[Direct answer]They cannot be equated or substituted: machine-generated responses are merely an information output, without a fiduciary relationship or professional responsibility for the conclusions. Formal legal opinions must be issued by trustees after verifying the facts and legal provisions item by item. Specifically, machine responses can be used for understanding general directions and organizing issue lists; however, once they are submitted to courts, counterparties, or investors, or used to decide whether to sue or how to present evidence, they cannot be relied upon solely. The differences lie in three aspects: whether the facts have been verified, whether the legal provisions are the current effective versions, and who is responsible for any deviations. The rules on entrustment contracts and trustees' obligations in the Civil Code of the People's Republic of China (effective January 1, 2021) only bind the parties who establish a trust relationship, and machine services do not automatically enter into this relationship. In conclusion, treat it as a clue, not a legal opinion.

1. Problem background and typical scenarios

After encountering disputes, many litigants' first reaction is not to seek legal help, but to type out the situation in the chat window first. Within seconds, they receive a well-organized response: which law applies, the statute of limitations period, what materials are needed, and an estimated claim amount. This response reads like a legal opinion, so some people directly use it to draft lawsuits, calculate the statute of limitations, and negotiate with the other party until they encounter discrepancies at some stage and then check the information. There are several common triggering factors. First, calculating the statute of limitations: the machine provides a general rule of three years, but fails to notice that the contract in hand is executed in stages or involves debt collection and account reconciliation during the period, which actually has a different starting point. Following the general answer until the deadline approaches will narrow the window for relief. Second, the version of legal provisions: the response cites a revised article or mixes departmental regulations with local interpretations, which are immediately pointed out by the other party during negotiations, greatly undermining the credibility. Third, organizing evidence: the machine lists a comprehensive set of materials but fails to prioritize the disputed issues in the case, resulting in the two or three key documents that determine the outcome being buried in the long list. Fourth, the form of legal documents: the automatically generated content lacks signatures, stamps, or letterheads, and is not accepted as official materials when submitted to the court or third parties. There are also scenarios that occur afterward: litigants send letters, provide responses, or sign supplementary agreements based on the machine's suggestions, only to later find a statement in the documents that is unfavorable to them. They want to explain that they followed the system's recommendations, but cannot produce the generation records or prompts. They cannot even recall what they asked or what the system responded to. Such situations of first following the machine's advice and then retroactively gathering evidence often come to light when the evidentiary deadline approaches.

II. Key points of dispute

The disagreements focus on three points, each involving different judgments. The first point concerns the nature of the output. Machine responses are generated based on the questions, which involve language-level inference without accessing the original documents, verifying facts, or taking responsibility for the conclusions. Legal opinions, however, must be based on the trustee reviewing the materials, identifying the disputed issues, searching for current effective regulations, and issuing opinions with their signature. Although both may appear to be written texts, their legal weight differs: the former is a clue, while the latter is a professional outcome. Treating clues as outcomes often leads to disputes. The second point is about who should bear the responsibility for errors. Machines may mistake invalid provisions as current law, apply local standards to other regions, or invent non-existent clause numbers. When parties act based on these recommendations and suffer losses, they later find that service agreements typically state that the advice is for reference only and does not constitute legal opinions. In contrast, trustees issuing opinions must bear corresponding liability for errors caused by gross negligence. The provisions of the Civil Code of the People's Republic of China (implemented on January 1, 2021) regarding agency contracts and the trustee's fault liability serve as the basis for this distinction; without a trustee relationship, there can be no such liability. The third point concerns whether the output can be used as evidence. Defenses, legal briefs, and evidence lists submitted to the court must be signed, correspond to specific facts, and be submitted within the evidentiary period. A text without sources, signatures, or verifiable generation conditions is typically only used as a draft for the parties to organize their thoughts. The provisions of the Civil Procedure Law of the People's Republic of China (amended on December 24, 2021 and implemented on January 1, 2022) regarding litigation documents and evidentiary deadlines determine the form and timing of formal materials entering the legal process. Taken together, the issue is not whether the machine's responses are correct, but whether this output qualifies to enter the legal relationship.

III. Professional analysis by Hengluo Law Firm

When facing a machine-generated response, it's more efficient to first check whether it addresses general rules or the specific case's conclusions, rather than reading it sentence by sentence. Even if the general rules are sound, there are still three crucial steps to verify their applicability to the actual case: where the facts come from, whether the norms are the latest version, and whether the conclusions align with procedural milestones. **First, verify the facts**. The machine only knows what's stated in the question, but lacks access to the original contract, correspondence, payment receipts, or chat records. If the question omits mentioning that the contract was supplementarily signed once, the entire reasoning could be invalidated. Manually checking every factual statement in the response against the original documents is the most time-consuming and indispensable step in human review. **Second, verify the norms**. Retrieve the current version of the legal provisions mentioned in the response (e.g., article titles, clauses, effective dates) and ensure they haven't been amended, replaced, or misinterpreted as judicial interpretations. Particular attention must be paid to statutes of limitations, jurisdiction, and burden of proof, as even minor version discrepancies could overturn entire conclusions. This also explains why logical coherence in the response doesn't guarantee its practical reliability. **Third, verify the procedural milestones**. General rules only become meaningful when applied to the specific case's timeline and procedural stages: when does the statute of limitations begin, which evidence proves interruption, how many days remain before the evidentiary deadline, and when should written opinions be submitted. The machine provides general rules, while human reviewers adapt them to the specific case. There's often a disconnect between the user's and reviewer's perspectives: the former emphasizes the response's logical clarity and consistency with online sources, while the latter prioritizes original evidence, current norms, and acceptable document formats. Only contracts, payment receipts, acknowledgments, and current legal provisions are admissible evidence, not screenshots of dialogues. Evidence types are also divided into two categories: those proving the machine's output (e.g., generated records, prompts, timestamps, and screenshots) and those proving the actual case's facts (e.g., contracts, performance records, correspondence, and ownership documents). For extended discussions on the interface between machine output and human review, please refer to the following link: [Link].Special services and other case pagesSearch for it.

IV. Practical suggestions and legal basis

把机器答复放进工作流程，宜先定三条边界，越往后越不能省人工。第一条边界是用途分级：仅用于了解术语、梳理问题清单、准备咨询提纲的，可以直接用；凡是要写进起诉状、答辩意见、谈判函、补充协议，或据此决定是否起诉、如何报价的，一律先经人工复核再落笔。分级写成一张两列表，左列列用途，右列写是否需要复核，团队里谁都照表执行，比每次临时判断稳妥。第二条边界是留痕方式：每次生成都保留提示词、完整回复、生成时间与模型版本，截图之外最好另有可导出的记录；日后若需说明当时依据什么判断，这些是唯一能还原现场的材料，举证期限内能否提交、以何种形式提交，宜在复核时一并确认。第三条边界是复核清单：至少核四项，事实有无原件支撑、法条是否现行有效、时效起算与中断节点是否对、文书签署与提交形式是否合规。受托人的核验义务由委托合同与执业规范确定；若争议进入诉讼，向法院交纳的案件受理费按《诉讼费用交纳办法》（2007年4月1日施行）分段累计，与复核服务费分列。三条边界定完，机器是提速工具，人工仍是责任关口。

5. Henglu Law Firm's Advice

There are several habits that, the earlier you establish them, the more convenient they will be. Don't let the machine decide the time for you: for such key points as the start of the statute of limitations, the deadline for providing evidence, and the deadline for responding, always verify them with the original documents and current provisions. The numbers provided by the system should only be regarded as a reminder. Don't treat a long list as an evidence index: materials should be selected based on the focus of the dispute. What really matters are often just a few documents. Completeness doesn't always mean being relevant. Leave a record on the same day: generated records and reminder words may not be retrievable after a while. Save them immediately after asking questions, which is more reliable than retroactively adding them. There's one point that's easily overlooked: for every document sent to others, the signature and name at the bottom determine who is responsible for it. No matter how smoothly the machine writes it, it can't replace this column. Write these points into a checklist and paste it into the process, which is more reliable than relying on memory every time.

VI. Frequently Asked Questions

Q1: Can the thesis defense outlines automatically generated by large models be used as legal opinions stamped with the law firm's official seal?

No. A legal opinion letter stamped with the law firm's official seal is a professional document issued by a trustee after reviewing the materials and signing it. Machine-generated defense points lack a trustee relationship, fact verification, and a signature of responsibility, so they do not formally belong to this category of documents. There are two levels of judgment: if it's used for personal reflection, it has value; but if it's submitted to the court, the other party, or investors, it must be manually verified and reissued. Even if the content happens to be correct, the lack of a signature, letterhead, list of supporting materials, and issuing entity makes it difficult for the recipient to accept it as an official opinion. The provisions of the Civil Code of the People's Republic of China (effective January 1, 2021) on entrustment contracts only bind the two parties who establish a trustee relationship. Machine services do not automatically enter this relationship. The conclusion is straightforward: use it as a drafting reference, but don't directly submit it as an official opinion.

Q2: To preserve the machine's responses as evidence for rights protection, which generated records and prompt words need to be fixed and retained?

There are two types of records that need to be preserved, and missing either of them will make it impossible to reconstruct the scene. The first type is the generation-side records: complete prompts, round-by-round responses, generation time and model version, exportable dialogue files, and screenshots taken at the time. These records show what the machine actually said and under what conditions it said it. The second type is the fact-side originals: contracts, performance certificates, correspondence records, and ownership documents, which are used to determine whether the factual premises in the responses hold true and also serve as the basis for asserting rights in the future. These two types of records should be stored separately, and screenshots should not be mixed with contracts in the same folder. A common omission in practice is only saving the conclusion screen but not the original question and subsequent follow-up questions, making it impossible to clarify the specific scenario being asked about at the time. The provisions of the Civil Procedure Law of the People's Republic of China (amended on December 24, 2021 and implemented on January 1, 2022) regarding the form of evidence and the deadline for providing evidence determine the format and submission timeline of these records. It is more reliable to first create a comparison table by subject, time, and document type, and then fill in the gaps item by item rather than randomly searching through chat records later.

Q3: Could the reviewers please check the AI draft article line by line? Is the fee calculated per thousand characters or per document?

Both options are common, and the choice depends on whether the verification of the boundaries can be completed in one go. Charging by the thousand words is suitable for situations where the initial draft has a clear scope and only requires error correction and legal clause verification; while charging by the number of documents is suitable for cases that require comprehensive services including fact verification, material selection, and issuing review notes. Regardless of the pricing method, it is advisable to clarify before commissioning whether it includes checking for invalid legal clauses, comparing the draft with the original sentence by sentence, or providing subsequent rounds of revision. Here, it is important to clarify the following points: the fees for using machine tools, the service fees for manual review, and the case acceptance fees paid to the court after entering litigation are three separate expenses. The first two are agreed upon by the commissioning parties, while the latter is accumulated in stages according to the "Measures for the Payment of Litigation Fees" (implemented on April 1, 2007) and should not be combined into a total amount for price comparison. To compare the two pricing methods, it is necessary to list the number of words in the initial draft, the number of legal clauses, and the verification rounds before discussing the price.

Q4: If the machine misquotes the legal provisions, when should the manual correction be completed before the expiration of the evidence-presentation period?

The revision must be completed before the deadline for providing evidence, as the closer the trial date is, the more passive you will be. Once the invalidated legal provisions are included in the submitted written opinions, the amendment is not just about changing a few words. It often requires re-checking the entire reasoning, submitting supplementary explanations, or even applying for correction of the transcript, all of which take up procedural time. It is advisable to work backwards in three steps: First, check the legal provisions and implementation dates mentioned in the responses against the current legal texts one by one, and mark those that have been invalidated or do not match the current version; Second, recalculate the starting and interrupting points of the affected statute of limitations to confirm whether the claims are still within the time window; Finally, resubmit the revised opinions in the form of a legal document. The general civil first-instance procedure usually lasts six months, the summary procedure three months, and the second-instance procedure generally lasts three months. The announcement or appraisal may take longer, and the trial time limit is governed by the Civil Procedure Law of the People's Republic of China (amended on December 24, 2021 and implemented on January 1, 2022) and judicial interpretations. Separately record the deadline for revision in your schedule, and don't wait until the other party points it out before taking action.

Q5: Today, let's first take a screenshot of the three-year statute of limitations that AI calculated incorrectly, and then check which original contract to confirm the starting date of the statute of limitations?

Today, first take a screenshot, then verify the calculation date. Don't reverse the order. I found that the three-year statute of limitations was calculated as two years. The first thing to do is to fix the original question, the complete reply, and the generation time. Save a copy of the screenshot and another copy that can be exported for records. This step can be completed on the same day. If you wait too long, the records may not be exportable. After fixing them, go back to the rules of the Statute of Limitations and Calculation in the Civil Code of the People's Republic of China (implemented on January 1, 2021) and check the performance period, installment arrangements, collection and reconciliation records in the original contract to confirm the exact calculation date and whether there were any interruptions in the middle. First, import the generation records, then adjust the contract and transactional vouchers, and finally write the system's conclusions and the results of manual recalculation side by side in a table. After completing these three steps, the corrections will be traceable. If you do it in reverse, you won't even be able to explain where the mistake occurred in the original version.

The above Q&A is mainly based on the judicial practice in the Beijing area. There may be differences in the adjudication standards in different regions. It is recommended to consult a professional lawyer based on the specific circumstances of the case.

If you are currently facing similar problems, it is recommended that you consult a professional lawyer as soon as possible with the relevant evidence materials, in order to avoid missing the statutory time limit.Specialised services Legal services·Beijing Henglu Law Firm

Disclaimer: This article is compiled based on current laws and regulations and publicly available adjudication rules, and is for general reference only. It does not constitute a formal legal opinion on specific cases. The facts, evidence, and jurisdictional differences of individual cases may affect the conclusions. In the event of a dispute, please consult a practicing lawyer promptly.