The real divide in legal AI isn't the model

Put two firms on the same AI assistant and you'll get two different qualities of answer. The gap isn't the technology — it's what the technology is allowed to access.

Give two law firms anywhere in the world the same AI assistant, point it at the same question, and you won't get the same answer. Not because the model is different— but because of what sits behind it and fuels the responses. One firm's assistant draws on precedent its knowledge team has spent years vetting, tagging, and keeping current. The other's draws on whatever happened to be saved in the nearest folder. Both answers will sound equally confident and convincingly articulate. However, only one of them is correct. I'm writing this from Sydney but ask a knowledge team in London or Toronto and they'll tell you the same thing: the gap has nothing to do with geography.

Every firm is shopping from the same shelf

Copilot, Harvey, Legora, Claude — the frontier models stopped being anyone's private edge a while ago, and most firms now have at least one credible AI assistant live for their lawyers, often more than one. So the real competitive question was never which logo sits on the licence. It's what that model is allowed to see and access. Handed a stale draft or a document nobody would stand behind, an AI assistant answers with exactly the same confidence as one handed the firm's genuine best work. There is no way to tell the difference from the output alone.

I asked the person who builds this for a living

I put that question to my colleague Alex Smith, Senior Director of Product – Search, Knowledge & AI at iManage, who spends his working days thinking about what AI tools should and shouldn't be allowed to see. “Context is important to lawyers because legal work sits in a combination of experience in the context of the type of work, the type of area you work in, and context can be very important for things like jurisdiction,” he told me. “Otherwise, you're going to get answers which come from a lot of different places — not always the great ones.”

That instinct shows up everywhere in how Smith talks about the problem. According to the iManage Knowledge Work Benchmark Report 2026, professionals lose an average of 37 minutes a day just searching for documents — time that should be spent working, not hunting for something to work from. “I kind of see a context window as like a box,” he says. “And if the right things don't end up in the box, then AI is going to reason on the wrong things.” Leave that to chance, he argues, “and we're kind of almost throwing things blind into that window.”

The divide is a knowledge gap, not a model gap

The same Benchmark Report found that organisations at the highest level of knowledge maturity are nearly twice as likely to report year-over-year revenue growth as those at the lowest, and that more than a third have already had a documented policy breach traced back to ungoverned AI use. Smith sees the same pattern from the product side. “There is a lot of risk involved” when information is left to move on its own, he says — which is why his advice to knowledge teams is blunt: “Knowledge needs to run over the firm's best materials, their brand and their core.”

What Insight+ is actually built to do

This is the case for treating knowledge as governed infrastructure, not a filing exercise — and it's exactly what iManage Insight+ is built for. “Collections is how you build knowledge around pieces of work that lawyers consistently do, and putting context around that,” Smith explains. Firms decide which precedent, template, and matter history an AI assistant is allowed to draw from, so a question about an obligation or a deal structure gets answered from what the firm actually knows, not from whatever a lawyer happened to have open.

None of this happens on its own. Every precedent worth trusting has been selected, tagged, and kept current by a knowledge team that was doing this work long before anyone mentioned AI models. What's changed is what that curation now feeds. Smith describes Insight+ connecting into a firm's other tools through MCP, or Model Context Protocol: “MCP allows you to communicate without moving information.” The firm's knowledge stays exactly where it's governed — the AI comes to it, not the other way round.

The advantage was never the model

For firms with lawyers spread across multiple offices and time zones — Sydney to Singapore, London to New York — that distinction matters more than it might elsewhere. Properly governed context is reachable by anyone with permission, regardless of which office they sit in or how many years they've been at the firm. Allens, one of Australia's largest firms, has already built exactly this kind of foundation, using curated knowledge collections to underpin its AI strategy across the firm. Whichever model a firm licenses next, the lesson holds: the edge was never the model. It was always what the model was allowed to know.