BLOG

AI & tenancy advice: Helpful tool or hidden risk?

05/03/2026 • Brendan Ross

Whether it’s ChatGPT, Copilot, Claude or Gemini, AI tools are becoming increasingly part of everyday information-seeking and problem-solving. They are always online, and can help break down complex material and offer quick explanations when people are under pressure. It can feel like talking to a confident person, which – as social creatures – we find trust-worthy. They can also create professional and confident-sounding documents, which is an area people can struggle with when trying to address a legal issue or go to the NSW Civil and Administrative Tribunal (NCAT).

With their growing presence and influence it’s important to understand the limitations – especially if you are thinking about using AI in relation to navigating a legal dispute. AI is not a source of legal advice, and it cannot grasp the context, history or emotions behind a tenancy issue. Knowing how AI works, and where its limits are, is important for safe use. It also reinforces why human experience, judgment and advocacy remain central to resolving tenancy problems.

What happens to your information when you use AI

When you type information into or upload a file to a public AI tool, the content is usually stored on external servers. It may also be used to train future AI models — even if you delete the conversation (Digital NSW, 2025). Most public AIs are run by overseas companies and the data is stored overseas, so Australian privacy laws may not apply (Guidance on privacy and developing and training generative AI models, Office of the Australian Information Commissioner). Although this is not true of companies with an Australian link, such as having an Australian office.

Important: Never upload personal or sensitive information (like your date of birth, home address, driver’s licence or other identification numbers) into public AI tools.

What AI actually does (and doesn’t do)

Current AI tools work by recognising patterns in huge amounts of text and predicting what words should come next. This is why they can explain ideas or summarise information, even though they do not genuinely understand your personal situation.

These tools are examples of 'generative AI' — systems that create new text based on what they have learned from their training data. The large language models (LLMs) behind them draw on information from many different sources and time periods. This allows them to respond quickly, but it also means they can produce information that is outdated, incomplete or simply wrong (CSIRO, 2020).

Because AI is trained on existing information, it can repeat the biases or gaps contained in that data (AHRC, 2020; CSIRO, 2020). As a result, AI often overlooks or misunderstands the experiences of renters from diverse backgrounds.

AI can be a useful way to explore issues or prepare questions, but it has important limits. These limits become especially stark in tenancy matters, where context, evidence and lived experience often determine the outcome.

Some definitions

Artificial Intelligence (AI) – Technology that allows computers to perform tasks that usually require human thinking — such as learning, reasoning, and making decisions.

Generative AI –  creates new text, or even pictures or sounds, based on the patterns in its training data.

Large Language Models (LLMs) – a type of generative AI trained on huge amounts of text. LLMs predict the next likely word in a sentence, which is why they can produce helpful-sounding answers even though they do not “understand” your personal situation. ChatGPT, Gemini and Copilot are examples of generative AI using LLMs for their text responses. They may use other models for multimedia.

The NSW Government's Digital NSW has a useful summary of other common terms that may be helpful.

What AI could miss in tenancy matters

Tenancy issues often depend on the kinds of details that AI cannot reliably recognise like history, verbal agreements, or the behaviour of a landlord or agent. These details can completely change the nature of a dispute, and they often sit outside what an AI tool can understand or accurately interpret (South-East Monash Legal Service, 2024).

AI tools can summarise text you upload, but they cannot understand the full story behind it. They cannot see the mould in your bathroom, hear the tone of a phone call, or recognise when a conversation felt intimidating or unsafe. They cannot directly weigh evidence, understand stress or urgency, or identify how discrimination, disability or financial pressure might be shaping your housing situation – although it may be able to guess that a person in similar circumstances faces similar frustrations.

Because AI is trained on information from many places, it may also mix up NSW tenancy law with rules from other states or countries. An answer may sound confident, but still be inaccurate or unsuitable for your circumstances. You can attempt to avoid this by being very specific about your situation. The challenge is you might not know, and neither does the AI, all the questions that may be relevant.

While AI can sound reassuring, it has no obligation to be correct, fair or confidential (Digital NSW, 2025; NCAT Procedural Direction 7, 2025). A helpful-sounding answer may still be incomplete, misleading or unsuitable for your situation.

Putting AI to the test

We tested a relatively simple question, with a couple of traps, on 3 of the most common and publicly available tools – OpenAI's ChatGPT, Google's Gemini and Microsoft's Copilot. We used the free basic version of each of these tools, without being signed in. This does restrict the model and potentially reduce its usefulness. We've compiled the three answers into this document so you can see the full responses provided by each.
 

Our prompt to all three was:

"I'm renting in NSW. My landlord has told me they want to move back in. They gave me 41 days notice by text. When do I have to move?"

We're looking for five elements in the answers. So how did they do?

  1. Identify that a landlord moving back in is a ‘reasonable ground’ requiring a notice with particular information, required documentation and evidence.

    All three models produced some mistakes on this point. ChatGPT and Copilot both missed the requirement to give a 'tenant information statement' and ChatGPT additionally missed the requirement for an evidence statement from the owner.

    Gemini was the closest to a correct answer, but over-reached by suggesting that the evidence statement needed to be a statutory declaration. This is a problem because a person may assume a notice wasn't valid when it was, and be frustrated when the Tribunal or somewhere else doesn't act as the model might predict. This was the first hallucination – use of a statutory declaration wasn't mentioned in the sources Gemini provided.

  2. Identify that notice by text message (SMS) is not usually valid. While difficult, it can be made an agreed service.

    All three models produced a generally correct answer here, sharing an assumption that SMS notice would not be valid, but identifying that it is technically possible to make SMS an agreed method of service. None expanded on the extra detail about why it is difficult to use SMS, but that's not really necessary.

  3. Identify the correct length of notice is either 60 or 90 days, depending on how long your current fixed term (if you have one) is.

    Gemini predicted this answer fairly well, identifying the need for more information but asserting the rules accurately here. Both ChatGPT and Copilot missed the possibility that a notice could be 60 days if the person has been renting with a current fixed term of 6 months or less and gave the impression that notice always needed to be 90 days. Again, this may lead a renter down the wrong path in disputing a notice.

  4. Identify that tenant does not have to move without an NCAT order.
    Both Gemini and Copilot informed the user that without an NCAT order, the person does not have to move. Copilot was more accurate here, stating that a person does not have to move unless both elements – a valid termination notice, and an NCAT order – have been met. 

    While ChatGPT identified that the notice was likely invalid, it appeared to assume the next step would be for the tenant to dispute the notice at NCAT. While this is an option available to tenants, it is rarely the next step.

  5. Use and provide authoritative sources like legislation, or at least reliable, legally-checked sources like government sites or legal practices.

    All three models used the NSW government's websites – mainly the nsw.gov.au pages on renting, with one use of Legal Aid, as their main source, linking to relevant pages. We would call this a reliable rather than an authoritative source because while a lot of effort is put into ensuring information is accurate it is also trying to communicate in plainer language. This means websites like this still rely on an interpretation by government workers, rather than what the legislation actually says or how a court or tribunal will interpret it.

    Co-pilot stopped there, but the other two also included at least one real estate agency website and this is slightly more concerning. While real estate agents have professional licencing requirements we do occasionally find errors on real estate pages, or at least interpretations of the law that are more favourable to the agent/landlord than the tenant. In this case, the websites were largely just summarising the changes in a straight-forward way and none of the sources used in these responses had significant issues beyond some gaps in detail.

 

So all three models got things broadly right, but with some mistakes. Those mistakes may have had significant impact on a person asking the questions, and certainly are things that we would ensure Tenant Advocates are trained to avoid. 

All three models offered a next step of writing back to the landlord, but creating new documents is where mistakes can start to compound. If you take action based on incorrect information then things can get really tricky. This is also the area where the benefit of an Advocate talking through your options with you and helping you select the most strategic and appropriate course for you can't be easily replicated by AI. 

What the Tribunal Says About Using AI

Procedural Direction 7 (2025) by the NSW Civil and Administrative Tribunal (NCAT), sets out how AI can – and cannot – be used in Tribunal matters. The Direction recognises that AI tools can be helpful, but it also highlights serious risks.

NCAT allows people to use AI for simple tasks, such as:

  • drafting or summarising information
  • preparing chronologies
  • reviewing documents

However, there are strict limits.

You must not:

  • use AI to prepare evidence, witness statements, statutory declarations or anything meant to reflect a person’s own account
  • upload confidential, sensitive or restricted information into public AI tools
  • rely on AI to check citations, legislation or case references

NCAT also warns that AI tools may:

  • invent cases, legislation or quotes (“hallucinations”)
  • rely on outdated, incorrect or irrelevant information
  • include biased data
  • store prompts or uploaded information, risking confidentiality breaches

If AI is used to draft written submissions, you must personally check that every reference, legal authority and factual statement is correct. AI cannot do this verification for you.

We think these tips are pretty solid – they acknowledge the limits of AI while recognising the potential benefits in organising information.

The importance of human engagement with resolving tenancy issues

AI is becoming more common. These tools can save time, but they cannot understand or respond to the human experiences behind tenancy problems – things like trauma, financial stress, discrimination, disability, cultural background or language barriers (South-East Monash Legal Service, 2024; Centre for Culture, Ethnicity & Health, 2025).

AI may help explain information, but it cannot build trust, recognise fear or support someone through a stressful dispute. It cannot replace the safety, dignity, and human understanding that comes from speaking with someone who listens and sees the whole person.

This is why Tenant Advocates at Tenants Advice and Advocacy Services remain essential. Advocates bring legal knowledge, negotiation skills, cultural awareness, empathy and trauma-informed practice. They understand the pressures tenants face and can help people navigate conversations, conflict and systems that AI cannot meaningfully interpret.

As AI continues to evolve, experts emphasise that trustworthy systems must be fair, transparent and guided by human oversight (AHRC, 2021; CSIRO, 2020). In tenancy work, AI can improve access to information, however decisions, advice, and advocacy must remain human-led.