P

PatentSight+

The models are the engine of named capabilities layered on a platform that functions without them, which is the B band. Four AI capabilities are published and each is distinct: Protege, a generative assistant taking plain-language business questions to structured visual answers, in general availability since May 2026; AI-Powered Features supplying chart-level explanations and faster analysis; an AI Classifier building custom technology views; and TechDiscovery, added October 2024, allowing patent search from ordinary words and phrases.

Underneath sits a platform with an independent existence and a longer history: a harmonised database of more than 90 million patent family records, manually checked ownership normalisation, legal-status tracking, more than 100 attributes and measures including the Patent Asset Index, full-text and field search with Boolean, proximity and wildcard syntax, a similarity search, saved search history, and the visualisation tools the product was originally known for.

Strip out every model and the analytics platform remains complete. What is worth recording is the direction of travel: the vendor positions Protege as changing who can use the platform rather than what it does, describing it as replacing complicated filters with simple questions so business leaders and colleagues outside IP can reach the same data. That is an access change on a stable analytical core, which is the B band exactly. Verified 13 September 2026.

Grounding is real, documented and unusually strong, short of a measured accuracy figure, which is the B band. The grounding mechanism goes beyond citation to reproducibility, and that distinction is the finding on this row. Every Protege answer is accompanied by the full search query the assistant constructed, stated to be fully reproducible within PatentSight+, so a user can rerun the analysis and obtain the same result rather than merely being pointed at a source.

Around that, the assistant explains each step of the analysis, displays the queries, contextualises results and suggests next steps, with the vendor describing full transparency into how insights are generated so they can be understood and validated. Answers are grounded in the curated harmonised dataset and in published metrics including the Patent Asset Index methodology, so the analytical logic is inspectable as well as the data.

What is absent is measurement of the assistant's own accuracy. No precision, recall or error rate is published for Protege, no test set is described, and no failure mode is named. The figures the vendor does publish measure effort rather than correctness, and are hedged and user-attributed: users reported reductions in manual analysis effort of up to 70 to 90 per cent and up to three times more output. Recorded and not credited: an on-demand session is advertised on independently validated accuracy for AI-driven patent classification, but no result is published, and under R25 it is named as what would move this row. Verified 13 September 2026.

A written commitment that the model works alongside the user, with real and specific review surfaces, short of the full control structure, which is the B band. The commitment is stated in the vendor's own idiom rather than as a disclaimer: Protege is said to keep professionals in the driver's seat, and human oversight is one of the five RELX Responsible AI limbs the product page states govern all AI in the platform. The review surfaces behind it are concrete and are the strongest part of this row.

The assistant explains every step of its analysis, displays the full queries it constructed, contextualises the results and suggests next steps, and each answer carries the query so the user can reproduce and inspect it. The support documentation includes guidance on prompting and on validating AI-generated insights, which is instruction on how to supervise the tool rather than a statement that supervision happens. Vendor material quotes a named customer specifically on this point, valuing seeing how the assistant reasons through a problem.

What the A band requires is not published. No threshold is stated at which any capability acts without review, nothing distinguishes modes with different levels of automation, no confidence signal attaches to an individual answer, and nothing describes what happens after an output is found wrong or how a user reports it. Nor is anything published about the AI Classifier or the enrichment models, whose outputs enter the dataset rather than arriving as answers a user inspects. Verified 13 September 2026.

Named customers with attributed substance, short of joining the names to measured figures, which is the B band. Two customers are named and both say something specific rather than offering praise. Siemens describes using the platform to track patent quality over time against competitors, accounting for acquisitions and divestitures, and notes that the quality indices reflect differences between the United States, German and Chinese markets, which is a substantive account of what the metrics do.

Christopher Hauke, Head of Strategic IP and Innovation at Schott Pharma AG and Co. KGaA, is quoted by name and role on the value of seeing how Protege reasons through a problem, saying it makes the analysis transparent and gives him confidence in the results. Beyond individual customers, the AI Insider Program is described as having developed Protege in collaboration with hundreds of participating innovator organisations, with early-access feedback shaping its agentic reasoning, which evidences a structured user base rather than a handful of references.

The figures are the weakness and the note states why they are not graded higher. Reductions in manual analysis effort of up to 70 to 90 per cent and up to three times more output are attributed to users rather than measured by the vendor, are expressed as ranges with an upper bound, and carry no method, baseline or sample. They are recorded as reported and are not treated as measurement. No figure is joined to either named customer. Verified 13 September 2026.

Substantive published commitments across most limbs, defeated on two, which is the B band. The commitments are specific and several are unusual. Training is excluded in express terms on the product page: queries and usage data are never used to train AI models, repeated as a statement that customer data is never used to train AI models. Data is stated not to be shared outside the customer's organisation. And a cross-product commitment is published that this corpus has not seen elsewhere: customer data entered into Protege is not shared with other LexisNexis products unless explicitly communicated and authorised, which matters because the vendor operates seven adjacent IP products a buyer might otherwise assume share a data layer.

Retention is addressed both as policy and as a control, the vendor citing robust data retention and deletion policies while the platform lets a user disable automatic saving of search queries and compare, combine, export or delete saved queries. Personal data is processed under a published Privacy Policy and Data Processing Addendum, and the General Terms provide for return of Subscriber Files on request within sixty days of termination.

Two A limbs fail. Privilege and work product are not addressed by name, which under R33 forecloses A on its own. And no model provider is identified, so nothing states what any third party sees of a query, which on a tool that ingests strategic questions about unannounced acquisitions and licensing positions is the gap that matters. Verified 13 September 2026.

A contractual scope limit sits in the terms while the marketing describes the product in decision terms and widens the audience, which is the C band. The contractual limit is real: the General Terms grant a non-exclusive, non-transferable, limited right to access and use the online services and materials for research purposes, which frames the product as research rather than advice, and the terms address legal professionals separately.

Against that, the marketing is written in the language of determinations rather than of information: decision-ready insights, high-value strategic decisions, and worked examples that are squarely professional questions, including identifying top acquisition targets in a sector and finding potential licensing partners for a technology. The audience is deliberately widened rather than left ambiguous, which is what places this in the C band rather than above it.

The vendor states the point openly, positioning Protege as expanding access to patent data to key players in other parts of the organisation and as letting business leaders reach insights without complex filters. So non-lawyers are an intended audience for output that feeds licensing, acquisition and portfolio decisions. Nothing published addresses what a business user should not do with a Protege answer, or where a patent attorney's judgement must intervene.

Recorded and not credited because it engages the audience rather than the line: the Responsible AI page observes that many users of IP analytics are lawyers and reasons from that, which is graded on the bar guidance signal. Verified 13 September 2026.

A published governance framework, externally defined and connected to this product by scope, short of testing results and a named owner, which is the B band and the strongest governance position in this lane. The framework is the RELX Responsible AI Principles, and what distinguishes this record from its neighbours is that the connector required by R16 is present and explicit rather than assumed: the product's own AI page states that all AI in PatentSight+ is governed by those principles, and the TechDiscovery release states the module was developed in accordance with them.

The five limbs are enumerated on the product surface rather than referenced by title, covering real-world impact, minimising bias, transparency in how insights are generated, human oversight, and governance with strong privacy and security practices. A Responsible AI page sits on the IP estate itself and argues the case for self-regulation rather than merely asserting compliance. Structured feedback is real: the AI Insider Program gives participating organisations early access, focus groups and feedback channels, which is a mechanism for surfacing problems before general release.

What the A band asks for is missing. No individual, committee or function is named as accountable for model behaviour on this product. No pre-release testing regime is described. And although bias minimisation is named as a principle and an on-demand session advertises independently validated classification accuracy, no evaluation result is published, so a buyer can read the commitment and not the evidence. Verified 13 September 2026.

Substantive published policy covering most of the ground, short of the full set, which is the B band, and the coverage here is broader than most records reach. Training exclusion is express and repeated. Encryption is stated for data at rest and in transit, with privacy-by-design named as a development practice. Audit practice is specific rather than asserted: LexisNexis cloud environments, products and security programmes are stated to be audited annually, with the audit scope enumerated as encryption, backup, disaster recovery, access controls, data destruction, breach prevention and confidentiality safeguards.

Incident practice is published and tiered, with policies and procedures covering technical, administrative, business and executive escalation. Subprocessors are handled properly and this is the strongest limb: the Data Processing Addendum incorporated into the terms points to a maintained public subprocessor list, commits to updating it at least fourteen days before any change takes effect, and gives the customer a right to object within fourteen days with reasons.

Retention and deletion policies are stated, and the platform gives the user direct control over saved search queries. Two things hold it off A. No retention period is stated anywhere, only that policies exist. And access control at product level is thin: single sign-on across LexisNexis applications is described as being introduced rather than available, and no role or permission model is published. Verified 13 September 2026.

A published agreement addresses service quality and termination while nothing was located on the allocation of risk for AI output, which lands in the C band. What is established is real and favours the customer in two places. The General Terms give the subscriber a remedy for degradation rather than only a disclaimer: where a change materially degrades the service or has a material adverse effect, LexisNexis has thirty days to cure the condition, failing which a termination right arises, with changes made for regulatory or compliance reasons carved out.

And data is returnable, the terms providing that on request made before or within sixty days after termination the subscriber will be given a file of its Subscriber Files in a mutually agreed format. Termination mechanics are published on both sides, including ten days' notice for convenience by LexisNexis and immediate suspension for breach. Personal data obligations are addressed through an incorporated Data Processing Addendum.

What is not established keeps this at C and the limit is recorded honestly rather than treated as an absence: the warranty and limitation of liability clauses of the General Terms were not recovered through the R8 ladder, so nothing is asserted about their contents in either direction. Separately, and this is a finding rather than a retrieval gap, nothing AI-specific was located anywhere in the published terms. No provision addresses the accuracy of an AI-generated insight, no indemnity reaches a decision taken on one, and no insurance position is published. Verified 13 September 2026.

Integration is claimed and partly named, without a documented catalogue or configuration detail, which is the C band. What is established sits mostly inside the vendor's own family. Single sign-on across LexisNexis applications is described as an enhanced login experience allowing users to move between LexisNexis products without re-authenticating, which is a real interoperability step, though the vendor describes it as being released over a period rather than as available, so under ground rules section 2 it is recorded and not credited as shipped.

Data can leave the platform: search queries and results can be exported, and a separate product, IP DataDirect, exists as a data feed, with an older factsheet referencing a developers portal and proof-of-concept access. TechDiscovery and the AI Classifier operate as modules within the same platform rather than as integrations. What the higher bands require was not established. No connector catalogue is published, no external system is named as supported, no API documentation for this product was reached, and nothing describes the direction of flow or the configuration required for any connection.

The gap has practical weight for the buyer: a corporate IP department typically runs a docketing or IP management system alongside its analytics, and nothing published states whether portfolio data can move between them. Verified 13 September 2026.

Cloud delivery is evident and neither the tenancy model nor the region is stated, which under R38 is the C band with both co-equal limbs absent. The delivery model is clear: a browser-based subscription platform accessed through LexisNexis online services, governed by General Terms that define the arrangement as access to online services rather than as licensed installed software. Hosting is described only by character, the vendor stating that it partners with trusted, enterprise-grade cloud providers and continuously monitors and improves its infrastructure.

That is a statement about the class of provider rather than an identification of one, and it carries no location. Nothing published names a data centre region, a country of processing, or an option to elect one, which is a real gap for a platform sold across Europe and Asia to customers with data residency obligations, and it sits oddly beside the detailed data protection material the same estate publishes. Tenancy is equally unaddressed: nothing states whether a customer's saved searches, exports and Protege conversation history sit in a shared or isolated environment, and no single-tenant or private deployment option is offered or refused.

Recorded and expressly not credited under ground rules section 3: the annual audit scope includes backup and disaster recovery, which is resilience practice rather than a residency statement, and the cloud providers' own certifications are infrastructure. Verified 13 September 2026.

Certification is referred to without a named standard held for this product, and the one standard that is named is promised rather than held, which is the C band. The critical sentence is in the product support documentation and is recorded verbatim in effect: Protege in PatentSight+ follows the same enterprise security framework used across LexisNexis products and will be ISO 27001 certified. That is future tense, and under ground rules section 2 future tense is not evidence of a shipped capability, so it is expressly not credited.

Read plainly it says the opposite of what a skimming buyer would take from it: as of this verification the product is not certified on its own account. What else exists is real but unnamed. The vendor states that it undergoes regular third-party audits to maintain industry-leading certifications without identifying any of them, and that LexisNexis cloud environments, products and security programmes are audited annually, with the audit scope enumerated in useful detail across encryption, backup, disaster recovery, access controls, data destruction, breach prevention and confidentiality safeguards.

So audit practice is described and its coverage is specific, while the resulting attestations are not named, dated, scoped or obtainable. No trust centre carrying reports was located for this product. The distance to a higher grade is small and entirely within the vendor's control: naming the certification actually held, its scope and its period would move this row immediately. Verified 13 September 2026.

The vendor refers to purpose-built AI without identifying what sits underneath, which is the C band. The references are confident and comparative but never specific. Protege is described as purpose-built for patent intelligence, as delivering insights beyond what general large language models and AI-powered IP workflow tools can provide, and as having agentic reasoning refined through early-access feedback. TechDiscovery, the AI Classifier and the enrichment models are each described by function.

Not one of them is attributed to a model, a version, an architecture or a provider. The comparison against general large language models implies the assistant is something other than a wrapper without saying what it is, and nothing distinguishes models built in-house from models licensed and fine-tuned. Nothing states where inference runs, what any provider retains of a query, or whether a customer would be told if the underlying model changed.

The contrast with the same vendor's data disclosure is the point worth recording: the corpus behind the platform is described in exhaustive detail, down to 90 million patent families, ownership normalisation and more than 100 named attributes, while the models reading it are described only by purpose. Recorded and expressly not credited under ground rules section 3: the reference to trusted, enterprise-grade cloud providers names a class of infrastructure supplier, not a model supplier, and no provider is identified in any case. Verified 13 September 2026.

No pricing information is published at any level, including the unit of charge, which is the D band. The page inventory was taken under R20 across the product estate, the AI pages, the Responsible AI page, the resources and news sections, the support and product support centres and the LexisNexis legal and terms estate. There is no pricing page for this product and no figure, band or unit appears anywhere on the vendor's own surfaces.

The commercial routes published are a demonstration request and a contact form. What the terms confirm is that pricing exists as a negotiated contractual document rather than a public one: the Subscription Agreement is defined as the General Terms plus Supplemental Terms plus the applicable rates set forth in the Price Schedule, and the terms distinguish transactional pricing plans from fixed-price and fixed-term arrangements without publishing either.

So a buyer can establish that two different charging shapes exist and nothing about what either costs, which is not published pricing structure for this product within the meaning of R10. Under R10's closing discipline no structure means no row, so no VendorPricing row is written. Recorded and expressly not credited under R41: third-party listings describe customisable pricing available by quote and subscription-based term pricing, which corroborates the absence rather than establishing any structure. Verified 13 September 2026.

Coverage is described with real substance across data, buyers and use cases, with the boundaries left open, which is the B band. Data coverage is quantified: more than 90 million patent family records, harmonised and enriched with manually checked ownership normalisation, legal-status tracking and more than 100 attributes and measures. Buyer coverage is named across an unusually wide range, the vendor addressing corporations, law firms, regulators and academic users, and stating expressly that the assistant is intended to widen access from IP specialists to business leaders and colleagues elsewhere in the organisation.

Use cases are enumerated concretely rather than abstractly, the vendor publishing worked example questions covering acquisition targets in a named sector, the evolution of a named technology landscape, licensing partner identification and portfolio fee reduction, alongside benchmarking, due diligence, competitive analysis, litigation risk and non-practising entity identification. Jurisdictional reach is evidenced through a named customer describing quality indices that account for differences between the United States, German and Chinese markets.

What is left open is the limit in every direction. No technology area is named as better or worse covered, no jurisdiction is stated as out of scope, nothing describes what the AI Classifier handles poorly, and nothing distinguishes what this platform covers from the adjacent LexisNexis IP products a buyer might otherwise assume are included. Verified 13 September 2026.