A survey by the UK charity JUSTICE and the Administrative Fairness Lab asked people with a recent legal problem where they turned for help. A quarter of 18-24 year olds had consulted an AI chatbot rather than a lawyer. The issues driving them there were housing, employment and family matters, the areas where legal aid has been contracting for years across common-law jurisdictions. People who cannot afford a lawyer and cannot get one funded type their problem into a chatbot and take whatever comes back.
Young people dominate the data, and for predictable reasons. An 18-24 year old has grown up with search engines as the default starting point for any question, and has lived through the normalisation of AI assistants for everything from travel planning to mental health support. The leap to legal advice feels like a continuation. The cost pressures make it rational. Legal aid in Australia has not kept pace with demand for decades, and the categories where it is available have narrowed. A young worker with an unfair dismissal problem or a tenant facing eviction will look for help where they can find it, at a price they can afford. A free chatbot is always available and speaks with conviction.
The output from a generic chatbot is fluent and confident, with no source the user can verify. People who turn to chatbots for legal help are often looking for reassurance as much as answers. This is the register where chatbots sound most authoritative and are least reliable. A model trained to produce plausible, empathetic prose will generate it whether or not the underlying legal content is accurate. A tenant asking whether their eviction notice is valid needs a correct answer and reassurance. The chatbot provides the reassurance reliably. The correct answer is the part it cannot guarantee. When the two arrive together in the same confident tone, the user has no way of distinguishing the genuine from the fabricated, and the reassurance makes the inaccuracy more persuasive.
The survey is British. The access-to-justice gap it maps is one Australian practitioners will recognise. Pro bono services, community legal centres, and duty lawyer schemes here operate against the same structural backdrop: a shrinking legal aid net, a rise in self-represented litigants, and tribunals carrying caseloads that were never designed for unrepresented parties. The categories the survey identifies are the bread and butter of state tribunals like NCAT, VCAT and QCAT, and of the Fair Work Commission's jurisdiction. These are the forums where self-represented litigants are most common and where duty lawyer services are thinnest. A tenant arguing about a bond dispute at NCAT, or a casual worker testing whether their dismissal was unfair, is unlikely to have had legal representation. The survey data, read in an Australian context, would predict the same pattern here.
We are already seeing the downstream effect. People bring chatbot advice into proceedings. A tenant who has been told their eviction notice is invalid may arrive at NCAT prepared to argue the point, only to find that the notice period the chatbot cited does not match the actual legislation. Most of these matters never generate a published decision. The gap between the chatbot's confident answer and the real law surfaces for the first time at the tribunal counter, when a decision-maker applies the actual statute.
We think the practical implication for Australian practitioners doing pro bono or duty-lawyer work has been understated. Most clients are doing what any reasonable person would do when they cannot afford a lawyer and the alternative is nothing. The implication is operational. The person walking into your duty lawyer session has probably already had a conversation about their case with a chatbot. They may have formed views about their prospects and their rights based on what the chatbot told them. They may have been told something confidently that is wrong. An intake process that does not ask what AI tools the client has already consulted is working with an incomplete picture.
A practitioner who asks a client what they have already looked up online or which AI tool they consulted, and hears a summary of what the chatbot said, is in a better position than one who does not ask. Sometimes the chatbot's answer will be broadly correct. Often it will not. The practitioner can then verify it and move forward on ground they trust. The damage happens when the chatbot's confident but incorrect answer calcifies into the client's understanding of their case before a lawyer ever sees them. By the time they reach a duty lawyer or a pro bono clinic, that understanding may have shaped the documents they have gathered and the arguments they believe are available to them. Undoing a wrong but confidently delivered view takes longer than giving the right answer the first time. The client who has been told by a chatbot that they have a strong case for unfair dismissal may be less receptive to the duty lawyer's assessment that they are outside the minimum employment period. The chatbot's confidence has already set the anchor.
This is where the distinction between fluent AI and grounded AI matters. A generic chatbot produces confident prose from a training process optimised for plausibility. It has no corpus of Australian law behind it and no citation you can open or verify. The training data is whatever the model was built on, broad internet text that may or may not include Australian legal content, with no way for the user to know which authorities shaped the answer. If a duty lawyer asks the client what the chatbot told them about their unfair dismissal claim, the answer is untraceable. The client cannot name a case or point to a section. They have a confident summary drawn from sources they cannot identify, and nobody has checked whether it is accurate. The only way to verify it is to do the research from scratch. In a duty lawyer session where you have twenty minutes per client, scratch research is not always possible.
Verification done on a platform built for Australian law looks different. Habeas' Search Engine scans over 300,000 Australian cases and pieces of legislation in seconds, with results grounded in a closed dataset of legitimate Australian legal sources, so they are verifiable and traceable, never hallucinated. A practitioner who needs to check whether the chatbot's advice on a residential tenancy notice period or an unfair dismissal eligibility threshold was correct can run the question through a system that returns answers tied to primary law. The citation resolves to a real case or statute. The practitioner can open it and show the client where the chatbot went wrong, or where it happened to be right.
The access-to-justice gap is real, and the JUSTICE survey describes it in detail. People who cannot get a lawyer will reach for whatever is free and sounds authoritative. We are not claiming that Habeas solves the access problem. A specialist research tool is not a substitute for the legal aid funding that has been stripped out of the system over decades. Habeas gives the practitioner on the other side of that conversation a tool that grounds their own work in verified Australian law, with citations they can stand behind. If the client's chatbot advice is wrong, the practitioner needs to be able to show why, and show it quickly. Foundational research processes that used to take a full morning can now be completed in minutes. That matters when you are seeing six clients in a duty lawyer session and need to verify six different pieces of advice the internet gave them.
The JUSTICE survey is a window into something already happening at scale. Practitioners who build their intake process around the question of what their client's chatbot already told them will catch errors before they reach a tribunal. Those who do not will catch them later, when correction is more expensive and the client's confidence in the wrong answer has had longer to set.
Habeas is built for Australian legal research. Every result is traceable to its source. See for yourself at habeas.ai.
If you want to try for yourself or get in contact, book a demo with us here. We also offer the capacity for self-serve individuals to sign up, and subscribe or register a free trial at app.habeas.ai.
The legal research in this article was conducted and every citation verified using Habeas, the Australian legal AI research platform.
Hero image: Michael D Beckwith on Pexels
