Legal research has always needed to be accurate. What’s changed is what accuracy alone now fails to protect you from. AI has accelerated research dramatically. But it’s also exposed a new class of risk: answers that sound right, look polished, and move fast, without being defensible. In today’s environment, that’s not a productivity gain. It’s a liability.
As Sarah Jacobson, Director of Knowledge Management at MinterEllison, put it in our recent webinar, Beyond answers: a new way to do legal research:
“Defensible research is work that could withstand scrutiny at every level. A partner could hand it to a client, a barrister could rely on it in court, and a regulator could follow the trail.”
That bar is higher than it used to be, and AI is one of the reasons why.
The hidden risk: confidence you haven’t earned
One of the most dangerous qualities of generative AI is how authoritative it sounds. Outputs are fluent, structured, and decisive.
- That polish can mask fundamental problems:
- Case citations that don’t exist
- Principles that are mischaracterised
- Nuance lost through over‑confident summarisation
- In the past, there were things lawyers could safely assume.
“In the past, if you got a junior to do legal research … and they gave you a case citation, you would assume that case citation is correct,” Sarah notes. “You can’t even assume the case citation is correct anymore. You’re starting right at the very beginning – does this case actually exist?”
That shift matters. Because when errors slip through, they don’t stay theoretical. Advice goes to clients. Submissions get filed. Business decisions get made.
And accountability still sits with the lawyer.
Why verification now starts much earlier
Traditionally, senior lawyers reviewed research for quality and relevance. Today, they’re often verifying the fundamentals that used to be taken for granted.
That changes how work gets done:
- juniors need tools that surface sources, not just summaries
- seniors need visibility into reasoning, not just conclusions
- firms need confidence that AI use won’t introduce ‘silent risk’
As Sarah explains, even subject‑matter expertise isn’t always enough if you can’t see the underlying material:
“If you’re looking at a case summary, an expert isn’t going to be able to say it’s wrong without reading the case. There’s a level of basic manual verification that’s going to be required for some time.”
Speed without traceability doesn’t save time. It just pushes the work downstream – often when the stakes are higher.
What firms should expect from AI‑assisted research
For firms navigating AI adoption, the goal isn’t automation for its own sake. It’s research you can stand behind.
That means AI should support legal judgment by:
- showing the research path, not hiding it
- linking directly to primary law
- making verification part of the workflow, not an afterthought
- “Technically correct” research may tick a box. But defensible research considers nuance, competing arguments, and context.
“You can have something that’s technically 100% correct,” Sarah says, “but it’s not necessarily something you can confidently take to a client.”
That distinction is where professional judgment still matters most.
The bottom line
AI is embedded in legal work. Courts are paying attention. Clients expect transparency. And the cost of getting it wrong is rising.
The firms that will thrive aren’t the ones that adopted AI first. They’re the ones that built governance, verification, and culture around how it’s used.
Because in an AI‑first firm, accountability hasn’t moved.
It still sits with the lawyer.
Watch webinar on demand.
Sarah Jacobson joins Amy Hope (BlueScope), Tej Kadiyala (Thomson Reuters) and James Jarvis (Thomson Reuters) to unpack what defensible, AI assisted legal research looks like in practice, and how firms are adapting workflows to meet rising expectations. Watch now →
