When the Judge Did Not Write the Judgment: AI Generated Judicial Orders and the Right to a Reasoned Decision in Indian Courts
There is a quiet crisis developing in Indian courtrooms, and most litigants do not know it is happening. Judgments are being delivered that run into dozens of pages, cite multiple precedents, reproduce cross-examination at length, and arrive at findings yet on careful reading, they reveal a peculiar quality: the evidence was never actually analysed. The facts were summarised. The law was stated in general terms. The conclusion was reached. But the connective tissue of judicial reasoning the independent, fact-specific application of law to evidence is absent.
This is increasingly recognised as a signature of judgments drafted with the assistance of artificial intelligence tools. As AI writing models become accessible to anyone with an internet connection, their use inside the judiciary whether by officers themselves or their staff is no longer a hypothetical concern. It is a present reality that practising advocates across India are beginning to encounter in impugned orders before appellate courts. The matter is not merely academic. It goes to the heart of what a litigant is entitled to: a decision made by a thinking, reading, reasoning human mind that has genuinely grappled with their case.
This article examines the legal framework governing the duty to give reasoned judgments, identifies the specific markers that distinguish AI-assisted drafting from genuine judicial reasoning, explains the appellate remedies available to a litigant who has received such an order, and offers a framework for raising these submissions effectively before higher courts.
Legal Background
The Constitutional Foundation of Reasoned Judgments
The duty of a court to give reasons for its decision is not merely a procedural formality. It is a constitutional imperative rooted in the principles of natural justice and the right to a fair hearing guaranteed under Article 21 of the Constitution of India. The Supreme Court has, across decades of jurisprudence, consistently held that a reasoned order is a precondition of judicial legitimacy. Reasons are the mechanism by which a court demonstrates that it has heard, considered, and decided and by which an appellate court can meaningfully review what the trial court did.
Under the Code of Civil Procedure, 1908, Order XX Rule 5 specifically mandates that a judgment must contain the points for determination, the decision thereon, and the reasons for the decision. This provision is not satisfied by the mere recitation of the parties’ respective cases followed by a conclusion. The “reasons” referred to in Order XX Rule 5 must reflect the court’s independent engagement with the evidence and the applicable law, applied to the specific facts before it. A judgment that reproduces a party’s pleading, states the law in general terms, and then announces a conclusion without the intervening step of applying the law to the facts does not satisfy this requirement.
In criminal proceedings, Section 354 of the Code of Criminal Procedure, 1973 (now Section 392 of the Bharatiya Nagarik Suraksha Sanhita, 2023) imposes an identical obligation, requiring that every judgment contain the points for determination, the decision, and the reasons. The principle is universal across jurisdictions: a decision without reasons is not a judicial act in the full sense of the term.
What Amounts to a “Reasoned” Judgment
The distinction between a judgment that contains reasons and one that merely appears to contain reasons is fundamental. True judicial reasoning involves four interconnected steps: identifying the precise factual dispute that requires resolution; assessing the quality and credibility of the evidence bearing on that dispute; applying the applicable legal standard to those facts; and explaining why the evidence, so assessed, satisfies or does not satisfy that legal standard.
Each of these steps requires original engagement with the record. A court cannot perform the second step assessing the quality of evidence by reproducing the evidence verbatim and then skipping to a conclusion. Yet this is precisely the pattern that AI-assisted drafting produces. The model is given the pleadings and the recorded evidence, asked to generate a judgment, and produces a document that includes all the required structural elements facts, issues, evidence summary, findings, order but performs no independent analytical work between the evidence and the findings.
How to Identify an AI-Drafted Judgment
The Cross-Examination Problem
The most reliable diagnostic indicator of AI-assisted judicial drafting is the treatment of cross-examination. Cross-examination is the crucible of civil and criminal evidence in Indian courts. It is where the party’s narrative is tested, inconsistencies are exposed, admissions are extracted, and the truthfulness of the witness is challenged. A judgment that genuinely engages with the evidence will invariably grapple with the cross-examination citing specific admissions, explaining why certain answers damaged the witness’s credibility, and tracing how particular answers led the court to prefer one version of events over another.
An AIgenerated judgment, by contrast, will typically reproduce the cross-examination in full sometimes across many pages and then proceed to decide each issue as though the cross-examination had not occurred. The cross-examination is present in the document. It was not ignored in the sense of being withheld. But it was not processed. The findings on each issue make no reference to specific admissions, offer no explanation for why damaging answers were discounted, and do not trace any logical path from the cross-examination to the conclusion. The structural presence of the cross-examination provides an illusion of completeness while concealing a complete absence of reasoning.
This pattern is diagnostically significant because it reflects a technical limitation of large language models: they are highly capable of summarising and reproducing content they are given, but they do not “read” evidence the way a judicial mind does. When an AI model is given a case file and asked to write a judgment, it will include the evidence because the prompt or the structural template requires it, but it will derive its conclusions from the general legal framework rather than from the specific admissions and contradictions in that particular record.
Mechanical Issue-Framing and Formulaic Transitions
AI-drafted judgments tend to exhibit a rigid, template-like structure in their treatment of issues. Each issue is addressed in a separate section that opens with a formulaic sentence variants of “So far as Issue No. ___ is concerned, this Court deems it fit to mention here that…” and closes with an equally formulaic conclusion declaring the issue answered in the affirmative or negative. This consistency is superficially impressive but analytically empty. Human judges drafting organically vary their approach to different issues depending on the nature of the dispute raised by each one. Some issues are complex and require extended analysis; others are straightforward and deserve brief treatment. A judgment that treats every issue with the same template structure regardless of its difficulty is exhibiting an AI characteristic.
Similarly, the excessive use of certain phrases “in the considered opinion of this Court,” “this Court believes,” “in light of the foregoing discussion,” “for the sake of convenience,” “to avoid repetition of facts” in recurring combinations across a single judgment is a known marker. These are phrases that AI models trained on Indian legal text deploy as judicial voice filler. A human judge uses them occasionally; an AI model uses them at nearly every transition.
Block-Quotation of Precedents Without Application
Another reliable indicator is the verbatim reproduction of passages from cited judgments sometimes including quotes-within-quotes from even older cases embedded in the cited judgment followed by no genuine application to the facts at hand. AI models trained on legal databases are highly capable of locating and reproducing passages from cited cases with precision. What they consistently fail to do is then take that quoted principle and trace, sentence by sentence, how it applies to the specific facts of the case before the court. The quotation is present; the application is a single conclusory sentence: “In the present case, the same principle applies.”
A human judge rarely quotes at such length and, when doing so, invariably follows the quotation with a substantive paragraph explaining the parallel between the facts of the cited case and the facts before the court.
Misapplication of Legal Principles Across Factual Contexts
Perhaps the most jurisprudentially significant marker of AIgenerated judicial reasoning is the systematic over-generalisation of legal rules without factual adaptation. AI models learn legal propositions in their general form and apply them to any factually proximate situation, without performing the necessary contextual analysis that determines whether the rule in fact applies.
A concrete example illustrates this clearly. Section 17 of the Registration Act, 1908 requires compulsory registration of instruments that create, declare, assign, limit, or extinguish rights in immovable property. This is a well-settled proposition, and an AI model will correctly state it. However, the application of this rule requires the court to first determine whether the instrument in question actually operates upon immovable property or upon something else for instance, a contractual entitlement to money representing the consideration for the acquisition of immovable property. A settlement agreement among co-owners determining the percentage in which compensation money from land acquisition will be distributed operates upon a monetary entitlement, not upon immovable property itself. A court applying Section 17 to such a document without performing this preliminary analysis is committing a legal error of the kind that AI over-generalisation predictably produces.
What the Courts Have Said
The duty to give a reasoned judgment has been affirmed consistently across the judicial hierarchy. The Supreme Court in State of Maharashtra v. Vithal Rao Pritirao Chawan, (1981) 4 SCC 129, held that the giving of reasons is an essential element of judicial proceedings and that a judgment without reasons is a judgment that cannot be reviewed or appealed in any meaningful sense. The court emphasised that reasons serve a double purpose they discipline the judge’s own thinking and they inform the losing party of the basis on which they were defeated, enabling a considered decision on whether to pursue an appeal.
In Kranti Associates Pvt. Ltd. v. Masood Ahmed Khan, (2010) 9 SCC 496, the Supreme Court elaborated at length on the requirement of reasoned orders in quasi-judicial and judicial proceedings, observing that the absence of reasons strikes at the very root of fairness and is a ground for setting aside an order irrespective of the ultimate correctness of the conclusion reached. The court noted that where a statutory authority or a court is required to give reasons, the absence of those reasons is not a mere irregularity capable of being cured on appeal it is a fundamental defect.
On the related question of appreciation of evidence, the Supreme Court has consistently held that a first appeal court is entitled and indeed obliged to re-appreciate the entire evidence independently, and that a judgment of the trial court which has not genuinely engaged with the evidence does not carry the persuasive weight that a reasoned judgment would. The appellate court in such cases is effectively sitting as the first court to have actually analysed the record.
The legal framework, therefore, already provides the tools needed to address AIgenerated judicial orders. The principles are in place. What is new is the factual context in which they must be applied.
Rights and Remedies
First Appeal Under Section 96 CPC
Where the impugned order is a decree passed by a civil court, the primary remedy is a First Appeal under Section 96 of the Code of Civil Procedure, 1908. A First Appeal is a complete re-hearing on facts and law. The appellate court is entitled to re-appreciate the entire evidence independently, draw its own inferences, and arrive at findings different from those of the trial court.
This is critically important in the context of AI-drafted judgments: because the trial court has effectively not analysed the evidence, the First Appeal court is not reviewing a considered judicial finding it is stepping in to perform the analysis that should have been performed below. The appellant’s advocate should frame grounds specifically around the failure to analyse cross-examination admissions, the non-application of legal principles to specific facts, and the formulaic treatment of issues.
The grounds of appeal should be drafted with surgical precision: identify the specific admission in cross-examination that was ignored; state the finding on the relevant issue; and submit that had the Learned Judge actually applied the admission to the issue, the finding would necessarily have been different. This is not merely an abstract submission about judicial methodology it is a substantive ground going to the correctness of each finding.
Revision Under Article 227 of the Constitution
Where the order is not a decree but an interlocutory order, or where the error is of jurisdiction rather than merely of fact-appreciation, a revision petition under Article 227 of the Constitution before the High Court is available. The High Court’s superintendence jurisdiction under Article 227 extends to examining whether a subordinate court has exercised its jurisdiction legally and in accordance with law and a judgment that fails to independently apply the law to the facts, or that applies a legal rule without performing the necessary factual analysis, is a judgment that has not been made “in accordance with law” in the full sense.
Oral Submissions Before the Appellate Court
When the issue of AI-assisted drafting is to be raised in oral arguments, it is important to calibrate the submission carefully. The submission should be framed not as an accusation of judicial misconduct but as an explanation for a pattern of errors that is otherwise difficult to account for. An experienced judge who has not seen the phenomenon may be sceptical of a bare assertion that “the judgment was written by AI.” What is compelling is a specific, mapped analysis: “My Lord, the cross-examination of the plaintiff spans nine pages of the impugned judgment. In those nine pages, the plaintiff made the following three admissions which directly contradict her case on Issues 4, 5, and 6. Not one of those admissions is mentioned anywhere in the court’s discussion of those issues. We submit that this is not an oversight the evidence was reproduced but not processed, and this is consistent with the specific limitation of AI drafting tools which summarise content without analytically engaging with it.”
This submission serves a dual purpose: it explains the specific errors in the judgment and it invites the appellate court to exercise its independent fact-finding jurisdiction without deference to the trial court’s findings. In the context of a First Appeal where independent re-appreciation is the standard anyway, this framing is both legally sound and practically effective.
The Broader Question of Judicial Accountability
The use of AI tools by judicial officers in drafting judgments is not yet addressed by any formal guideline, circular, or statute in India. The Judicial Officers’ duties under the service rules of various High Courts require the independent exercise of judicial functions, and there is a strong argument that outsourcing the analytical core of a judgment to an AI system even while reviewing and approving the output falls short of this requirement where the AI output itself contains errors of fact-application and law that the officer failed to catch.
This is a question that is likely to come before the High Courts and the Supreme Court in the coming years. Advocates who identify and competently argue such cases before appellate forums will be contributing to the development of a legal framework that protects litigants from the specific category of injustice that arises when a court, nominally, decides a case, but in reality delegates the decision to a machine that does not read the record.
Conclusion
The right to a reasoned judgment is not a technicality. It is the substance of what it means to have your case heard by a court. When that reasoning is absent replaced by a structurally complete but analytically hollow document a litigant has received the form of justice without its content. They have been given a verdict, but they have not been given a decision.
The tools to challenge such judgments already exist in Indian law. First appeals permit full re-appreciation of evidence. The grounds for challenging non-reasoned orders are well-settled. What is required is the ability to identify when a judgment falls into this category and the skill to frame the challenge with the specificity and precision that persuades an appellate court.
The emergence of AI in judicial drafting is neither the end of the world nor a problem that will resolve itself. It is a new category of legal challenge that requires a disciplined, evidence-based response. For the practising advocate, the first skill to develop is the ability to read a judgment not only for what it says but for what it fails to do and to translate that failure into grounds that a court can act upon.
If your case has been decided by an order that runs to many pages but never engages with the specific evidence that was placed before the court, you have not necessarily lost your case. You may simply have received a first draft.
