Lexis+ with Protégé UK Review: Legal AI, General AI and Vault

JSH LAW / LEGAL AI PRODUCT RESEARCH

England and Wales perspective · Sources reviewed 23 October 2026 · Documentary assessment

The name change and the current product

The UK product page records the change from Lexis+ AI to Lexis+ with Protégé in the first quarter of 2026. Older articles, demonstrations and studies may therefore describe an earlier version under the previous name.

The supplier is LexisNexis, part of the RELX group. The product is built around its professional legal-content and workflow services, rather than being a standalone public chatbot. Confirm which sources, modules and integration rights are included in the proposed subscription: membership of the same product family does not make every resource available automatically.

A useful buying brief identifies the actual bottleneck. Is the team trying to find relevant authorities, produce first drafts, interrogate a collection or reuse internal knowledge? Those needs overlap, but they require different trial questions and different measures of success.

Legal AI and General AI need distinct review rules

LexisNexis’ General AI explanation distinguishes broader reasoning and web context from the legal-source workflow. It describes moving from general drafting to legal validation within the platform.

This is a practical source hierarchy. It is not a claim that one mode never makes mistakes or another is unusable. The point is to identify the source of each material proposition before it becomes advice, a contract term or a court submission.

Give reviewers a visible way to record the mode and source set used. A legal conclusion should not become untraceable when it is copied from a mixed-purpose conversation into a final document.

What the agentic workflow changes

The agentic AI material describes planning, research, drafting and iterative checking. These are vendor descriptions of the workflow, not evidence that the final answer is always correct.

The potential benefit is that a complex task can be broken into connected stages. The supervision problem is that an early mistake may then pass through several stages and become a polished final draft. Ask to inspect the research basis and unresolved assumptions rather than relying on a statement that the system checked itself.

An effective professional review should be able to stop the process, narrow the question and replace an incorrect assumption. It should also preserve the distinction between a system-generated plan and the lawyer’s approved plan for the matter.

The Vault and the retention distinction buyers should not miss

On the review date, the current UK page described up to fifty Vaults with 1,500 documents per Vault. It distinguished persistent Vault storage from small, unsaved uploads, and described a ninety-day conversation-history period. These are documentation figures, not a contractual capacity guarantee for every account.

That distinction matters more than the headline number. Deleting a conversation, ending a session and deleting a persistent document collection are different actions. A practice needs a retention schedule identifying each data class and the person responsible for closing it down.

We found older indexed product material with a lower document figure. Buyers should use the current account documentation and written contract rather than assuming a cached comparison is still accurate. Also test whether the proposed collection actually processes: scans, embedded images, damaged files and lengthy tables can affect the quality of the usable text.

Two hypothetical UK workflows and one deliberate boundary test

A source-grounded research note

Use an invented England and Wales problem with a known statutory provision, a relevant appellate decision and a qualification that changes the outcome. Ask for the legal position, the competing argument and the fact that would need clarification before advice could be finalised.

The human reviewer should open each authority and record the precise proposition it supports. The test fails if a genuine citation is used to support something it does not say. Repeat with a different procedural setting to see whether the system recognises that the broad principle alone is insufficient.

A Vault containing conflicting documents

Create synthetic emails, a draft agreement, a signed agreement and a later amendment. Ask for a table of operative obligations, recording the document version and the reason for treating it as current. Include an obligation that appears in the draft but was removed before signature.

The useful output would make the version decision visible. The unacceptable shortcut is a merged summary that combines mutually inconsistent drafts as if they were one agreement. Nothing in the appearance of a persistent repository establishes that the collection has been legally reconciled.

Moving from a general explanation to a legal conclusion

Ask General AI to produce a plain-English board briefing from fictional business facts. Then identify each legal proposition requiring checking through the legal-source workflow. Test whether the approved final note preserves the verified qualification when it is shortened.

This exercise is about a handoff between modes. It is not a claim that either mode has completed the trial, nor a suggestion that the general model should determine legal merits without source verification.

UK suitability and the limits of global branding

A UK-targeted professional offering is a stronger starting point for domestic research than a library whose published coverage is confined to US law. It still needs a source-level assessment for the practice area, jurisdiction and date involved.

Ask for a demonstration in the areas the team actually handles. For family work, distinguish private children proceedings, financial remedies and public-law processes rather than treating “family law” as one undifferentiated subject. Check access to the relevant procedural material and the ability to inspect it outside an AI summary.

The product page itself includes international examples and terminology. That does not invalidate the UK offering; it is a reminder to check the selected jurisdiction and source set rather than infer them from the page’s national domain or the tone of the output.

Information governance: resolve the service-specific position

LexisNexis’ public materials describe private models, encryption, access controls and restrictions on training with customer data. Its agentic page makes a UK-hosting statement, while the broader product page describes European primary and backup locations. These descriptions should be mapped to the actual module and customer agreement.

Do not combine different statements into an invented promise that every component is always processed only in the UK. Obtain a written data-flow description covering models, storage, backups, support access and document-management connectors. Ask which terms apply to General AI, Vault and integrated firm knowledge.

Permissions must survive the integration. Test whether a user excluded from a matter can retrieve its content indirectly through a general question, shared prompt or saved output. Personalisation also deserves scrutiny where an earlier matter could influence a later answer.

For children’s information and private court papers, assess authority, necessity and restrictions before live uploading. A supplier’s security design does not itself determine privilege or the court’s communication rules. Start with a synthetic file and the hub’s governance questions.

Pricing and the cost of overlapping subscriptions

The UK product page describes customised pricing based on organisation size, capabilities and content access. We have not obtained a quotation. Do not invent a flat monthly price or assume an existing research licence includes every AI workflow.

Ask for the cost of the actual source collection, users, integrations, onboarding and exit. If General AI access is part of the purchase rationale, identify which other subscriptions could actually be removed and which still serve a separate purpose. Multiple model access is not a saving if the organisation keeps paying for all its previous tools.

The value measure is the completed, checked task. Record source-checking time, corrections, quality of the final advice or draft and the effort required to maintain the internal knowledge. A broad platform can be worthwhile while a particular module remains unnecessary.

What a fair accuracy assessment would require

Use a preregistered set of questions with known answers, not only examples selected after the product succeeds. Include missing information, a misleading premise, an authority from another jurisdiction and a document that should be excluded by permissions.

Record both false statements and material omissions. A refusal to answer outside the evidence is not automatically a failure; an answer which conceals uncertainty may be. Keep the model configuration, sources and human corrections so the trial can be repeated after a meaningful product change.

Earlier academic work on legal AI provides a reason for scrutiny, but cannot be converted into a current Protégé UK error rate. This review supplies no numerical accuracy ranking and makes no claim to have run the exercises above.

Alternatives and JSH Law’s assessment

CoCounsel Legal is the natural comparison for publisher-backed UK research. Vincent by Clio warrants evaluation for its source collection and cross-border needs. Claude may support document synthesis, but should not be assumed to supply the same legal-source service.

Our assessment: worth a controlled professional pilot where legal research and document work need to be joined up. The strongest implementation will keep Legal AI, General AI and matter documents visibly distinct while allowing a deliberate handoff between them. Confirm retention, hosting and source entitlements before treating the platform as the practice’s default workspace.

Does General AI inside Lexis become authoritative law?

No. The environment and its data controls do not determine whether a particular legal proposition is supported by the right authority.

Is Vault retention the same as conversation retention?

No. The supplier describes different storage classes. Establish the deletion and retention position for each class in the purchased service.

Sources and methodology

Checked on 23 October 2026: the UK product page, agentic workflow explanation and General AI documentation. We have not inspected a customer tenant, audit report or negotiated contract. Account-specific hosting, limits and permissions require confirmation. Review again after a material source, retention, model or licensing change. The research hub explains the wider methodology.

If your need concerns the organisation and clarity of family-court papers rather than buying research software, see JSH Law’s document support and Book a Consultation to identify a defined next task.