Harvey vs Legora vs Luminance: Legal AI Tools for UK Firms Compared

The Short Answer

The tools fall into three groups, and the right group matters more than the right brand:

• General legal AI platforms (Harvey, Legora): one assistant for drafting, research and document work across the firm. Both name large UK firms as customers.
• Contract review specialists (Luminance, Kira, Spellbook): built around reviewing and drafting contracts, from due diligence at volume to clause work inside Word.
• Research assistants from legal publishers (CoCounsel, Lexis+ AI / Protégé): answers grounded in Westlaw, Practical Law or Lexis content. That grounding is the direct answer to the fabricated-citation problem UK courts have been sanctioning.

None of them publishes a UK price, and none of the security pages we read names a UK-only hosting region. Both are questions to put to the vendor in writing.

The Comparison

What each vendor or a named customer publicly states, as at 26 October 2026:

Harvey

Harvey's UK story starts with Allen & Overy's launch partnership in February 2023, rolled out to more than 3,500 lawyers across 43 offices. Its customers page now shows A&O Shearman, Macfarlanes, CMS, Lewis Silkin and Gowling WLG among others. The Bar Council's November 2025 guidance names it alongside Copilot, Gemini and Perplexity, which tells you how mainstream it has become.

On data, Harvey's security page states it is hosted on Azure, offers processing in the EU and Switzerland or Australia, and contractually prohibits model providers from training on customer data. It does not name a UK region. There is no public pricing page.

Legora

Legora was called Leya until February 2025. The same month it announced a firmwide rollout at Bird & Bird, and Trowers & Hamlins appears on its customers page. In Google's Keyword Planner for the UK (October 2026), “legora” sits in a higher monthly search band (10,000-100,000) than any other tool on this page.

Its security page states that customers control where their data is stored and that it will not use customer data to train or fine-tune models. No UK region is named and there is no public pricing.

Harvey or Legora? On public information they are close: both are general platforms, both name large UK firms, neither publishes prices, and both put training restrictions in writing. The deciding questions are practical ones only a pilot answers: how each handles your document types, how it fits your document management system, and what your lawyers actually use after the first month.

Luminance

Luminance describes itself as born out of the University of Cambridge and states it has more than 1,000 customers in over 70 countries. It is a contract review and due diligence product first, which suits firms whose main volume is reviewing agreements rather than drafting or research. Pricing is by demo only, and we could not confirm a named UK law firm customer or a hosting statement from a primary source. Ask for both.

Kira, Spellbook, CoCounsel and Lexis+ AI

• Kira, now part of Litera, is positioned for high-volume, high-stakes review and due diligence. It is the other name to compare against Luminance.
• Spellbook works inside Microsoft Word for law firms and in-house teams. It is the only tool here that publishes its pricing basis (team members on the licence), and it offers a 7-day trial.
• CoCounsel Legal launched in the UK in January 2026 with research grounded in Westlaw and Practical Law.
• Lexis+ AI and Protégé: LexisNexis launched Protégé for UK lawyers in July 2025. Its in-house product page states data centres in Europe and that inputs are not used for training.

If your firm already subscribes to Westlaw or Lexis, the publisher's own assistant is a natural first pilot, because its answers point at sources your lawyers already use.

What to Ask Every Vendor Before You Sign

The SRA's August 2026 warning notice and the Law Society's guidance turn most of these into obligations rather than preferences. Get the answers in the contract:

1. Is it a closed system? Who outside your firm can see inputs and outputs, including the vendor's staff and its model providers?
2. Training: is training on your data prohibited for the vendor and every model provider behind it?
3. Location: where exactly is client data processed and stored, and can you choose? Check the answer against your client engagement terms.
4. Retention and deletion: how long are prompts, documents and outputs kept, and can you delete a matter's data on request?
5. Sources: does every answer link to the document or authority it came from, and what happens when it finds nothing?
6. Logs: can you export who used it, on which matter, with which documents?
7. Exit: what can you take with you, in what format, if you leave?

Why this matters is covered in AI for UK law firms: SRA rules and what to automate, including the court cases that made citation checking a firm-level duty.

How to Run a Fair Pilot

Vendor demos use documents the tool has seen before. A pilot that tells you something uses yours:

1. Pick one workflow and 30-50 documents your lawyers have already reviewed, so the right answers are known.
2. Run each shortlisted tool on the same set and score it against the lawyers' answers: what it found, what it missed, and what it invented.
3. Count misses separately from false alarms. In contract review a missed clause costs far more than a flagged one a lawyer dismisses.
4. Include documents where the answer is “not there”. A tool that always finds something is guessing.
5. Measure lawyer time per document with and without the tool, on the same people.

This is the same evaluation I use on my own systems, where I publish retrieval results including the weak suites.

When Building Beats Buying

Platforms are general by design. A custom build earns its cost when the documents are specialised, the work has to happen in Word with your own clause logic, or you need full control of where client data goes. I built a clause analysis system for an investment fund law practice for exactly those reasons: six clause categories in 150-200 page partnership agreements, analysed against the firm's own library, inside a Word add-in, with zero-retention processing. The architecture and the build-or-buy trade-offs are in my implementation guide for law firms.

Most firms end up with both: a platform for everyday drafting and research, and a custom system for the one workflow that is specialised enough to be worth it.

Frequently Asked Questions

Next Steps

If you are shortlisting tools or deciding whether one workflow justifies a build, I offer a 45-minute call at no charge. I'm an independent AI consultant and fractional CTO in the UK, and I'll tell you if buying is the better answer. Book a call.