Legal research was the first part of legal work that generative AI touched, and it is where the gap between a real product and a plausible-sounding one is widest. It is also the area where the market structure changed most in the last eighteen months.
The thing that actually distinguishes these products
Grounding. Whether the system’s answers are tied to a licensed body of primary law that it can cite and you can open, or whether it is generating text that resembles legal analysis.
This is not a subtle technical distinction. It is the difference between a tool that can be relied on with verification and a tool that produces the fabricated citations now recorded in more than 2,000 court decisions worldwide. A grounded system retrieves a case and summarises it. An ungrounded one predicts what a case would look like.
When evaluating anything in this category, the first question is what corpus sits underneath, and whether every proposition in an answer links to a document you can read in full.
The incumbents
Thomson Reuters and RELX, through Westlaw and LexisNexis, still hold the position that matters most: licensed primary law, editorial enhancement, and the citators that tell you whether a case is still good. That asset is expensive to build and has not been disrupted.
Thomson Reuters has been moving quickly on the AI layer. It expanded CoCounsel Legal into the United Kingdom in January 2026, put agentic AI into more than 200 law schools the same month, and launched the next generation of CoCounsel Legal in August 2026. It has also partnered with model providers rather than building everything in house, including the expanded Anthropic partnership connecting Claude to CoCounsel.
The strategic logic is simple. If the model is the commodity and the content is the moat, then the incumbents are defending the right asset.
The new shape of the middle market
The most consequential structural change was Clio’s $1 billion acquisition of vLex, completed in November 2025 alongside a $500 million Series G at a $5 billion valuation.
vLex brought more than a billion editorially enriched documents across 110 jurisdictions. Clio brought a practice management platform used by legal professionals in more than 130 countries. The combination puts credible primary law research inside the system where smaller firms already run their matters, which is a materially different proposition from asking those firms to buy a separate research subscription.
For solo and small firm practitioners, that is the most important development in legal research pricing in a decade.
The AI-native layer
Harvey, at an $11 billion valuation as of March 2026, and Legora, at $5.55 billion with 800 customers across more than 50 markets, are not straightforwardly research products. They are workflow platforms that include research alongside drafting, review and increasingly agentic task execution.
That distinction matters at procurement. Buying one of these to replace Westlaw is a category error. Firms running them successfully generally run them alongside a primary law subscription, not instead of it.
How to evaluate a research tool properly
The demo problem in this category is severe, because a fluent wrong answer and a correct answer look identical to someone who does not already know the law.
The only reliable method is to test on questions you know the answer to.
- Take five questions from your own recent matters where you know the correct authority, including one where the answer is genuinely unsettled.
- Check whether every proposition links to a source you can open. If a claim has no link, treat the tool as ungrounded for that claim.
- Test the unsettled question hard. Good tools say the position is unclear and show you the competing lines. Weak ones pick a side confidently.
- Check currency. Ask about something that changed in the last three months. Find out what the actual content update lag is.
- Test outside the core jurisdiction. Coverage of England and Wales, Canada, Australia or the EU is often far thinner than United States coverage, and vendors rarely volunteer this.
- Look for the citator. Whether an authority is still good law is the single most consequential fact about it, and it is the function most commonly missing from AI-native tools.
What has not changed
Verification. Every authority relied on in a filing gets opened and read, regardless of which system produced it. That was true when research meant a library, it was true when it meant a keyword search, and the growing sanctions record shows it is true now.
What has changed is speed to a first orientation on an unfamiliar area. That is a real and substantial gain. It is not the same as the answer.
Related: what the AI hallucination sanctions record tells lawyers and what the 2026 funding rounds mean for buyers.
Where the legal industry reads first.
Enjoyed this article? Get the biggest legal industry updates, deals, appointments, insights and expert interviews in your inbox, free.
No spam. Unsubscribe anytime.