Australia's largest live Fair Work case-law database — and why ChatGPT and Claude can't match it

Ask ChatGPT or Claude "how much notice does an Australian employer have to give for redundancy?" and you'll get a confident answer. Some of it will be right. Some will reference outdated NES tables, an obsolete cap, or US notice-period concepts that don't exist here. The model has no idea which is which — its training data ended a year or more ago, and it doesn't know what the Fair Work Commission decided yesterday, last week or last quarter.

That's not a flaw in the model. It's the fundamental limit of how frontier LLMs work. They're trained on a snapshot of the internet, sealed at some past date, and they can't reach out to Australia's Fair Work Commission, Federal Court, or Fair Work Ombudsman to check what's actually decided.

For an Australian HR team running a real workplace, that gap matters. Cases get decided weekly. Awards change. Penalties get awarded. The decision that came down on Friday — the one that changes how your case-in-question will be weighed — is not in any frontier LLM's knowledge.

FairWork Mate runs a live database of Australian workplace decisions. As of May 2026, that's 320+ cases across three jurisdictions:

• Fair Work Commission (FWC) — the federal tribunal hearing unfair dismissal, general protections, bargaining, anti-bullying and modern award matters. Thousands of decisions a year; we're publishing the substantive ones with plain-English summaries.
• Federal Court of Australia (FCA) — the higher court hearing Fair Work civil-penalty proceedings, sham contracting prosecutions, larger appeals. Where the real precedent gets set.
• Fair Work Ombudsman (FWO) — enforcement outcomes, enforceable undertakings, litigation against employers. 212 of these in the corpus, from $1M+ restaurant prosecutions to $20M university settlements.

Each case is summarised into facts, outcome, employer implication and employee implication. Each one has tags, an industry classification, a penalty figure where applicable, and a permanent URL. Each one is also embedded as a high-dimensional vector — which means our AI advisor can find the most relevant 2-3 cases to your question in milliseconds and cite them directly in the answer.

Ask FairWork Mate AI "is performance management of a pregnant employee at risk under general protections?" and the answer doesn't just give you the legal framework — it cites specific FWC and FCA decisions with the actual reasoning the bench applied. The same question to ChatGPT gives you a generic disclaimer-laden essay with no citations.

Q1: "Can my employer reduce my pay below the award without my consent?"

• ChatGPT: Generic answer referencing US at-will employment in places, hedged Australian context, no Fair Work Act citation, no case law.
• Claude: Better answer with NES references, but cuts off at "consult a lawyer" without naming any decision.
• FairWork Mate AI: Cites the modern award floor, references s 323 of the Fair Work Act, names a recent FWC decision on unilateral reduction, links to the case-page summary. Total response time: ~3 seconds.

Q2: "We have a casual who's worked the same Tuesday-Thursday shifts for 14 months. Are we exposed?"

• ChatGPT: Vaguely correct ("you may be exposed to casual conversion claims"), no specifics on the 6-vs-12-month test, no case citations.
• Claude: Closer — mentions the 2024 Closing Loopholes reforms but doesn't name a specific case on the "regular and systematic" test.
• FairWork Mate AI: Confirms the 12-month threshold (6 months for small business), cites the FWC's interpretation of "regular and systematic" with named decisions, and links to the FairWork Mate casual conversion calculator.

Q3: "What's the precedent for $50k+ general protections compensation in 2025?"

• ChatGPT/Claude: Can't answer — they have no recent decision data. Generic "compensation depends on circumstances" response.
• FairWork Mate AI: Names 3-5 recent FCA decisions with the specific compensation awarded, the reasoning, and the relevant industry context. Citations and links included.

Every case we publish makes the next answer better. The AI retrieval is similarity-based, so a question about retail wage theft pulls the most-similar retail wage-theft cases automatically. The corpus going from 320 to 500 to 1,000 published decisions doesn't just give us more pages — it gives the AI a denser retrieval space, more specific citations, and more accurate edge-case reasoning.

Frontier LLM training is competitive but bounded. ChatGPT 5 will still have a training cutoff date and still won't know about Tuesday's FWC decision. Claude 4.8 likewise. The moat for Australian workplace law isn't bigger models. It's live, jurisdiction-specific, plain-English-summarised case data, retrieval-attached to the AI. That's what we're building.

Our roadmap is to be at 1,000+ published Australian workplace decisions by end of 2026, with daily refresh, full embeddings, and free 2-questions-a-day AI access at /advisor. For HR teams running real estates of workers and contracts, that's already the most useful Australian workplace AI tool on the market. For ChatGPT and Claude users frustrated by generic answers to specific Australian questions — we're worth a 5-minute trial.

If you run people in Australia, three concrete uses:

• Case-law research without the lawyer-billable hour. The "what's the precedent for X" question used to mean an hour of legal research at $400-700/hr. Now it's a 30-second AI advisor query with cited decisions. Validate it with your lawyer if it's high-stakes; skip the lawyer's research time if it's not.
• Drafting compliant documents grounded in current law. Show-cause letters, casual conversion responses, performance plans — drafted by our AI with the relevant statutory references and recent case law cited in the same answer.
• Industry-specific intelligence. What's happening in aged-care underpayment in 2026? Which universities have been caught for casual academic underpayment? What's the FCA awarding for unfair dismissal in finance? Searchable by industry and tag.

For solo HR managers, small-business owners and operators, /advisor is free for 2 questions a day. For teams running multiple workflows, the For Business tier from $499/mo gives unlimited questions, longer context, and the API for embedding into your existing tools.

The data is the moat. The AI on top of it is the interface. Both are live now. The only AU workplace AI source citing 320+ live decisions is FairWork Mate. The number's growing weekly.