When AI Legal Advice Costs More Than a Lawyer

Asking artificial intelligence to prepare a legal case may appear to be an inexpensive alternative to engaging a lawyer.

A recent Court decision demonstrates how quickly that apparent saving can become a false economy.

A self-represented litigant used ChatGPT to prepare legal arguments challenging the refusal of his student visa. The resulting submissions looked sophisticated. They referred to legislation, identified legal principles and cited numerous decided cases.

There was just one problem: when the applicant appeared in Court, he could not explain what any of it meant.

His case was dismissed. He was ordered to pay more than $9,000 towards the other party’s legal costs, and the Court considered whether he should pay even more because of the additional work caused by his AI-generated submissions.

What happened when a litigant used ChatGPT to prepare his case?

The applicant asked the Court to review a decision affirming the refusal of his student visa.

Judicial review is a particularly technical area of law. It is not enough to argue that a decision was unfair, unreasonable or simply wrong. An applicant must identify a recognised legal error affecting the validity of the decision.

The applicant had been given an opportunity to file written submissions explaining his case. He did not comply with the timetable ordered by the Court. Instead, six days before the hearing, he provided lengthy submissions prepared with the assistance of ChatGPT.

The Court allowed him to rely upon them.

On paper, the submissions appeared to advance detailed legal arguments supported by numerous authorities. In the courtroom, however, the appearance of legal sophistication quickly unravelled.

When asked to explain the alleged legal errors, the applicant could not make a coherent argument. When asked about one of the cases cited in his submissions, he was unable to explain it. The Court concluded that he had no understanding of what was contained in the documents filed in his name.

ChatGPT had given him the words of a legal argument. It had not given him the legal knowledge required to understand or defend it.

Did ChatGPT get the law wrong?

The problems went well beyond a minor error or an incorrectly typed citation.

Some of the cases relied upon did not support the legal propositions attributed to them. One citation referred to a paragraph that did not exist. Another authority was used for a proposition it did not establish.

ChatGPT also appears to have confused two different cases with the same party name. The Court observed that it had probably referred to a 2026 decision when it intended to refer to—or distinguish—a different decision from 2019.

One of the applicant’s grounds of review was described by the Court as confusing and making no sense. The Court said that this ground perhaps best demonstrated the danger of relying upon generative AI to draft grounds of judicial review and legal submissions.

This is an important development from the more familiar examples of AI inventing cases that do not exist.

A fictitious case can be exposed by searching for it. A genuine case used incorrectly can be much harder to detect. The case exists, the citation looks convincing and the language sounds authoritative. Unless someone reads and understands the decision, the error may pass unnoticed.

We have previously discussed this broader problem in our article Can AI Replace Your Lawyer?.

The danger is not confined to AI inventing the law. AI can also misunderstand genuine law, take statements out of context or construct an argument that sounds plausible but has no proper legal foundation.

Did using AI cost the applicant his case?

It is impossible to know whether the applicant would have achieved a different result if he had engaged a lawyer.

The underlying case appears to have faced serious difficulties. The applicant had lived in Australia for many years, had completed a master’s degree and subsequently enrolled in courses he did not complete. The courses for which he later sought a visa were at a lower level than qualifications he already held.

The Tribunal had concluded that he was using the student visa system to maintain his residence in Australia rather than genuinely intending to remain temporarily for study. The Court found that conclusion was reasonably open on the evidence.

A lawyer could not have changed those facts.

However, that does not mean legal advice would have made no difference.

An experienced lawyer may have advised the applicant that the proposed challenge had poor prospects and that commencing or continuing the proceeding could expose him to a substantial costs order. That advice might have saved him from pursuing an unsuccessful case.

Alternatively, if there were any properly arguable grounds of review, a lawyer could have identified them, removed the arguments that had no legal foundation and presented the case in a form the Court could properly consider.

Instead, the applicant arrived at Court with lengthy submissions that he did not understand, authorities that did not support his arguments and at least one ground that made no legal sense.

AI may not have created the weaknesses in his underlying case, but it disguised those weaknesses beneath pages of legal language. It appears to have given him confidence that he had arguments which could not withstand examination in Court.

Can using AI increase what you have to pay the other side?

Using AI can increase a litigant’s exposure to costs when it creates unnecessary work for the Court or another party.

The applicant was ordered to pay the Minister’s costs fixed at $9,097.93 after his application was dismissed.

The Court did not say that the entire amount was caused by ChatGPT. It was a costs order arising from the unsuccessful proceeding generally.

However, the Court found that the applicant’s late AI-generated submissions required the Minister to undertake additional work and prepare supplementary submissions. The Court was therefore prepared to consider whether the applicant should pay costs above the ordinary scale.

What appeared to be a free or inexpensive substitute for legal representation left the applicant facing more than $9,000 in costs, together with the possibility of a further costs penalty arising directly from his use of AI.

That is how free legal advice can become very expensive.

Why is legal-looking language so dangerous?

The greatest risk is that AI can make a weak argument appear much stronger than it really is.

Legal documents have a recognisable style. They refer to legislation, use technical expressions and cite decisions from superior courts. Generative AI is very good at reproducing that style.

But legal language is not the same as legal reasoning.

A court application must identify the correct legal question, apply the relevant principles to the actual decision and explain precisely how any error affected the outcome. Adding more cases or more legal terminology cannot compensate for the absence of a valid argument.

The person whose name appears on a court document must also be able to explain and defend it. A litigant cannot assume that the Court will accept an argument merely because ChatGPT expressed it confidently.

In this case, the Court tested the submissions by asking the applicant to explain them. He could not.

The polished language did not strengthen his case. It exposed the fact that he did not understand the case he was asking the Court to decide.

Should I use AI to prepare court documents?

You should not file an AI-generated court document unless its contents have been independently checked by someone who understands the relevant law.

AI can assist with organising information, explaining unfamiliar terminology and preparing questions to discuss with a lawyer. It may also help a person create a chronology or identify documents that could be relevant.

That is very different from relying upon it to decide:

• whether you have a legally sustainable case;
• which court or tribunal has jurisdiction;
• what orders you should seek;
• which legal grounds are available;
• whether the evidence supports those grounds; or
• what financial risks you face if the case is unsuccessful.

Those decisions require more than the ability to generate convincing words. They require legal knowledge, strategic judgment and an objective assessment of the strengths and weaknesses of the case.

When should I speak to a lawyer?

You should obtain legal advice before commencing proceedings or filing important court documents, particularly if you have used AI to prepare them.

Seeking advice does not necessarily mean engaging a lawyer to conduct the entire case. A preliminary consultation may help determine whether there is a viable claim, what evidence is required and whether the likely benefit justifies the cost and risk of proceeding.

Sometimes the most valuable advice a lawyer can provide is that a case should not be pursued.

That advice may feel disappointing, particularly when ChatGPT has produced pages of apparently persuasive legal arguments. It may also save tens of thousands of dollars, months of work and the risk of paying the other party’s costs.

Artificial intelligence can generate an argument in seconds. It cannot guarantee that the argument is correct, that the cases say what it claims or that the person relying upon it has a case worth pursuing.

The applicant in this case tried to challenge a significant decision with submissions generated by ChatGPT. He was left unable to explain his own arguments, his application was dismissed and he was ordered to pay more than $9,000 in costs.

The legal advice that appears most expensive at the beginning of a case may prove considerably cheaper than discovering the problems at the end.

Asif v Minister for Immigration and Citizenship [2026] FedCFamC2G 1402