Most AI research questions have a straightforward answer. What's the limitation period? What does section 8 of the Act say? What's the CPR rule on this? For those, you want a fast, cited answer and you want to move on.
But some questions genuinely don't have a simple answer. The authorities conflict. The statute is ambiguous. The client's situation sits at the intersection of two different legal frameworks. For those, what you need isn't just a faster lookup, it's analysis that actually thinks through the problem.
That's the specific thing Extended Thinking is for.
What it actually does differently
Standard AI research retrieves relevant sources and synthesises them into an answer. It's fast and it works well for questions where the law is settled.
Extended Thinking does something additional: it shows you the reasoning, not just the conclusion. Before giving you an answer, it works through the question step by step, which sources are most relevant, how they interact, where there's genuine tension between authorities, what the competing interpretations are. You can see the chain of thought.
This matters for two separate reasons. First, the analysis is more thorough for complex questions. Second, you get an audit trail of how the conclusion was reached, which is useful when you need to be able to explain your research basis.
When it's worth using
The clearest use cases are questions where the law is genuinely unsettled: a Supreme Court decision that may have shifted how a lower court will read a statutory provision, an area where there are conflicting Court of Appeal decisions, a point of construction where the wording is ambiguous.
It's also useful for questions that span multiple legal frameworks. A commercial property dispute that involves landlord and tenant law, planning law, and a contractual claim isn't complicated because any single area is obscure, it's complicated because the interaction matters. Extended Thinking is designed to track those interactions explicitly.
And it's useful when you're advising on something where the stakes are high enough that you want to be confident you haven't missed a relevant argument, not just that you've found the obvious one.
When it isn't worth using
Standard research mode is faster and equally reliable for well-settled questions. If you want to know the current disrepair threshold under the Landlord and Tenant Act 1985, or what form a section 21 notice needs to take, Extended Thinking adds overhead without adding insight.
The honest version of this is: most day-to-day research doesn't need it. It's a tool for specific situations, not a default mode.
The compliance angle
There's a practical argument for using Extended Thinking on complex matters beyond the research quality: the visible reasoning chain is documentation.
When you advise a client on a complex legal question, the SRA expects you to be able to demonstrate the basis for that advice. Traditionally that's a research note, a file note, a record of the authorities you considered. Extended Thinking produces that record automatically, what sources were considered, how they were weighed, where uncertainty was identified.
Saved to the matter record, it shows that the research was thorough and that the advice was grounded in a genuine analysis of the applicable law. That's useful if the advice is ever questioned, and it's the kind of documentation that supervision frameworks should be generating anyway.
See Extended Thinking on Writford for a detailed look at how it works.