Yes, it is Dr Google

Self-represented litigants are now appearing with submissions drafted by large language models. Medicine faced a comparable change two decades ago, when patients began arriving with the results of internet searches, a phenomenon its practitioners named "Dr Google". The Australian legal literature and the courts' recent guidance on generative AI do not, so far as I have found, draw on that experience. This article does. It treats the comparison as a hypothesis and tests it against the medical evidence.

The internet-informed patient of the early 2000s presented the medical profession with a familiar set of problems: confident but partial information, longer consultations, and a redistribution of information and authority that practitioners did not welcome. The evidence accumulated over the following two decades suggests that the feared harms were smaller than anticipated and that the public emerged more capable. The same dynamics are now observable in Australian courts and tribunals. The professional frustration is real and, on the medical evidence, likely to persist. The law is nonetheless better placed than medicine to absorb the change, for a structural reason that the comparison makes visible.

Why people use it

People come to court without a lawyer "sometimes out of choice but more often out of circumstance".¹ The circumstance is predominantly cost. In the Federal Court approximately 15 per cent of matters on judges' dockets now involve a litigant in person. Allocations of such matters rose from 247 in 2024 to 405 in 2025, an increase of 64 per cent, and in the Court's employment work by 200 per cent. Commencing a matter costs an individual $1,530 before any legal fee is incurred. A Chief Justice and a law society president, within weeks of each other, identified the same barrier: the impact of cost and delay "must not be underestimated and must be continuously addressed",³ and community legal services are "often turning away people in need".⁴

The Fair Work Commission has now examined the question empirically. Its user research, published in August 2026, comprised interviews with eighteen unfair dismissal and general protections applicants who had used generative AI, together with surveys of 408 applicants and 211 respondents.² Cost was the most consistently cited reason for use. Quotations of $7,500 to $11,000 for representation were reported as common, in matters that typically settle for between $3,500 and $10,000. The survey data point in the same direction. Among applicants who had not used generative AI, use of a lawyer or paid agent was higher, which the authors read as evidence that those who could obtain professional advice did not require a substitute. The interview material records the economics as applicants experienced them:

Where the quoted cost of representation exceeds the expected value of the claim, representation is in practice unavailable to the applicant. The Commission's own concern, set out in the same report, lies in the systemic consequences. Intake and case management processes were designed for an environment in which filing carried cost and effort, marginal claims tended to filter themselves out before lodgment, and the quality of a document signalled the care and capability of its author. The report finds that generative AI has weakened each of these implicit controls at once. The barriers to filing have fallen, self-filtering has diminished, and the quality signal has degraded, because machine-drafted submissions carry the structural markers of competent preparation whether or not the underlying claim has merit.² This is, in substance, the complaint that medicine made twenty years earlier.

What Dr Google actually did

The medical literature is now extensive enough to separate what internet-informed patients were accused of from what they did. The anticipated harm, injury from self-diagnosis and delayed care, was investigated and largely not found. A mixed-methods systematic review of 25 studies published between 2007 and 2018 reported no substantial evidence of delayed care or direct patient harm.⁵ What the studies did record was a change in how patients prepared. Search was used before the consultation rather than instead of it. In a large United States survey, 41 per cent of those who had attempted a self-diagnosis reported that a medical professional later confirmed it.⁶ In a primary-care study, 56.8 per cent of patients reported greater confidence in their doctor after having searched first.⁷ The systematic review describes the consultation as having moved from an authoritative toward a collaborative model.⁵

Professional frustration, and its persistence

The profession did not experience this as an improvement, and the evidence of its frustration is as robust as the evidence of patient benefit. Sixty-four per cent of doctors surveyed reported that patient internet use lengthened consultations, and 44 per cent reported an increase in unnecessary investigations.⁸ In focus groups, physicians described the internet-informed consultation as "annoying", "irritating" and "frustrating", and characterised correction of patients' information as "damage control". One participant stated the difficulty precisely: "I do not mind patients coming in with information, but it is very hard if they present you with 60 sheets. Time is really at a premium".⁹

The judicial equivalent is already on the record. The President of the Queensland Court of Appeal has observed that some self-represented litigants "find spurious arguments on the internet" that "take up much more of a court's time than the matter warrants", and that "there are no easy solutions".¹⁰ The observations of a general practitioner in 2006 and an appellate judge in 2026 are near-identical in content. The underlying concern, professional time consumed by confident but unreliable material, is the same.

The frustration also appears to be durable. A systematic review of 22 studies published between 1990 and 2023 did not find physicians' views of internet-informed patients trending more positive over that period. It found instead attitudes that remained divided, with individual practitioners moving between participative and defensive postures depending on the patient in front of them.¹¹ The profession did not, over three decades, reconcile itself to the change. The relevant finding is not that experts learn to accept the phenomenon but that public capability increased while professional frustration persisted alongside it. On this evidence, friction at the point of contact is a standing cost of a more informed public rather than a transitional one.

What did shift was the distribution of information and authority. The same body of research records physicians conceding that medical authority was "no longer as sacred as it used to be", an observation the profession made about itself.⁵ The legal profession now confronts the same redistribution of information and authority.

Why the law is better placed

Once accepted, the comparison discloses a structural difference that favours the law. The defect in Dr Google was specific. A search result is an assertion delivered to a person who has no independent means of verifying it against an authoritative source, through a ranking system optimised for prominence rather than accuracy. Google acknowledged the limitation in 2013, when its search quality guidelines introduced a separate class of "Your Money or Your Life" content, health among it, to be judged on expertise, authoritativeness and trust rather than on relevance alone.¹² Reputation was substituted for verification because the ranking could not establish truth on its own. The consequence was a persistent ceiling on accuracy. Online symptom checkers returned the correct diagnosis first in approximately 34 per cent of cases in 2015 and 36 per cent in 2020.¹³ The patient could seldom reach the source to close the gap, because medicine's ground truth lies within the body and, for the most part, behind journal paywalls.

The law differs in one respect that bears directly on the risk. Its primary sources are public and free. The statute governing a dispute and the judgments deciding comparable cases have been published in full, free of charge, on AustLII since 1995 and on Jade, which adds citation analysis showing how each decision has since been treated.²³ An AI-generated statement about Australian law can therefore be tested against the law itself by the person who received it. The verifiable layer that medicine had to construct, and largely could not place in the patient's hands, already exists in the law and costs nothing to consult.

This bears on a common objection. It is argued that because an AI answer appears checkable, it manufactures a spurious authority, and that legal AI is for that reason more dangerous than Dr Google rather than less.¹⁴ The objection would be sound if the sources were closed. Because they are open, the appropriate response is not to withhold the tool but to teach verification. The Commission's research identifies several of its eighteen interviewees already doing so: supplying the tool with their actual documents and the relevant legislation, putting the same question to more than one tool to test consistency, instructing the tool to argue the opposing case, and cross-checking outputs against other sources. About a quarter of surveyed users reported never relying on the tool alone.²

Risk communication research supports this approach. An accurate hazard warning does not fail because it discloses the hazard. In an experimental study of jellyfish safety messaging, candid warnings did not reduce visitors' intentions to travel to a beach destination overall, although they changed the behaviour of respondents with high risk perception.¹⁵ Candid guidance about the limits of generative AI is therefore unlikely to deter its use, and need not. Its function is to shape how the tool is used by people who will use it regardless. The converse risk is also documented. A protective measure that is misunderstood can itself create the hazard. An enclosure that bathers believe excludes Irukandji, when it does not, produces a false perception of safety on which a negligence claim may be founded.¹⁶ A fluent, well-structured document generated without verification carries the same false signal. The Commission's finding that document quality no longer indicates preparation is that hazard in a procedural setting.

What the comparison cannot carry

The analogy fails at one point, the reversibility of error. Legal error can fail faster, and less reversibly, than medical error. A mistaken self-diagnosis is often corrected at the next appointment. A missed limitation period or a defectively pleaded defence may not be recoverable at all. Fabricated authority is not hypothetical, and it is not confined to the unrepresented. In Dayal, a practising solicitor tendered a list and summary of authorities generated with AI assistance. The cited decisions did not exist. The court referred the practitioner to the Victorian Legal Services Board and Commissioner, and the Board subsequently varied his practising certificate.¹⁷ A trained lawyer, under a professional duty to the court, did not verify the output. The users least able to scrutinise an answer are frequently those most dependent on it, and the resulting automation bias is documented in the literature.¹⁸ The failure in Dayal, however, was one of practice rather than capacity. These cases show that verification matters more in law than in medicine, and that teaching it is the substantive task, for the profession as much as for the public.

That task is educational, and it is not new to legal education. Law schools have been criticised for producing technicians rather than reflexive professionals,¹⁹ and the case for cultivating reflexive capacity, the habit of interrogating one's own sources and reasoning, has been made as a public duty of legal education rather than a pedagogical preference.²⁰ The self-represented litigant who checks a generated citation against the reported judgment is exercising that capacity. The question for courts, tribunals and educators is how to make the practice ordinary among people who will never attend a law school.

Where the courts have already led

Australian courts and tribunals have regulated the tool rather than prohibited it, and have done so by formal instrument within a short period: the Federal Court's practice note in April 2026, the family courts' practice direction in May, and the Fair Work Commission's guidance note, published in August and in effect from 20 October, among others.²¹ The consistent instruction is not that the tool must not be used but that the user is responsible for its output and must verify it. The Federal Court's statement is the most explicit. Its Chief Justice has said not only that self-represented litigants' use of AI is "now inevitable", but that it "has the capacity to help people produce higher quality documents, better summaries of facts, and more clearly expressed submissions", and that "the key is in how it is used, not the fact that it is used".¹

The administration of justice remains best served by competent representation, and nothing in this analysis suggests otherwise. The right of a person to appear on their own behalf is, in the High Court's description, "fundamental".²² A tool that assists a person to exercise that right, verified against the public and freely available law, enlarges the number of people able to take part in the legal system. As the Supreme Court of Western Australia has observed, adherence to the rule of law "provides the bedrock of stability for our society as a whole", and can "provide a model for the culture at large".³