Somewhere in the past week, a partner fielded a CFO question about the AI budget. The number doing the rounds gave it new urgency: Gartner is forecasting that legal technology budgets will double by 2028, driven by AI adoption across corporate legal departments and private practice. Coverage was predictable. The figure circulated, attracted the usual mix of excitement and unease, and was absorbed.
We think the more useful part came later in the release.
Buried under the spending forecast, Gartner characterised reliable AI research as one of the capabilities already delivering value to legal teams, tying that reliability, in their framing, to applications that integrate directly with primary research databases. Not applications that are powerful. Not fluent ones. Applications connected to primary sources.
That is a quiet but consequential distinction.
Before building an argument on analyst projections, we should say plainly: this is a forecast, not a settled fact. Gartner forecasts, like all market forecasts, carry a margin of uncertainty that tends to widen the further out you project. The profession has encountered confident analyst numbers before. Sometimes they resolve; sometimes they are revised, and often they capture a directional truth while scrambling the timeline.
The dollar figure itself tells Australian practitioners relatively little. What matters is the accompanying reasoning: if budget is about to move, how should the profession decide where to spend it?
Gartner's answer to that question is the part worth examining.
The criterion Gartner describes is a design distinction, not a feature claim. Their framing separates two architectures that look similar from the outside but behave very differently under pressure: applications that draw directly on primary research databases, and those that do not.
One deploys a general-purpose language model against whatever text it was trained on, generating answers from a statistical aggregation of that training. The output can be fluent and fast, and it can also be wrong in ways that are difficult to detect, because nothing anchors the answer to a specific, verifiable source. For legal research, that gap between fluency and accuracy is where professional exposure lives.
The other grounds each query against a controlled dataset of primary sources. The answer points back to a document. The document can be checked, and if the reasoning is flawed, the flaw is visible and correctable.
When Gartner ties reliability to primary database integration, they are describing this distinction without naming it. The market is beginning to formalise what good design actually looks like. That is what makes the buried criterion worth seizing.
Taken seriously, the criterion does work. A legal AI product that routes queries through a general-purpose model without grounding in primary sources does not meet it. One that surfaces answers with no traceable citation does not meet it. A product that indexes secondary commentary and calls it research sits on uncertain ground for the same reason: the chain back to primary law is either absent or obscured.
The field narrows faster than the headline spending number suggests.
Judgment still belongs to the lawyer. No retrieval architecture substitutes for careful reading and analysis on genuinely complex questions. What primary-source integration removes is the manual trawling that precedes judgment, and the opacity that buries sources in fluent but unverifiable summaries. That is the scope of the problem worth solving, and it is a significant one.
Habeas was built on this principle. The 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. Foundational research processes that used to take a full morning can now be completed in minutes. The output is a set of authorities the practitioner can open, read, and test directly against the question.
If Australian legal budgets follow anything like the trajectory Gartner projects, procurement decisions are coming. Not abstract decisions about AI readiness, but specific decisions about which products to trial and which to pay for.
The question that cuts through the noise is not "does this use AI?" Almost everything will. The question is whether the AI is anchored to primary sources you can verify. Gartner, quietly, has named that as the criterion for reliability.
We built for it. If the criterion is traceability, the test is straightforward: run a query on a matter you researched last week and follow the citation back to the source. Habeas.
The legal research in this article was conducted and every citation verified using Habeas, the Australian legal AI research platform.
