation itself.

The problem, however, is that the method for decomposing claims, the standards for extracting elements, and the actual drafting approach are not systematically organized in textbooks or case law. In other words, this is an area where methodology is built largely 👉 through practical experience.

Through handling a wide range of cases, I have gradually developed my own standards for claim decomposition, element extraction, and chart structure in a way that is suitable for claim construction.

2. Teaching the LLM the “Drafting Method” First

At this stage, what I do is simple. I organize the standards I use in practice 👉 the way I would teach a junior associate, and then register them as a source in NotebookLM.

A Critically Important Point

This step is not simply about “adding reference materials.” It is the process of pre-training the LLM on the interpretive framework and the output format so that it will reason under the same standards in all later interpretation tasks.

• This is the step where you inject, before adding any source materials, how the analysis should be conducted and how the results should be organized.
• If this step is skipped, the LLM will analyze each task under a different set of standards, which in turn prevents the overall work product from losing consistency.

3. Claim Chart Drafting Guidelines (Draft)

Below is a draft of the practical Claim Chart drafting guidelines that I prepared and asked the model to review.

DRAFT INSTRUCTION

**1. Methods for Claim Decomposition and Extraction for a Claim Chart Used in Claim Construction** Claims should be broken down into the smallest units that serve as the basis for invalidity and infringement analysis. The language of the claim should be separated according to context, while also being divided into sub-elements in a way that clearly reveals the invention’s distinctive features. The specific methods of decomposition and extraction are as follows. * **Structure- and Function-Based Decomposition:** Break the claim down into **structure**, which reflects the physical form of the elements and how they are connected, and **function**, which reflects the operation and role performed by each element. * **Identifying Organic Relationships:** Go beyond merely listing the components and derive the relationships showing how each component connects with the others and contributes to solving the technical problem. * **Extracting Interpretive Terms and Limitations:** From among the separated sub-elements, extract as the key interpretive targets those terms that **specifically require claim construction or are directly tied to the invention’s distinctive features**. In particular, major extraction targets include terms expressed functionally (for example, “means for ~”) that may need to be examined to determine whether limiting interpretation to the embodiments is required, as well as terms to which the principle of claim differentiation may apply because of dependent claims. **2. The Most Desirable Methodology for Drafting a Claim Chart** The most practical and desirable output format for visualizing claim construction results in a logical and clear manner is to prepare a **Claim Chart** in tabular form. For that purpose, I propose the following table format composed of **six items (columns)** as the output format. | No. | Claim Language | Claim Element (Interpretive Target) | Specification / Drawing Support | Interpretation Result and Legal Analysis (from the PHOSITA Perspective) | Notes (Prosecution History / Limiting Construction) | | :-- | :--- | :--- | :--- | :--- | :--- | | 1 | Decomposed claim language | (e.g., screw) | [paragraph number], [figure number] | (e.g., a low-speed pressing member having a spiral structure) | (e.g., vertical-type limitation based on an argument submission) | **[Drafting Guide and Explanation for Each Column]** * **Claim Language:** Set out verbatim the minimum unit of claim language as decomposed in a way that clearly reveals the invention’s distinctive features. * **Claim Element (Interpretive Target):** Extract from the decomposed language the core target terms or limitations that especially require interpretation or are written as functional expressions (e.g., “means for ~”). * **Specification / Drawing Support (Mapping of Intrinsic Evidence):** Identify and map the specific [paragraph numbers] and [figure numbers] in the detailed description of the invention that define or support the meaning of the term. * **Interpretation Result and Legal Analysis:** From the perspective of a person having ordinary skill in the art, apply intrinsic evidence first and clearly state the objectively and substantively established technical meaning of the claim element. * **Notes (Review of Prosecution History and Limiting Construction):** Record legal points of note, such as any intentional exclusion (prosecution history estoppel) reflected in the submission of arguments or amendments during prosecution, or any reason why, due to a clear disavowal of claim scope in the specification, limiting construction to the embodiments should apply.

This type of Claim Chart format allows intuitive one-to-one textual mapping between the relevant product (or prior art) and the claim elements in later infringement litigation or invalidity proceedings, making it the most effective practical analytical tool for preventing errors caused by improper narrowing or overbroad claim interpretation.

4. The Process of Validation → Refinement → Redesign

Rather than simply using the above guidelines as-is, I validated them in the following way.

“Are these guidelines sufficiently valid under actual legal doctrine and case law standards?”

To answer that question, I instructed the model in the chat to reflect the points requiring refinement and present again the most appropriate concrete Claim Chart format and drafting guide.

5. Refined Claim Chart Drafting Guidelines (Final Version)

The result is as follows, and this too was saved as a memo and then converted back into a source and integrated into the knowledge base. This process of turning good results into knowledge that can be continuously reused rather than a one-off answer is extremely important.

FINAL FRAMEWORK

**1. Methods for Claim Decomposition and Extraction for a Claim Chart Used in Claim Construction (Reflecting Special Issues)** When decomposing claims to apply the All Elements Rule (AER), which is central to infringement and invalidity analysis, it is essential to go beyond a simple listing of claim language and instead perform a more multidimensional extraction that reflects the legal characteristics of **special types of claims (functional claims, PbP claims, and numerical limitation claims)**. * **Identifying Structure, Function, and Organic Relationships:** Break the invention down into the physical structures that make it up and the functions those structures perform, while clearly deriving the relationships showing that each element does not exist in isolation but is organically connected with other elements to solve the technical problem. * **Extracting Targets for Limiting Construction of Functional Claims:** Because elements expressed in abstract or functional terms, such as “means for ~” or “~ step,” carry a risk of making the claim scope unduly broad, they should be extracted as targets for focused review to determine whether there are clearly unreasonable circumstances requiring the claim to be limited to the embodiments disclosed in the specification. * **Extracting the “Structure / Properties” of PbP (Product-by-Process) Claims:** Even when a manufacturing process is recited in the claim, the subject of the invention is still a “product,” so the manufacturing process itself should not be mechanically extracted as an independent element. Instead, the true technical element to be extracted is the “structure or properties of the product” ultimately defined by that manufacturing process. * **Extracting the “Critical Significance” of Numerical Limitation / Parameter Inventions:** Where a specific numerical range or a newly created parameter is involved, it should be separated as a core interpretive target in order to assess whether the specification enables the full claimed range without undue experimentation (support requirement), and whether the claimed range or parameter shows a remarkable effect (critical significance) that distinguishes it from the prior art. **2. The Most Desirable Concrete Claim Chart Format (Top-Tier Practical Template)** By upgrading the existing six-column structure, I propose a **seven-column framework that can fully map both Doctrine of Equivalents (DOE) defense / attack logic in cases of literal non-infringement and the analytical guidance for special types of claims**. | No. | Decomposed Claim Language (Including Organic Relationships) | Core Interpretive Target (Special Issue: Functional / PbP / Numerical) | Specification / Drawing Support (Mapping of Intrinsic Evidence) | Objective Technical Meaning and Result of Literal Interpretation (from the PHOSITA Perspective) | DOE Comparison: Identification of the Principle for Solving the Problem | Defense Logic: Prosecution History Estoppel and Grounds for Limiting Construction | | :-- | :--- | :--- | :--- | :--- | :--- | :--- | | 1 | (e.g., a cleaning unit that supplies cleaning water by electrolyzing filtered water) | cleaning unit (functional claim) | [paragraph 0045], [figure 3] | (e.g., an internal module that performs physical cleaning through electrodes without sterilizing chemicals) | (e.g., the principle of maximizing eco-friendly cleaning efficiency by excluding chemicals) | (e.g., conscious exclusion of a “chemical additive method” during argument submission) | | 2 | (e.g., a tablet manufactured by direct compression) | manufactured by direct compression (PbP claim) | [paragraph 0022] | (e.g., a porous tablet structure having a 15% inter-particle porosity formed by the direct compression process) | (e.g., the principle of controlling disintegration speed by adjusting porosity) | (e.g., structurally / physically different from tablets made by wet granulation) | | 3 | (e.g., a lens assembly having TTL ≤ 6.5 mm) | TTL ≤ 6.5 mm (numerical limitation) | [paragraph 0110], [Experimental Example 2] | (e.g., the physical limit of a miniaturized lens with a total track length of 6.5 mm or less) | (e.g., the principle of achieving an ultra-short focal length through refractive-index redistribution) | (e.g., target for lack of enablement across the full range of 6.5 mm or less) | **3. [Drafting Guide and Explanation for Each Column]** * **Decomposed Claim Language:** Set out verbatim the claim language decomposed into the minimum unit defining the scope of the invention, while describing it in a way that reveals not just a simple list but also the physical and functional relationships among the elements. * **Core Interpretive Target (with Special-Issue Label):** From the decomposed language, extract the key terms that present issues—such as functional expressions (“means for ~”), product-by-process (PbP) language, or numerical limitations / parameters—and identify the type in parentheses so the focus of the analysis is clear. * **Specification / Drawing Support (Mapping of Intrinsic Evidence):** Map the specific [paragraph numbers] and [figure numbers] in the specification that show where the term is defined under the lexicographer rule, where embodiments support a functional claim, where a manufacturing process in a PbP claim causes a specific change in physical properties, or where the technical threshold for a numerical limitation is demonstrated. * **Objective Technical Meaning and Result of Literal Interpretation (from the PHOSITA Perspective):** From the perspective of a person having ordinary skill in the art, apply the principle of construing the claim in light of the specification and describe the objectively and reasonably derived substantive meaning of the element. In the case of PbP claims, the description should not merely repeat the manufacturing process itself, but should clearly translate the structure or properties defined by that process. * **DOE Comparison: Identification of the Principle for Solving the Problem (New):** This column is used to assess possible infringement under the Doctrine of Equivalents when differences are found in a literal one-to-one comparison with the accused product. By considering both the content of the specification and the prior art existing at the time of filing, this column states in advance the “core of the technical idea underlying the specific means for solving the problem unique to the patented invention” (i.e., the principle for solving the problem) to which the relevant element belongs. * **Defense Logic: Prosecution History Estoppel and Grounds for Limiting Construction (New / Refined):** Record the history of any intentional exclusion (prosecution history estoppel) of specific subject matter reflected in arguments or amendments submitted throughout the prosecution of the original application and any divisional applications. Also record whether a functional claim should be limited to a specific embodiment because a broader reading would be clearly unreasonable in light of the specification, or whether a numerical limitation invention may be invalid for failure to satisfy disclosure requirements (such as enablement) because the full claimed range cannot be practiced without undue experimentation.

6. The Core Meaning of This Stage

The most important insight to gain at this stage is the following.

“A Claim Chart is not merely an organizational tool, but a structural mechanism that controls the accuracy of claim interpretation.”

And when using LLMs, the key is not simply getting good results, but building a system in which the results are always generated under the same standards.

7. Key Points for Practical Application

When this process is applied directly in practice, it can be summarized as follows.

1. First, establish the “drafting format” (the structure of the table).
2. Next, inject the “interpretive standards” (legal doctrine and drafting guidelines).
3. Then, input the “source materials” (the specification and evidence).
4. Finally, carry out the “claim interpretation and chart drafting”.

Most people reverse this order. Because they start by asking for the interpretation right away, the results become unstable and lose consistency.

This stage is not simply about creating a Claim Chart, but about designing how the LLM will think. When this work is done properly, all later analytical work becomes far more stable.

© 2026 ChinSu Lee. All rights reserved.

Mastering Patent Claim Construction with LLMs (2): Enhancing the Claim Interpretation Framework and Expanding the Knowledge Base

Mastering Claim Construction with LLMs: Refining the Framework and Expanding the Knowledge Base

How practical experience and AI can work together to evolve beyond simple analysis into an expert-level knowledge asset

Building on the previous post, this time I focused on refining the claim construction framework and expanding the surrounding knowledge system.

In the previous post, I supplied additional practical heuristics that I had separately compiled from experience and instructed the model to compare and verify them against the existing case law analysis. The framework I asked it to review was the “Five-Step Claim Construction Framework under Korean Case Law.”

System Directive

This is a five-step interpretive structure that takes the claim-centered approach (literal interpretation) as the governing principle, while supplementarily referring to the specification and drawings to determine the technical meaning in an objective and reasonable manner. Five-Step Claim Construction Framework under Korean Case Law 1) Step 1 (Principle of Literal Interpretation): The scope of the patent right is first defined based on the ordinary meaning of the terms used in the claim, as understood in the relevant technical field. 2) Step 2 (Reference to the Specification and Drawings): From the perspective of a person having ordinary skill in the art (PHOSITA), the claims are interpreted in light of the overall context of the specification. If the applicant explicitly defined a term in the specification (the lexicographer rule), that definition takes precedence. More specifically, the meaning of the term is determined through the detailed description of the invention so that the intended function and operation of the invention are properly implemented. The Korean Supreme Court has made clear that, when consulting the specification, courts must go beyond simply reading the surrounding context and must instead objectively and reasonably examine the “technical meaning” that the language was intended to convey, including the problem-solving principle and technical effect. 3) Step 3 (Prohibition on Undue Narrowing or Expansion): Even when consulting the specification, the claims must not be improperly narrowed based on specific embodiments, nor unreasonably expanded beyond what is supported by the specification. In practice, one of the most common errors in Korean patent litigation is confusing “interpreting claims in light of the specification” with “importing limitations from the specification into the claims.” The Supreme Court has taken a firm position against improper narrowing based on the description of the invention (for example, in the Cream case and the Display Structure case). 4) Step 4 (Reference to Prosecution History and Estoppel): If a particular feature was consciously excluded or narrowed during prosecution, later attempts to reverse course and broaden the interpretation are restricted. 5) Step 5 (Application of the All-Elements Rule and Doctrine of Equivalents): In infringement analysis, the all-elements rule (AER) applies, while the doctrine of equivalents (DOE) serves as a supplementary doctrine to capture design-arounds.

As a result, I received a favorable assessment regarding its consistency with the case law, along with suggestions for areas that still needed improvement. The purpose of this stage was not simply to add more information, but to verify whether the existing framework was sufficiently supported for practical use and systematically reinforce the parts that were still lacking.

1. Method of Knowledge Expansion: “Turning Results Back into Sources”

The key method used in this round of work was as follows.

• Save meaningful analytical outputs generated during the conversation as NotebookLM Studio notes
• Convert those notes back into sources and reinject them into the existing knowledge base

In other words, instead of merely consuming answers, this creates a structure in which useful outputs are continually accumulated, refined, and converted into knowledge assets. As this process repeats, the LLM gradually comes closer to functioning like a case-specific expert model.

2. Results of Framework Validation: Strong Structure, but Foundational Legal Support Still Needed

Based on an analysis of 23 sources, the five-step claim construction framework I had previously developed was confirmed to be a highly sophisticated and practical structure. At the same time, however, NotebookLM pointed out several important issues.

The current source set is overly concentrated on specialized issues such as product-by-process claims, numerical limitation inventions, and the fifth requirement of the Japanese doctrine of equivalents. As a result, while the framework’s overall skeleton is strong, the general body of Korean case law needed to support that structure is still relatively thin.

3. Summary of Areas Needing Reinforcement by Step

(1) Steps 1–2: The Lexicographer Rule

• Current status: The principle that claim terms may be interpreted according to definitions in the specification is partially reflected, but there is not yet enough explicit discussion of the lexicographer rule itself.
• What needs to be reinforced: The requirements for recognizing the rule and the level of “clarity” needed to displace the ordinary meaning of a term.
• Further research direction: Supreme Court decisions and academic materials analyzing the standard for clear claim-term definitions.

(2) Step 3: Interpretation of Functional Claim Language

• Current status: The possibility of narrower interpretation is partially reflected, but the applicable standard remains unclear.
• Core issue: It is difficult to determine when the general rule of literal interpretation should apply and when an exception permitting narrower interpretation should be recognized.
• Further research direction: A comparison of cases that accepted limiting interpretation versus those that rejected it, with emphasis on the underlying factual circumstances.

(3) Step 4: Prosecution History and Estoppel

• Current status: Some recent issues are well reflected, but the broader doctrine of prosecution history estoppel is still underdeveloped.
• Core issue: Compared with Japanese authorities, there is still not enough Korean case law grounding.
• Further research direction: Leading cases on conscious exclusion and estoppel, decisions addressing the full course of prosecution responses, and whether the dedication doctrine has been recognized in Korea.

(4) Step 5: AER and the Doctrine of Equivalents

• Current status: The explanation of the all-elements rule (AER) is still very limited, and the criteria for applying the doctrine of equivalents are only summarized at a high level.
• Core issue: There is still a lack of practical, case-usable standards for application.
• Further research direction: Cases involving omission-type infringement and indirect infringement, standards for determining whether the same problem-solving principle is present, and the way prior art should be considered.

4. An Important Insight: “Special-Issue Data Is Actually a Strength”

One interesting takeaway is that, despite the gaps noted above, a substantial portion of the current sources focuses on high-difficulty issues such as product-by-process claims, numerical limitation inventions, and in-depth Korean and Japanese doctrine-of-equivalents cases. That is not a weakness. If anything, it is a strength. Most practical frameworks cover only the general rules and tend to break down when they encounter specialized issues.

5. Proposed Expansion of the Framework

Taking that into account, the existing five-step structure could be expanded as follows.

“Step 6: Interpretation of Special Claim Types (PBP Claims and Numerical Limitation Inventions) and the Limits of Applying the Doctrine of Equivalents”

Adding this step would allow the framework to evolve beyond a simple theoretical summary into a practice-oriented structure capable of handling high-complexity cases.

6. Supplementing the Proposed Sources and Closing Remarks

I copied the reinforcement points identified above directly back into NotebookLM’s source window and then activated the deep research function to gather additional case law and doctrinal materials that had been missing. Through this process, I was able to strengthen the weaker parts of the existing framework and expand the knowledge base in a more balanced way.

What This Stage Ultimately Accomplished

1. Verified the structural completeness of the framework
2. Identified areas where doctrinal support was still weak
3. Confirmed the existing knowledge bias (specialized issues vs. general doctrine)
4. Established a repeatable process for knowledge expansion

In claim construction, what matters is not simply gathering a large volume of materials, but whether the framework is actually built to digest and organize those materials effectively. Strong results come less from the model itself and more from the design of the knowledge structure and the reinforcement process.

© 2026 ChinSu Lee. All rights reserved.

Can AI Be Your Paralegal? (Only if You Follow This 5-Step Verification Process)

 

Blogging_CS · Sep 20, 2025 · 10 min read

Generative AI promises to revolutionize the speed of legal research, but a critical pitfall lies hidden beneath the surface: “AI hallucinations.” Because AI can fabricate non-existent case law that looks authentic, legal professionals are now facing the paradox of spending more time verifying AI outputs than it would have taken to draft the work themselves.

This isn’t a hypothetical concern. In Mata v. Avianca, a case in the Southern District of New York, attorneys faced sanctions for submitting a brief containing fake judicial opinions generated by AI. Even more striking is Noland v. Land, where the California Court of Appeal sanctioned an attorney for filing a brief in which 21 of 23 case citations were complete fabrications. The penalty was severe: a $10,000 fine, mandatory notification to the client, and a report to the state bar.

These rulings send a clear message: before any discussion of technology, the user’s attitude and responsibility are paramount. Attorneys (including patent attorneys) have a fundamental, non-delegable duty to read and verify every citation in documents submitted to the court, regardless of the source. With the risk of AI hallucinations now widely known, claiming ignorance—“I didn’t know the AI could make things up”—is no longer a viable excuse. Ultimately, the final line of defense is a mindset of professional skepticism: question every AI output and cross-reference every legal basis with its original source.

---

A 5-Step Practical Workflow for Risk Management

Apply the following five-step workflow to all AI-assisted tasks to systematically manage risk.

1. Step 1: Define the Task & Select Trusted Data
   

   Set a clear objective for the AI and personally select the most reliable source materials (e.g., recent case law, statutes, internal documents). Remember that the “Garbage In, Garbage Out” principle applies from the very beginning.

2. Step 2: Draft with RAG (Retrieval-Augmented Generation)
   

   Generate the initial draft based on your selected materials. RAG is the most effective anti-hallucination technique, as it forces the AI to base its answers on a trusted external data source you provide, rather than its vast, internal training data.

   Use Case:

   • Drafting an Initial Case Memo: Upload relevant case law, articles, and factual documents to a tool like Google's NotebookLM or Claude. Then, instruct it: “Using only the uploaded documents, summarize the court's criteria for ‘Issue A’ and outline the arguments favorable to our case.” This allows for the rapid creation of a reliable initial memo.

3. Step 3: Expand Research with Citation-Enabled Tools
   

   To strengthen or challenge the initial draft's logic, use AI tools that provide source links to broaden your perspective.

   Recommended Tools:

   • Perplexity, Skywork AI: Useful for initial research as they provide source links alongside answers.
   • Gemini's Deep Research feature: Capable of comprehensive analysis on complex legal issues with citations.

   Pitfall:

   • Source Unreliability: The AI may link to personal blogs or irrelevant content. An AI-provided citation is not a verified fact; it must be checked manually.

4. Step 4: Cross-Verify with Multiple AIs & Refine with Advanced Prompts
   

   Critically review the output by posing the same question to two or more AIs (e.g., ChatGPT, Gemini, Claude) and enhance the quality of the results through sophisticated prompt engineering.

   Key Prompting Techniques:

   • Assign a Role: “You are a U.S. patent attorney with 15 years of experience specializing in the semiconductor field.”
   • Demand Chain-of-Thought Reasoning: “Think step-by-step to reach your conclusion.”
   • Instruct it to Admit Ignorance: “If you do not know the answer, state that you could not find the information rather than guessing.”

5. Step 5: Final Human Verification - The Most Critical Step
   

   You must personally check every sentence, every citation, and every legal argument generated by the AI against its original source. To skip this step is to abdicate your professional duty.

---

Advanced Strategies & Firm-Level Policy

Beyond the daily workflow, firms should establish a policy framework to ensure stability and trust in their use of AI.

• Establish a Multi-Layered Defense Framework: Consider a formal defense-in-depth approach: (Base Layer) Sophisticated prompts → (Structural Layer) RAG for grounding → (Behavioral Layer) Fine-tuning for specialization. Fine-tuning, using tools like ChatGPT's GPTs or Gemini for Enterprise, can train an AI on your firm's past work to enhance accuracy for specific tasks, but requires careful consideration of cost, overfitting, and confidentiality risks.
• Implement a Confidence-Based Escalation System: Design an internal system that scores the AI's confidence in its responses. If a score falls below a set threshold (e.g., 85%), the output could be automatically flagged for mandatory human review, creating a secondary safety net.
• Establish Principles for Billing and Client Notification: AI subscription fees should be treated as overhead, not directly billed to clients. Bill for the professional value created by using AI (e.g., deeper analysis, better strategy), not for the “machine’s time.” Include a general disclosure clause in engagement letters stating that the firm may use secure AI tools to improve efficiency, thereby ensuring transparency with clients.

---

Conclusion: Final Accountability and the Path Forward

The core of the AI hallucination problem ultimately lies in the professional’s verification mindset. The technologies and workflows discussed today are merely tools. As courts and bar associations have repeatedly warned, the final responsibility rests with the human professional.

“AI is a tool; accountability remains human.”

Only by establishing this principle and combining multi-layered verification strategies with a commitment to direct validation can we use AI safely and effectively. When we invest the time saved by AI into deeper legal analysis and more creative strategy, we evolve into true legal experts of the AI era. AI will not replace you, but the responsibility for documents bearing your name rests solely with you.

Frequently Asked Questions

Q: Can I trust the content if the AI provides a source link?

A: Absolutely not. A source link provided by an AI is merely a claim of where it got the information, not a guarantee of accuracy. The AI can misinterpret or distort the source's content. You must click the link, read the original text, and verify that it has been cited correctly and in context.

Q: What is the safest way to use AI with confidential client information?

A: The default should be to use an enterprise-grade, secure AI service contracted by your firm or a private, on-premise LLM. If you must use a public AI, you are required to completely anonymize all identifying information from your queries. Uploading sensitive data to a public AI service is a serious ethical and security violation.

Q: What is the most common mistake legal professionals make when using AI?

A: Skipping Step 5 of the workflow: “Final Human Verification.” Seeing a well-written, plausible-sounding sentence and copy-pasting it without checking the original source is the easiest way to fall into the hallucination trap, with potentially severe consequences.