OpenAI's Deep Research vs specialised legal AI

OpenAI just released Deep Research, an agent that searches the internet to answer complex questions. It's genuinely impressive—it can spend 30 minutes researching a topic, synthesise information from dozens of sources, and produce detailed reports.

For general research, it's a leap forward. For legal work, it's exactly the wrong approach.

The internet is not a legal database

Deep Research treats the entire internet as its knowledge base. That's powerful for questions like "what are the latest trends in renewable energy?" where breadth matters and approximate accuracy is fine.

Legal questions are different. When a client asks about limitation periods for breach of contract, you need:

The correct statutory provision (not a blog post summarising it)
Current case law on how courts interpret that provision
Any relevant exceptions or extensions
The specific procedural rules for the court you're filing in

An internet search might find all of these—mixed with outdated information, American law that doesn't apply, and confident-sounding articles that are simply wrong. For a general overview, that might be fine. For advice you're putting your practising certificate behind, it's not.

Different problems, different architectures

Deep Research is built for breadth. It casts a wide net, gathers everything potentially relevant, and synthesises.

Andri is built for depth and precision. We work exclusively with verified legal sources—official law reports, legislation.gov.uk, EUR-Lex, the Gazette, Companies House. Every citation traces back to an authoritative source, not to whatever the internet served up.

The difference shows up in how each system handles uncertainty:

Deep Research: Synthesises from mixed sources, may present unverified claims with confidence
Andri: If we can't verify something from authoritative sources, we say so. A qualified answer beats a confident wrong one.

What actually matters for legal research

Source verification. Can you cite it in court? If the answer relies on a blog post, a law firm marketing page, or a Reddit thread—regardless of how accurate it happens to be—you can't use it.

Temporal awareness. Law changes. A 2019 blog post about GDPR might be technically accurate for 2019 and dangerously wrong for 2025. Legal research tools need to know when sources were published and when law was amended.

Jurisdictional precision. "Contract law" is different in England, Scotland, the Netherlands, and California. Internet search doesn't naturally distinguish. A system built for legal work has to.

Citation integrity. Paragraph numbers, neutral citations, correct formatting. One wrong citation and your credibility is shot.

Where general AI falls short

We've tested general AI tools on legal questions. They're remarkably good at explaining concepts. They're also remarkably bad at:

Getting Dutch article numbers right (6:162 BW vs 6:126 BW—similar numbers, very different law)
Distinguishing binding precedent from dicta
Knowing which court a case was decided in
Recognising when a case has been overruled

These aren't edge cases. They're basic requirements for legal research that general-purpose tools weren't built to handle.

The right tool for the job

Deep Research is impressive technology. If you're researching market trends, scientific topics, or general knowledge questions, it's probably better than anything else available.

For legal work—where precision matters, sources must be authoritative, and mistakes have professional consequences—you need something built specifically for that purpose.

That's what we've built. Try it on your next matter.