When it comes to document review, people keep asking whether GenAI can “think like a lawyer.” But let’s be real—when was the last time anyone questioned:
- Why a predictive coding model assigned a document a 0.62 relevance score?
- How a CAL process prioritized likely relevant documents?
- Whether the keywords in a search were ever properly tested?
We don’t ask these questions nearly enough. And they matter way more than debating whether AI has “reasoning” skills.
Let’s break it down. Large Language Models might not “think,” but they follow rules and deliver results. Ever asked ChatGPT to:
- 𝘋𝘳𝘢𝘧𝘵 𝘢 𝘤𝘭𝘪𝘦𝘯𝘵 𝘦𝘮𝘢𝘪𝘭 𝘵𝘩𝘢𝘵 𝘴𝘢𝘺𝘴 “𝘱𝘦𝘳 𝘮𝘺 𝘭𝘢𝘴𝘵 𝘦𝘮𝘢𝘪𝘭”—𝘸𝘪𝘵𝘩𝘰𝘶𝘵 𝘢𝘤𝘵𝘶𝘢𝘭𝘭𝘺 𝘴𝘢𝘺𝘪𝘯𝘨 𝘪𝘵?
- 𝘞𝘳𝘪𝘵𝘦 𝘢 𝘸𝘦𝘥𝘥𝘪𝘯𝘨 𝘵𝘰𝘢𝘴𝘵 𝘵𝘩𝘢𝘵’𝘴 𝘤𝘩𝘢𝘳𝘮𝘪𝘯𝘨 𝘣𝘶𝘵 𝘯𝘰𝘵 𝘵𝘳𝘺𝘪𝘯𝘨 𝘵𝘰𝘰 𝘩𝘢𝘳𝘥?
- 𝘙𝘦𝘸𝘳𝘪𝘵𝘦 𝘵𝘩𝘦 𝘙𝘶𝘭𝘦𝘴 𝘰𝘧 𝘊𝘪𝘷𝘪𝘭 𝘗𝘳𝘰𝘤𝘦𝘥𝘶𝘳𝘦 𝘢𝘴 𝘢 𝘒𝘦𝘯𝘥𝘳𝘪𝘤𝘬 𝘓𝘢𝘮𝘢𝘳 𝘴𝘰𝘯𝘨?
- 𝘗𝘳𝘰𝘷𝘪𝘥𝘦 10 e𝘹𝘢𝘮𝘱𝘭𝘦𝘴 𝘰𝘧 𝘸𝘩𝘢𝘵 𝘢 𝘭𝘢𝘸𝘺𝘦𝘳 𝘮𝘪𝘨𝘩𝘵 𝘢𝘴𝘬 𝘊𝘩𝘢𝘵𝘎𝘗𝘛 𝘵𝘰 𝘥𝘰?
Chances are, you liked the output. Did the language model “reason”? No clue. But it understood what you wanted—𝒂𝒏𝒅 𝒅𝒆𝒍𝒊𝒗𝒆𝒓𝒆𝒅. And that’s really what matters. Let’s face it, nobody has ever said: “𝘐 𝘓𝘖𝘝𝘌 𝘵𝘩𝘦 𝘣𝘦𝘴𝘵 𝘮𝘢𝘯 𝘴𝘱𝘦𝘦𝘤𝘩 𝘊𝘩𝘢𝘵𝘎𝘗𝘛 𝘸𝘳𝘰𝘵𝘦 𝘧𝘰𝘳 𝘮𝘺 𝘧𝘳𝘪𝘦𝘯𝘥’𝘴 𝘸𝘦𝘥𝘥𝘪𝘯𝘨, 𝘣𝘶𝘵 𝘐’𝘮 𝘸𝘰𝘳𝘳𝘪𝘦𝘥 𝘪𝘵 𝘥𝘪𝘥𝘯’𝘵 𝘳𝘦𝘢𝘭𝘭𝘺 𝘳𝘦𝘢𝘴𝘰𝘯.”
Here’s the deal.
GenAI isn’t having an existential crisis over privilege calls. It’s not paralyzed with uncertainty over relevance. And if the process is designed well, it doesn’t over-lawyer or misinterpret (ignore) coding instructions.
The real question isn’t whether AI can reason—it’s whether we’ve designed a process to produce explainable, repeatable, and defensible outcomes. Still skeptical? There’s proof it works. On every document classification.
I’ve managed hundreds of predictive coding projects. And while TAR/CAL is a massive improvement over keyword searching, it still misses relevant docs—or surfaces irrelevant ones—with no clear way to explain why. Then again, two weeks later, humans often can’t explain why they marked a document responsive. That’s also a problem.
Transparency matters—whether it’s a machine-learning model or a first-year associate making the call. Good AI systems don’t just give results—they provide clear reasoning behind them.
So, back to that reasonable doubt….If we hold GenAI to a higher standard than the traditional review process, we’re asking the wrong question. The real test isn’t whether AI “reasons” like a lawyer—it’s whether it delivers better, faster, and more defensible results than the status quo.
And on that, there’s not much doubt at all.