JSH LAW / LEGAL AI PRODUCT RESEARCH
England and Wales perspective · Sources reviewed 23 October 2026 · Documentary assessment
Legal AI products solve different problems. This hub helps UK readers compare the actual work, the sources and the safeguards—before spending money, uploading a sensitive file or relying on a persuasive answer.
Who this library is for
The primary readers are legal professionals, in-house teams, support practitioners and informed court users trying to make a sensible technology decision. The same product will not be proportionate for all of them. An enterprise contract system may suit a corporate portfolio and have little immediate value to an unrepresented parent.
The reviews distinguish product procurement from broader commentary about the legal AI market and from practical guidance for litigants in person. Where the immediate problem is understanding an order or preparing a short document, start with our guide to legal technology worth paying for.
This is a permanent research directory. Product names, features, prices and terms change; each review records its research date and the questions that remain dependent on the buyer’s licence and deployment.
Compare the fifteen products by their main job
The categories below are editorial navigation aids, not exclusive technical classifications. Several products span more than one workflow. “UK proposition” means relevant supplier documentation exists; it is not a certification of legal accuracy or comprehensive coverage.
| Product and review | Main job to investigate | Likely audience | England and Wales question | Important limitation |
|---|---|---|---|---|
| Harvey | Research, document collections and workflows | Law firms and in-house teams | Which domestic sources and modules are licensed? | Broad capability does not establish complete factual reconciliation |
| Legora | Collaborative review, research and drafting | Teams working across documents | Which sources and review controls apply? | A populated review cell is not human approval |
| Microsoft Copilot for Legal | Word drafting and playbook review | Commercial Microsoft 365 users | How are jurisdiction and legal sources controlled? | Not jurisdiction-aware by default, according to Microsoft |
| CoCounsel Legal | Publisher-backed research and document work | Professional legal teams | Which UK release and content entitlements are included? | Later US features cannot be presumed available locally |
| Lexis+ with Protégé | Legal research, drafting and Vault analysis | Professional legal teams | Which mode, sources and storage rules apply? | General AI, legal sources and internal documents have different authority |
| Spellbook | Contract drafting, redlines and comparison | Contract lawyers and in-house teams | Are suggested terms appropriate for the governing law? | Contract assistance is not comprehensive domestic research |
| Luminance | Contract intelligence and negotiation knowledge | Enterprise legal and business teams | How are current policy and historic concessions separated? | Institutional memory can preserve the wrong precedent |
| Vincent by Clio | Source-based and cross-border research | Firms and legal departments | Which domestic collections and jurisdiction settings apply? | Global coverage does not establish every local entitlement |
| Clio Manage AI | Calendar, billing and matter administration | Practices using Clio Manage | How are orders, dates and approvals verified? | Extracting a date is not necessarily calculating a legal deadline |
| Everlaw | Disclosure and analysis of substantial evidence collections | Disputes and investigation teams | What collection and procedural framework governs the task? | Source-linked answers can still omit material evidence |
| Paxton | Litigation records, research and drafting | Principally US-facing legal teams | Is authoritative domestic coverage demonstrated? | Published research library is US-focused |
| Darrow | Case discovery and litigation-risk portfolios | Specialist claimant, compliance and insurance teams | Does the legal route and data support a UK use? | A risk signal or valuation is not proof or a guaranteed recovery |
| Wordsmith | In-house intake, contracts and legal workflows | General counsel and legal operations | Which sources, permissions and escalation rules apply? | Convenient self-service can obscure an approval boundary |
| ChatGPT | Flexible organisation, analysis and drafting | Individuals and organisations, depending on plan | Which workspace, sources and controls are being used? | General-purpose assistance is not authoritative legal advice |
| Claude | Document synthesis, analysis and drafting | Individuals and organisations, depending on plan | Which processing path, source pack and data terms apply? | A large upload does not establish complete visual or factual review |
Legal research and professional knowledge workflows
Harvey AI
Primary functions: professional assistance, substantial document collections and reusable workflows. Audience: teams with repeated legal analysis and drafting work. Jurisdiction: confirm the actual domestic source collection, rather than infer coverage from international positioning. The review examines the important distinction between Assistant, Vault and structured review, including documented limitations and the cost of implementation.
Legora
Primary functions: shared document review, research and drafting. Audience: law firms and in-house teams needing a visible review process across collections. Jurisdiction: source entitlements and the applicable law remain separate checks. The review focuses on table-based extraction, collaboration, reviewer responsibility and the need to preserve source references outside the platform.
CoCounsel Legal
Primary functions: research and related document workflows connected with Westlaw and Practical Law. Audience: professional legal teams. Jurisdiction: a UK launch is documented, but regional releases and package scope must be checked. The review distinguishes the confirmed UK proposition from later global announcements and examines citation support, matter context and licence cost.
Read the CoCounsel Legal UK review.
Lexis+ with Protégé
Primary functions: legal-source research, drafting, general AI and stored-document analysis. Audience: professional teams combining research and document work. Jurisdiction: inspect the UK source entitlement and selected mode. The review explains why Legal AI, General AI and Vault content require different evidential and retention checks.
Read the Lexis+ with Protégé review.
Vincent by Clio
Primary functions: source-based research, document workflows and reusable methods. Audience: firms and legal departments, including cross-border teams. Jurisdiction: vLex publishes UK coverage, but the purchased collection and jurisdiction selection still matter. The review examines the Clio acquisition, Skills, Memory and the risk of carrying stale assumptions into a new task.
Read the Vincent by Clio review.
Contract drafting and contract intelligence
Microsoft Copilot for Legal
Primary functions: document analysis, drafting, tracked edits and playbook review within Microsoft’s commercial environment. Audience: teams already working in Microsoft 365. Jurisdiction: Microsoft expressly documents default jurisdiction limitations. The review considers the playbook, licence, tenant controls and the difference between a neat redline and a sound legal decision.
Read the Microsoft Copilot for Legal buying review. Our separate market analysis addresses influence and access to justice.
Spellbook
Primary functions: drafting, review, comparison and playbooks, with distinct editor and multi-document workflows. Audience: professional contract teams. Jurisdiction: test the governing law and commercial assumptions rather than rely on international availability. The review separates Word, Google Docs and Associate, and examines version control, source selection and customised pricing.
Luminance
Primary functions: enterprise contract analysis, negotiation and institutional knowledge. Audience: organisations managing recurring work across a portfolio. Jurisdiction: contract familiarity does not replace current legal analysis. The review examines how historic concessions, approved policy and automated action should be controlled, and when the implementation burden is unlikely to be proportionate.
Litigation, disclosure and case discovery
Everlaw
Primary functions: substantial electronic evidence collections, review, analysis and litigation work products. Audience: disputes and investigation teams. Jurisdiction: UK product availability does not determine the applicable disclosure regime or court directions. The review starts with preservation and collection, then examines Deep Dive, retrieval completeness, permissions, production and cost.
Paxton AI
Primary functions: a connected litigation record, legal research and drafting. Audience: principally US-facing teams on the published evidence. Jurisdiction: the reviewed source library emphasises US federal and state law; comprehensive domestic coverage was not established. The review includes the public dollar pricing and explains why a UK coverage gap is a purchasing obstacle, not a minor caveat.
Darrow AI
Primary functions: detecting possible legal exposure, evaluating opportunities and managing litigation portfolios. Audience: specialist claimant firms, compliance and insurance users. Jurisdiction: a domestic legal route and data foundation need separate proof. The review distinguishes a signal, an allegation, a viable claim and a valuation, drawing on official material and independent reporting.
Practice administration and in-house legal operations
Clio Manage AI
Primary functions: matter administration, proposed calendar actions, draft billing and client updates. Audience: practices working in Clio Manage. Jurisdiction: the operative order and procedural basis must control a deadline. The review explains the move from Duo, the published UK starting price and the approval gates needed before an administrative proposal becomes an external action.
Read the Clio Manage AI review.
Wordsmith
Primary functions: legal intake, contracts, research and connected business workflows. Audience: general counsel and in-house legal teams. Jurisdiction: verify the actual domestic sources and approved company position. The review examines Slack and Word integration, escalation, permissions and why model-provider zero retention is not a complete account of platform storage.
General-purpose AI assistants
ChatGPT
Primary functions: adaptable organisation, analysis, explanation and drafting, depending on plan and tools. Audience: individuals and organisations with different governance needs. Jurisdiction: legal propositions still require authoritative checking. The review compares personal and business contexts, file storage, pricing and action boundaries. It is distinct from our ChatGPT Work preparation guide.
Read the ChatGPT legal-work review.
Claude
Primary functions: document synthesis, drafting and analysis, with plan-specific tools. Audience: individuals and organisations. Jurisdiction: a large context window does not supply authoritative domestic law. The review examines PDF processing, exact quotations, persistent context, commercial retention and the distinction between Enterprise seat fees and usage charges.
Read the Claude legal-work review.
How we assess products and handle evidence
Each review starts with a real workflow and an intended reader. We then examine current official product material, help documentation, commercial information and security or privacy disclosures. Independent reporting and research are used where relevant and identified separately.
We distinguish five kinds of statement:
| Evidence type | How to read it |
|---|---|
| Documented function | The supplier describes a capability; availability may depend on licence and configuration |
| Vendor performance or security claim | A proposition requiring supporting evidence, not an independent JSH Law finding |
| Independent report or study | Evidence limited to its method, sample, date and product version |
| Hypothetical trial | A proposed evaluation using invented material; no result is implied |
| Editorial assessment | A reasoned judgment about potential fit, limits and what remains to be established |
We do not claim to have bought, operated, benchmarked or security-audited all fifteen products. We do not publish fabricated demonstrations, customer results or comparative accuracy scores. The original informal survey that prompted the collection is not used as UK market share or verified industry adoption data.
Where information is unavailable, the review identifies the gap. An inaccessible source inventory, private contract or unexamined audit report should not be replaced with a confident assumption. A qualified documentary assessment can still be useful if the missing evidence is clear.
England and Wales: test the source before trusting the style
“UK law” can conceal differences between England and Wales, Scotland and Northern Ireland. International products may also foreground US federal or state law. Specify the actual jurisdiction, date, practice area and procedural setting.
For a legal proposition, inspect the authority itself. Establish that it exists, is current, supports the proposition and applies to the problem. A correct quotation from the wrong procedural context can still mislead. A citation link is an aid to review, not the conclusion of review.
Academic research, including Hallucination-Free?, demonstrates why source-grounded legal AI needs scrutiny. That study examined earlier US systems; it cannot be repurposed as an error rate for the current UK products in this directory. Ask for evidence matched to the current task and configuration.
Professional duties, data protection and confidentiality
The SRA’s 17 August 2026 warning addresses regulated firms and individuals, including inaccurate work and confidentiality risks. It does not make a product’s outputs self-verifying. JSH Law is not an SRA-authorised solicitors’ firm; we refer to the warning as relevant professional context, not as a claim of our own firm authorisation.
The ICO’s AI guidance and lawfulness chapter require attention to the purpose and basis of processing. Sensitive and criminal-offence information raise additional requirements. The ICO flags that relevant guidance is under review following the Data (Use and Access) Act; do not treat an older summary as a complete statement of every current deployment issue.
Before live use, the organisation should establish:
- Authority and purpose: what information may be processed, why, and under which legal and contractual basis.
- Data flow: the platform, model providers, subprocessors, support access, connected systems and processing locations.
- Controls: permissions, authentication, retention, deletion, incident response and any necessary impact assessment.
- Human responsibility: who verifies the output, approves action and retains the record of the decision.
- Exit: how source files, work product and review history remain usable when the service ends.
No-training, encryption, model-provider zero retention and platform deletion are different controls. None should be expanded into an invented guarantee of privilege or automatic UK GDPR compliance. Obtain the actual terms and appropriate specialist advice where necessary.
AI in family justice and evidence preparation
Technology can help people organise a difficult record, identify missing information and prepare focused questions. It must not turn an allegation into a finding, infer a child’s wishes from a pattern of messages or present a generated narrative as authenticated evidence.
Private children proceedings have rules and orders governing communication of information. FPR Part 12 and Practice Direction 12G permit some communications for specified recipients and purposes, subject to the relevant framework. That is not blanket authority to upload the papers to any AI service. Check the particular material, recipient, purpose and order before transmission.
Use wholly synthetic cases for experiments. Changing names in a real case may leave a family identifiable and does not necessarily resolve legal restrictions. Preserve original digital files, timestamps and context; keep extraction, inference and legal analysis separate.
For litigants in person, the proportionate answer may be a document register, chronology and targeted support rather than a new enterprise subscription. Our digital-evidence guide and prompt-governance article explain those practical boundaries.
A controlled evaluation before spending or deploying
Choose one task and prepare a human-checked answer ledger using invented material. Include a material contradiction, missing date, adverse source or escalation issue. State the unacceptable failures before running the trial.
Measure correctness, omissions, source fidelity, permissions, correction effort and the final usable output. Test an export and an account that should not have access. Introduce a changed document or rule and check whether the result updates for the right reason.
Then compare total cost: licence, source subscriptions, implementation, training, migration, usage, supervision and exit. A first draft produced quickly is only one part of that calculation. Keep pricing in its original currency and billing context; a US public tariff is not a UK quotation.
Our six practical evaluation questions are: what task; which sources; what data; which permissions; who checks; and what completed result? This is an editorial decision framework, not a legal test or product accreditation.
Related reading and further research
- Which legal technology is worth paying for as a litigant in person?
- Microsoft Copilot for Legal and the future of legal AI
- ChatGPT Work and safe preparation for family proceedings
- Before you press submit: AI prompt governance
- AI evidence, deepfakes and digital provenance
Future research should add recorded trials with synthetic inputs, preserved prompts, raw outputs and correction logs. A proposed test should become a claimed result only after it has actually been run and reviewed.
Review dates, corrections and practical support
The initial collection’s documentary research date is 23 October 2026. Recheck promptly after changes to ownership, product names, regional availability, sources, pricing, data terms or action permissions. A change to the page’s appearance alone is not a substantive product review.
The next editorial check is targeted for November 2026. That is a maintenance recommendation, not a claim that an automated monitoring task has been created. Corrections should identify the affected product, proposition and source so the existing review can be updated without creating a competing page.
If your immediate need is organising family-court evidence, clarifying a document task or preparing for a hearing, JSH Law’s services may be relevant. Book a Consultation with a short description of the problem and next deadline. Do not send confidential papers through a public form before an appropriate route and scope have been agreed.
Sources and scope of the hub
The individual reviews contain the product-specific source links, pricing qualifications and unresolved procurement questions. Common sources are the SRA warning, ICO guidance, Family Procedure Rules, PD12G and the academic study linked above. No unsupported star ranking, industry-adoption percentage, security certification or guaranteed saving is attributed to JSH Law’s work.
The collection provides general product information and independent editorial judgment. It is not case-specific legal advice, a supplier recommendation for every reader, or a substitute for an organisation’s legal, security and procurement assessment.