D
Derwent Patent Search

Derwent Patent Search, sold until recently as Derwent Innovation, is Clarivate's patent research platform for prior art, patentability, freedom-to-operate and invalidity work. Its distinguishing asset is the Derwent World Patents Index, a curated database in which human editors rewrite patent disclosures into standard English: descriptive titles and abstracts for more than 70 million invention families setting out an invention's novelty, use and advantage, produced by a team the company describes as over 800 patent editors, alongside subject-matter classification coding, manual indexing and invention-centric family building that groups documents by invention rather than by filing.

Around that sit enhanced full-text patent data from 76 jurisdictions and bibliographic data from 109, covering more than 178 million patent publications; the Derwent Patent Citation Index for forward, backward, literature and examiner citations at family level; more than 300 searchable fields with classification browsing across CPC, IPC, Japanese FI and F-terms and DWPI's own manual codes; and correlated litigation case data from more than 140 jurisdictions drawn from Darts-ip, a Clarivate company.

Five capabilities are AI-driven. AI Search, launched in December 2024, uses a language transformer model trained on DWPI itself to take free text such as an invention disclosure and return relevant records from more than 160 million patents, aimed at first-pass patentability and freedom-to-operate searching. Data enrichment uses AI models to fill missing bibliographic fields such as assignee. Optimized Assignee with Ultimate Parent resolves current ownership up the corporate tree.

And AI-derived legal status prediction supplies estimated expiration date, remaining life and dead-or-alive status across DWPI and INPADOC families. The company states the platform and its data are used by 40 global patent and trademark offices. Clarivate Plc is a public company listed on the New York Stock Exchange as CLVT, headquartered in London, whose intellectual property portfolio also includes CompuMark for trademarks, Darts-ip, and a separate newer product, Derwent Patent Monitor, launched in November 2025.

Vendor siteLondon, United Kingdom
Last verifiedSeptember 13, 2026
Compare with other vendors

Capability grades

All 15 axes, graded from public sources on the date shown. Hover a grade to see what the letter means on that axis.

BB on AI CentralityThe models are the engine of a core capability, layered on a product that would still function without them as a document or workflow system.

AI Centrality

How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.

The models are the engine of core capabilities layered on a product that would function without them, which is the B band. There is more AI here than the seed's single line suggests, and it is worth listing because the record's other rows depend on it: AI Search, a language transformer model trained on DWPI, launched December 2024 across more than 160 million patent records; AI data enrichment filling missing bibliographic fields such as assignee; Optimized Assignee with Ultimate Parent resolving current ownership; and AI-derived legal status prediction supplying estimated expiration date, remaining life and dead-or-alive status.

What keeps this at B is that the platform underneath is a curated database and a classical search engine, and the vendor sells it that way. AI Search is presented as one search mode among several, alongside keyword and Boolean searching across more than 300 fields, classification browsing, the citation index and Themescape mapping. Remove every model and a buyer still has DWPI's editorially written invention summaries, family building, full text from 76 jurisdictions and litigation data from 140-plus.

The human layer is the product's own headline claim, being over 800 patent editors rewriting disclosures into standard English, and the AI is trained on their output rather than substituting for it. Verified 13 September 2026.

Source: Vendor Published
BB on Citation Accuracy and Hallucination DisclosureGrounding is real and documented, with linked primary sources and a described retrieval method, short of published accuracy figures an outsider can test.

Citation Accuracy and Hallucination Disclosure

Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.

Grounding is real and documented with a described retrieval method, short of any measured accuracy, which is the B band. R15 governs the shape of this row and the note says so, because the usual limbs bite differently: the product retrieves patent records from a closed curated index rather than generating propositions about the law, so a returned result is a real document with an identifier by construction and the fabrication risk this axis was written for does not arise in the same form.

The retrieval method is described rather than asserted. AI Search is stated to use a language transformer model trained on DWPI, to work from free-text input such as an invention disclosure, and to be the only AI patent search trained and powered by that index, with the vendor explaining why that matters: the model learns from editorially written summaries of novelty, use and advantage rather than from raw patent prose.

What is absent is measurement. No recall or precision figure is published for AI Search, no test set is described, no benchmark against keyword or Boolean searching is offered, and nothing states what the model misses. On a tool sold for first-pass patentability and freedom-to-operate work, recall is the metric that matters and none is given. Recorded and expressly not credited as measurement: a named customer's statement that relevant records now appear at the top of the list rather than after reviewing 50 to 200 results is a testimonial, graded on operational evidence. Verified 13 September 2026.

Source: Vendor Published
CC on Autonomy and Oversight ModelAutonomy is claimed and oversight is asserted without a mechanism. Human in the loop appears as a phrase rather than a described control.

Autonomy and Oversight Model

What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.

A mode structure is published and nothing else about oversight is, which lands in the C band. What exists is real but thin. The user chooses between search modes, AI Search sitting alongside keyword, Boolean and advanced modes and classification browsing, so a customer can decide per search whether a model is involved at all, and the vendor describes AI Search as helping to validate results and ensure relevant records are not missed, which positions it as a complement to conventional searching rather than a replacement.

That is a published distinction between how the product can be run, and it is the only oversight material located. Everything the higher bands ask for is absent. No statement describes what a searcher should do with AI Search results before relying on them, no confidence signal or ranking explanation is published, nothing describes how the model behaves when a query falls outside its training, and nothing addresses what happens when a search misses relevant art. R15 is applied and named: a search engine returning ranked records does not act unattended, so the limb asking at what threshold the system acts alone does not bite and is recorded as inapplicable rather than failed.

But the limbs that do bite on a retrieval tool, being what the user is told about the reliability of a ranking and how to check it, are unaddressed, and the stakes are real because a missed prior art reference is the failure mode a patentability opinion turns on. Verified 13 September 2026.

Source: Vendor Published
BB on Operational and Outcome EvidenceReal deployment evidence with substance, short of full attribution or measurement: a named customer without figures, or figures without the named customer.

Operational and Outcome Evidence

Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.

Real deployment evidence with substance, short of the dating and method the A band requires, which is the B band. Two kinds of evidence sit here and both are first-party. Institutional adoption is stated as a figure: the platform and its DWPI data are used by 40 global patent and trademark offices, which for a prior art tool is the strongest form of validation available, since those are the bodies that examine patentability for a living.

And a named individual at a named organisation gives an attributed before-and-after in the launch release: Susan Johnson, patent agent and IP researcher at Boston Scientific, describing other tools as requiring review of up to 50 or even 200 results to find the most relevant records, against AI Search placing the most relevant at the top. That is a named customer, an identified role and a concrete comparison in one artifact, which is more than most records in this lane manage.

Three things hold it off A. Neither claim is dated as a deployment. No method accompanies the comparison, so the 50-to-200 figure describes the witness's experience of other products rather than a measurement of this one. And the 40 patent offices are counted rather than named, so a reader cannot verify which. The customer stories and case study material on the wider Clarivate estate were not reached, the site returning a bot-detection block, and are what would move this row. Verified 13 September 2026.

Source: Vendor Published
CC on Privilege and Confidentiality PostureConfidentiality is asserted in general terms, or the commitment lives only in a sales conversation and cannot be read in advance.

Privilege and Confidentiality Posture

How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.

Confidentiality is asserted in general corporate terms and the question this product actually raises is unaddressed, which is the C band. What is published sits at company level: a privacy policy describing Clarivate as processor for customer-entered data with separate data processing agreements concluded with the controller, personnel bound by confidentiality terms, and service providers contractually held to the same standard.

The published Clarivate Terms treat each party's confidential information as protected and define Clarivate IP as including confidential information and trade secrets. None of that addresses the specific exposure. AI Search works from free-text input, and the input the vendor invites is an invention disclosure, which is unpublished, novelty-destroying if it escapes, and often privileged in the hands of a patent attorney.

Nothing published states whether a search query is stored, for how long, whether it is logged against a user or an organisation, whether it is used to improve the model, or whether it is visible to Clarivate personnel. For a tool whose whole purpose is to be given an unfiled invention before anyone else sees it, that is the central confidentiality question and it is not answered anywhere located. Privilege and work product are not addressed.

No model provider position exists because no provider is named. Retention and deletion are unstated. Recorded as a retrieval limit rather than an absence: the confidentiality section of the current Clarivate Terms was not recovered through the R8 ladder. Verified 13 September 2026.

Source: Vendor Published
CC on UPL and Professional Responsibility PostureA boilerplate disclaimer sits in the terms while the marketing describes the product in advice terms, or the intended audience is left ambiguous.

UPL and Professional Responsibility Posture

Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point.

A boilerplate disclaimer sits in the terms while the marketing describes the product in decision terms, which is the C band. The disclaimer is real and is contractual: the Clarivate Terms disclaim all warranties other than those expressly stated, naming accuracy, completeness and currentness among the excluded ones, which on a prior art database is a pointed exclusion because those three properties are exactly what a searcher relies on.

Against that, the product is marketed in the language of professional determinations rather than of information retrieval: enabling IP professionals to make reliable innovation decisions, supporting accurate first-pass patentability searches, informing freedom-to-operate opinions, and helping users assess patentability with speed and confidence. Freedom-to-operate opinions and patentability assessments are legal work product, and the marketing places the tool inside them.

The audience is not ambiguous, which is what keeps this off the floor: the vendor names patent researchers, patent attorneys and other IP and R&D professionals, so it is explicit that non-lawyers are intended users too. What the higher bands require is absent. Nothing published addresses where the tool's output stops and professional judgement begins, no supervision or competence dimension is offered, no statement addresses an R&D user relying on an AI Search result without counsel, and no jurisdiction or professional rule is named. Verified 13 September 2026.

Source: Vendor Published
CC on AI Governance and Bias DisclosureResponsible AI principles are published without a mechanism, a testing regime, or anything a buyer could audit.

AI Governance and Bias Disclosure

Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.

Governance is described as a structure under construction, with no mechanism, owner or testing disclosed, which is the C band. What exists is a corporate-level statement rather than a product one: Clarivate reports multi-functional committees collaborating on a comprehensive AI Governance Framework, with those committees guiding segment leaders as generative AI is woven into product offerings. That is more than nothing, because it identifies a governance structure and locates responsibility with named committees rather than with no one, but it describes a framework being built rather than one that governs.

Nothing else was located. No AI principles or responsible-AI page exists on the product estate, no individual or function is named as accountable for model behaviour, no pre-release testing regime is described, and no evaluation result is disclosed. Bias is not addressed in any form. That omission is worth naming precisely on this product rather than generically: AI Search is trained on DWPI, a corpus written by human editors, so any systematic tendency in that editorial output, whether toward particular technology areas, jurisdictions or drafting conventions, would propagate into what the model retrieves.

The vendor markets the DWPI-trained provenance as the product's central advantage and publishes nothing about testing whether the model inherits the index's blind spots as well as its strengths. R16 is noted: nothing connects the corporate framework to this product in scope terms. Verified 13 September 2026.

Source: Vendor Published
CC on AI Safety and Data StewardshipA generic privacy policy covers the product without addressing what happens to documents and prompts after processing.

AI Safety and Data Stewardship

Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.

A corporate security programme is published without addressing what happens to what a user puts into the product, which is the C band. The programme is real. Clarivate publishes an Information Security Program page stating that the ISO 27000 family, and ISO/IEC 27001 specifically, forms the basis of its information security management system, with documented policies, standards and controls conforming to that standard and an in-house team covering cloud, application and information security.

A Trust Center brings together information security, privacy and security compliance material, and a defined incident response process is stated: identify, investigate, contain, and take corrective action based on severity and impact. Staff are subject to a code of conduct. What is not published is anything product-specific about data handling. No retention period for search queries, no deletion commitment, no statement of whether AI Search inputs are logged or reused, no subprocessor list, and no description of access controls within the product.

The gap matters here because the input is an invention disclosure. Two things are recorded and expressly not credited under R16 and ground rules section 3: the detailed certification and penetration-testing material published under the Clarivate umbrella belongs to Ex Libris, a different business unit serving libraries and academia, and nothing connects it to this product; and the Trust Center states plainly that specific certifications and reports may vary by product and environment. Verified 13 September 2026.

Source: Vendor Published
BB on AI Liability and RecourseA real published position on liability, short of the full picture: commonly a stated indemnity without scope or caps.

AI Liability and Recourse

What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.

A real published position on liability, short of the full picture, which is the B band. The Clarivate Terms are published and address the commercial limbs directly. Warranties exist rather than being wholly disclaimed: stated warranties plus any product-specific warranties included in the order are identified as the exclusive warranties, with everything else excluded, and remedies are specified rather than left open.

For software, if a valid warranty claim cannot be rectified within a reasonable period the customer may cancel the licence by written notice and receive a refund calculated on five-year straight-line depreciation from the order's effective date; for professional services, Clarivate will rectify on written notice. Liability is capped and the cap is drawn precisely: aggregate liability of each party, and of Clarivate's third-party providers, for all claims including breach of statutory duty, tort and negligence, is limited to actual direct damages up to the amounts payable in the twelve months before the first incident, measured against the specific Product that is the subject of the claim.

Excluded losses are mutual, covering lost profits, business, revenue, anticipated savings, data and goodwill, and indirect or consequential loss. What is absent keeps it off A. The disclaimer expressly excludes warranties of accuracy, completeness and currentness, which are the properties a prior art search rests on; no AI-specific provision was located; no vendor indemnity was located; and no insurance position is published. Verified 13 September 2026.

Source: Vendor Published
CC on Practice Systems Integration DepthIntegrations are listed as logos or marked as coming, with no documentation an implementer could use.

Practice Systems Integration Depth

How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.

Integration is claimed and the mechanism is named without a documented catalogue or configuration detail, which is the C band. What is established comes mostly from the agreement rather than from a product surface, which is itself telling. The Clarivate Terms contemplate that Licensed Information may be made available by API or data feed, subject to information service terms, and that APIs may be offered to configure hosted and installed software or to allow it to interoperate with third-party programs, described as Client Configurations, with API and data feed keys not to be shared, used across multiple interfaces, or used to mimic Clarivate functionality.

So an API exists and its commercial boundaries are published. Within the product, integration is largely internal: DWPI and the citation index are integrated with patent full text from multiple authorities including machine translations for China, Korea and Japan, and litigation data is correlated in from Darts-ip. What the higher bands require was not established. No connector catalogue or developer documentation was reached, no direction of flow is described for any connection, and not one external system is named as supported, which is a real gap given that the buyer typically also runs an IP management or docketing system and Clarivate itself sells adjacent IP products.

The estate returned a bot-detection block and the integration surfaces were not recovered through the R8 ladder. Verified 13 September 2026.

Source: Vendor Published
CC on Deployment Model and Data ResidencyCloud delivery is implied and neither the tenancy model nor the region is stated.

Deployment Model and Data Residency

Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.

Cloud delivery is evident and neither the tenancy model nor the region is stated, which under R38 is the C band, both co-equal limbs being absent. The product is a hosted browser-based research application, described by the vendor as a proprietary patent research application accessed through the Clarivate platform and licensed by subscription, with the Clarivate Terms distinguishing hosted software from installed software and contemplating both.

So the delivery model is clear enough. Everything a buyer with constraints would ask is not. No data centre region is named for this product, no country of processing is stated, no regional election is offered, and nothing describes whether a customer's saved searches, alerts and work files sit in a shared or isolated environment. No single-tenant, private or on-premises option is offered or refused. Recorded and expressly not credited under R16 and ground rules section 3: Clarivate publishes detailed data centre and residency material covering the United States, Europe and Asia Pacific, but it belongs to Ex Libris, a separate business unit serving libraries and academia, and no scope statement connects it to Derwent.

The Trust Center itself makes the point, stating that specific certifications and reports may vary by product and environment and directing buyers to an account representative for product-specific documentation. Verified 13 September 2026.

Source: Vendor Published
BB on Security Certifications and Trust CenterCertification is real and stated, short of accessible evidence: a named standard without scope, date, or a way to obtain the report.

Security Certifications and Trust Center

Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.

Certification is real and stated, short of accessible evidence and short of any scope covering this product, which is the B band. A named standard is claimed at company level and claimed properly: Clarivate states it has adopted the ISO 27000 family, and ISO/IEC 27001 specifically, as the basis of its information security management system, with policies, standards and controls documented and implemented to conform to it, and its trust material states that the security programme aligns with ISO 27001 and SOC 2, standards it describes as requiring documented controls, independent audits and continuous improvement.

A Trust Center exists and is public, gathering information security, privacy and security compliance in one place. Two things prevent A, and the vendor states the first itself. Scope is not published per product: the Trust Center says specific certifications and reports may vary by product and environment and directs buyers to contact an account representative for product-specific security documentation, so a buyer cannot establish from public material whether Derwent Patent Search sits inside any certification boundary.

And no report, certificate, audit period or auditor is obtainable without that conversation, which under R5 is the sales-gated tier. Recorded and not credited under R16: the specific certifications set out in detail on the Clarivate estate, covering ISO 22301, 27001, 27017, 27018, 27032 and 27701, are Ex Libris's and no connector reaches this product. Verified 13 September 2026.

Source: Vendor Published
BB on Model Supply Chain DisclosureThe supply chain is partly disclosed: providers named without change notification, or architecture described without the providers.

Model Supply Chain Disclosure

Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.

The architecture is described without the providers, which the B band names expressly as one of its two shapes. What is disclosed is more than most records manage on the training side and nothing at all on the supply side. The architecture is stated: AI Search uses a language transformer model, described elsewhere on the estate as a large-language AI model. The training corpus is named precisely and is the point of the product: the model is trained and powered by DWPI, the vendor's own editorially curated index, and the vendor markets it as the only AI patent search trained on that data, explaining that this is what lets it recognise related concepts across patent publications.

Being able to name what a model was trained on, and for that corpus to be a proprietary asset the vendor controls, is a genuinely strong disclosure and is rare in this corpus. What is missing is everything downstream of that. No model is named, no version or family is given, no provider is identified, nothing states whether the model was built in-house or licensed and fine-tuned, nothing says where inference runs, and no commitment exists to notify customers if any of it changes.

The same silence covers the other AI capabilities: data enrichment, Optimized Assignee and legal status prediction are each described as AI-derived with nothing said about what performs them. Under ground rules section 3 the DWPI training disclosure is spent here on architecture and is separately the substance of the corpus provenance row. Verified 13 September 2026.

Source: Vendor Published
DD on Commercial TransparencyNo pricing information published at any level, including the unit of charge.

Commercial Transparency

Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.

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 from the product estate recovered through the R8 ladder and from the site's own navigation: product pages for patent search and adjacent Derwent offerings, a landing page for the current release, the legal centre, the trust centre and the newsroom. There is no pricing page. The only commercial route on the product page is a free trial request form, and the wider estate resolves to contact and demonstration requests.

Nothing published states the charging model, so a buyer cannot establish even the shape of it: not whether the subscription is priced per named user, per concurrent seat, per module, by search volume or by corpus coverage, and not what the AI Search capability costs relative to the base platform or whether it is included. The Clarivate Terms confirm that fees sit in an order form or ordering document rather than anywhere public, and add a term worth naming because it bears on what a buyer can compare: the agreement treats pricing and discounts as confidential information the customer must not disclose.

Under R10's closing discipline no structure means no row, so no VendorPricing row is written. Recorded and not credited under R41: a third-party listing states that no pricing is published and that buyers must contact the vendor, which corroborates rather than establishes the absence. Verified 13 September 2026.

Source: Vendor Published
BB on Firm and Practice CoverageSegment and practice coverage is described with substance, short of the boundaries: what is supported is clear, what is not is left open.

Firm and Practice Coverage

Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.

Coverage is described with real substance and quantified precisely, with the boundaries left open, which is the B band. Jurisdictional coverage is stated with more precision than anywhere else in this lane, and each figure is separately scoped rather than rolled into one claim: DWPI curation across 60 jurisdictions and 132 million patents, enhanced full text from 76 jurisdictions, bibliographic data from 109, more than 178 million patent publications, and correlated litigation case data from more than 140 jurisdictions.

Language coverage is addressed through machine translation and full-text integration for major authorities including China, Korea and Japan, with the interface available in Chinese, English, Japanese and Korean. Buyer coverage is named rather than implied, the vendor identifying patent researchers, patent attorneys and other IP and R&D professionals, and evidencing the institutional end with 40 global patent and trademark offices.

Use cases are enumerated: patentability, freedom to operate, invalidity, prosecution, portfolio monitoring, licensing and monetisation, and litigation support. What is left open is the limit in every direction. No technology area is named as better or worse covered, and the 41 new DWPI manual codes added for emerging areas imply the taxonomy lags new fields without saying by how much. No customer size floor is stated, and nothing distinguishes what this product covers from the adjacent Derwent Patent Monitor. Verified 13 September 2026.

Source: Vendor Published
Sources on file

3 public documents

The public pages on file for Derwent Patent Search, with the recorded signals each one supports and the date it was last read. Open any of them and check the reading against the record.

Legal Signals

What each signal means

A signal records what public sources say on the date shown. It is not a grade and it is not a recommendation. Where a signal reads Not addressed, it means the index did not locate the material in public sources on that date, which is a statement about disclosure rather than about the product.

Confidentiality and Privilege

Client Data in Training

Can material a lawyer puts into this product be used to train a model?

Terms silent

A published agreement or policy exists and none of it addresses the question either way, or the document that would answer it could not be read and the summary names the retrieval limit. The summary states which shape the silence takes: an improvement right granted that never names training, or no improvement right granted at all.

No located public material addresses whether what a customer puts into the product is used to train or improve models, in either direction, which is this value. The silence is specific rather than general, and the reason it matters is the input. AI Search invites free text, and the text the vendor names is an invention disclosure: unpublished, novelty-destroying if it escapes, and frequently privileged. Nothing published says whether that query is retained, whether it feeds model improvement, or whether it is used in aggregate.

What the vendor does state about training runs the other way and is recorded so the value is not misread: the model is trained and powered by DWPI, the company's own editorially curated index, and is marketed as the only AI patent search trained on that data. That is a statement about the training corpus being proprietary rather than a statement that customer input is excluded from it, and the two are not the same. R43(1) was run.

The published Clarivate Terms were recovered in substance through the R8 ladder and no training term was located in either direction; a general provision permits Clarivate to use data relating to provision and performance of the products, but the confidentiality and data sections of the current Terms were not fully recovered against bot detection, so this is recorded as silence rather than as a contractual position.

Source: Vendor PublishedAs of Sep 13, 2026Evidence

Prompt and Output Retention

How long does the product keep what a lawyer typed, and can that be set to zero?

Not addressed

No located public material states how long prompts and outputs are retained.

No located public material states how long search queries or their results are kept. Nothing on the product estate, the Trust Center, the Information Security Program page or the recovered portions of the Clarivate Terms addresses retention of what a user types into the platform. That gap has an unusually sharp edge on this product. The material a patent searcher enters into AI Search is an invention disclosure or a claim under consideration, which is confidential by definition and commercially destructive if it surfaces before filing, and the platform also stores saved searches, alerts and work files across a matter.

Nothing published states whether any of it persists, for how long, whether it is retained after a subscription ends, or whether a customer can require its deletion. Adjacent material is recorded and not credited because it answers a different question: the privacy policy describes Clarivate acting as processor for customer-entered personal data under separate data processing agreements concluded with the controller, which is a framework for personal data rather than a retention commitment for search inputs, and the Trust Center describes a corporate security programme without product-level data handling.

The surfaces read on the date shown were the product pages recovered through the R8 ladder, the Trust Center, the Information Security Program page, the privacy policy and the Clarivate Terms.

Source: Vendor PublishedAs of Sep 13, 2026Evidence

Ethical Walls and Matter Segregation

Does retrieval respect the firm’s ethical walls, or can the model read across them?

Not addressed

No located public material addresses walls or matter level segregation.

No located public material describes how access to work is controlled within a customer organisation. Nothing published sets out a role model, permission levels, administrator capabilities, or how saved searches, alerts and work folders are shared or withheld between users of the same subscription. The privacy policy notes that an account administrator or the subscribing organisation may have access to a user's data, which describes an exposure rather than a control.

The concern is real rather than formal on this product. A patent firm running searches for competing clients in the same technology field has an ordinary conflicts problem, and a corporate IP department may need to wall a search relating to an unannounced acquisition or an unfiled application from colleagues. Nothing published states whether a searcher's queries or saved work are visible to others on the account, or whether they can be restricted.

Recorded and not credited: Clarivate publishes corporate-level statements that personnel access is limited to those whose roles require it and that staff are bound by confidentiality terms, which is vendor-side employee access control rather than a customer-facing permission model. The surfaces read were the product pages recovered through the R8 ladder, the Trust Center, the Information Security Program page and the privacy policy.

Source: Vendor PublishedAs of Sep 13, 2026Evidence

Third Party Request and Subpoena Notice

If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?

Not recorded

This signal has not been recorded for this vendor yet. It is not a finding either way.

Accuracy and Authority

Primary Law Corpus Provenance

Where does the law in this product come from, and does the vendor have the right to use it?

Sources named and licensed

The vendor names its primary law sources and the licence or public domain basis for each, with an update cadence.

The corpus is named precisely, the vendor owns it, and it is described in more detail than any other record in this corpus manages, which is this value and the strongest row on this record. The source is the Derwent World Patents Index, Clarivate's own proprietary asset rather than licensed third-party content, so the rights position is ownership and the vendor states the AI is trained and powered by it. What DWPI consists of is set out at length: descriptive titles and abstracts for more than 70 million invention families, written by human editors to summarise an invention's novelty, use and advantage in standard English, produced by a team the product page puts at over 800 patent editors; subject-matter classification coding and manual indexing; and invention-centric family building in which human indexers group documents by invention rather than by filing, including non-Convention patents and Chinese dual filings that carry no linking data.

Around it, enhanced full text from 76 jurisdictions and bibliographic data from 109 covering more than 178 million publications, all of which is public patent documentation, and correlated litigation data from Darts-ip, a Clarivate company. So the closed loop is unusual and worth naming: the model is trained on an index the vendor built by hand and continues to own. One discrepancy is recorded rather than resolved: the editor count is given as over 800 on the product page and as 850-plus in a November 2025 release.

Source: Vendor Publishedthe only AI patent search trained and powered by DWPIAs of Sep 13, 2026Evidence

Good Law Verification

Does the product tell you when the authority it just cited has been overruled?

Own treatment signal

The vendor computes and surfaces subsequent history itself, with the method described.

The vendor publishes its own status signal and surfaces it inline, which is this value, and R15 governs the translation because the patent analogue of good law is not case treatment. For a patent searcher the question a citator answers is whether the authority still stands, and here that means whether a patent is in force, when it expires, and how much life remains. The platform answers all three and does so as a derived signal rather than as raw data: AI-derived estimated expiration date and remaining life, and an AI-derived dead-or-alive status, offered across both DWPI and INPADOC families, alongside legal status data and legal status alerts.

Optimized Assignee with Ultimate Parent performs the adjacent function of resolving who currently owns the right, following the corporate tree to ultimate parent, which is the ownership half of the same question. Correlated litigation case data from more than 140 jurisdictions shows which patents in a result set have been litigated, with history and detail per case, which is the closest patent equivalent to negative treatment.

Two limits are recorded. The status signals are stated to be AI-derived and no accuracy or confidence information is published for them, on determinations that bear directly on freedom-to-operate conclusions. And nothing published describes how often the underlying legal status data is refreshed.

Source: Vendor PublishedAI-derived Estimated Expiration Date/Remaining Life and Dead/AliveAs of Sep 13, 2026Evidence

Refusal and Uncertainty Behaviour

What does the product do when the answer is not in the corpus?

Not addressed

No located public material addresses what the product does when it cannot ground an answer.

No located public material describes what the system does when it cannot produce a reliable answer. Nothing states that AI Search signals low confidence in a ranking, reports that a query falls outside what the model handles well, distinguishes results it is confident about from results it is not, or tells a user when a search should be supplemented by conventional keyword or classification searching. The same silence covers the AI-derived signals: nothing indicates when an estimated expiration date or a dead-or-alive determination is uncertain, or when an Optimized Assignee resolution could not be made confidently.

The consequence is specific to prior art work and runs the wrong way. A search tool's characteristic failure is silent omission: the relevant reference that never appears in the ranking. Unlike a generative tool, which produces a visibly wrong answer a reader can question, a retrieval system that misses simply returns a shorter list, and nothing published tells a searcher when that has happened or how to detect it. The vendor's own framing acknowledges the risk obliquely, describing AI Search as helping users validate results to ensure relevant records are not missed, which places the burden of catching omissions on the searcher without describing any mechanism the system provides.

The surfaces read were the product pages recovered through the R8 ladder, the December 2024 launch release, the Trust Center and the Clarivate Terms.

Source: Vendor PublishedAs of Sep 13, 2026Evidence

Fabricated Citation Record

Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?

None located

No court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product has been located as of the date shown. This is a statement about the public record on that one subject, not a finding about the product, and this signal is not a litigation history.

Searched on 13 September 2026 against the company name, the current product name and the former product name Derwent Innovation, across reporting and trackers covering court decisions on AI-generated fabricated citations. None located. No decision, sanction or disciplinary referral names Clarivate, Derwent, Derwent Patent Search or Derwent Innovation. Context is recorded because the field searched now includes patent practice specifically, so the absence was tested against directly comparable material: in a Kansas patent infringement action a judge fined counsel for a licensing company a total of 12,000 dollars across four lawyers after briefs contained case citations, quotations and representations of authority fabricated by generative AI, with one attorney admitting to using a general-purpose assistant without verification and referred for state disciplinary attention.

General-purpose assistants rather than patent search platforms are what that and the other reported instances describe. Under R119 this signal records fabricated legal citations in filings and nothing else. One point of product context: this platform retrieves patent records from a closed curated index rather than generating citations to legal authority, so the exposure this signal tracks is structurally low, and a failure here would take the form of a missed reference rather than an invented one.

Source: Bar Guidance or Court RecordAs of Sep 13, 2026
Professional Responsibility

Bar Guidance Alignment

Has the vendor engaged in public with the ethics opinions its buyers are bound by?

Not addressed

No located public material engages with bar or ethics guidance.

No located public material engages with bar or professional conduct guidance. No bar association, patent bar rule, ethics opinion or professional regulator is named or mapped to the product, in any jurisdiction, and no statement asks customers to use the platform consistently with their own professional obligations. That is a notable omission on this product class specifically, because patent practice is one of the few areas with its own dedicated professional regime: registered practitioners before the United States Patent and Trademark Office are governed by that office's rules of professional conduct, and the duty of disclosure of material prior art is a defining obligation of the field.

A tool sold to help find prior art sits directly against that duty and engages none of it. The vendor's closest approach is descriptive rather than normative, describing the platform as helping users find the right prior art when the stakes are high and as supporting reliable innovation decisions. The regulatory engagement that exists elsewhere on the corporate estate runs to information security and data protection rather than to professional conduct.

Recorded and not credited because it addresses the customer's institutional character rather than any guidance: the platform is stated to be used by 40 global patent and trademark offices, which is adoption evidence and is graded on the operational row.

Source: Vendor PublishedAs of Sep 13, 2026Evidence

Billing and Fee Posture

Does the vendor address what happens to the bill when the work takes an hour instead of six?

Not addressed

The product sits inside a lawyer to client fee relationship and no located public material addresses billing, fee or disclosure treatment, with no savings claim published either.

Nothing published addresses what happens to the bill when AI-assisted work takes an hour instead of six, which is the floor, and the question arises squarely on this product. Prior art searching is classically billed to a client, either as attorney time or as a search cost passed through, and the vendor's entire pitch is compression of exactly that work: finding more relevant patents in less time, saving time when evaluating patentability, and placing the most relevant records at the top of the list rather than after reviewing 50 to 200 results.

That is a direct claim to reduce chargeable search effort. Nothing follows from it in disclosure terms. No per-matter record of AI-assisted searching is described, nothing marks a result set as AI Search-derived rather than keyword-derived for the purposes of a search report or fee note, and no guidance is published on fee or disclosure treatment for a firm passing search costs to a client. The gap is compounded by the pricing position, since with no charging model published at all a firm cannot establish what the platform component of a matter cost would be, let alone how AI-driven efficiency changes it.

Recorded and expressly not credited under R21 and R24: the Clarivate Terms treat pricing and discounts as confidential information the customer must not disclose, which restricts what a firm may tell anyone about its costs rather than addressing the client's bill.

Source: Vendor PublishedAs of Sep 13, 2026Evidence

Outside Counsel Guideline Readiness

Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?

Not addressed

No located public material supports a client side disclosure obligation.

None of the three artifacts a firm would need is published, which is the floor. There is no subprocessor list: the privacy policy identifies processors by category only, naming cloud computing providers, web analytics services and payment processors as examples, and no entity is identified as processing customer content for this product. There is no model provider statement, because no model, version or provider is named anywhere on the estate for AI Search or for the AI-derived enrichment and legal status capabilities.

And there is no forwardable client-facing pack: no data processing addendum is published, the privacy policy stating instead that Clarivate concludes separate data processing agreements with controllers, so the instrument exists and is negotiated rather than available in advance. The Trust Center compounds the position rather than relieving it, stating that specific certifications and reports may vary by product and environment and directing buyers to an account representative for product-specific security documentation, which is exactly the sales conversation this signal asks whether a firm can avoid.

Recorded and not credited under R16 and ground rules section 3: the detailed certification set published under the Clarivate umbrella belongs to Ex Libris, a separate business unit, and Darts-ip is named as a data source rather than as a processor of customer content.

Source: Vendor PublishedAs of Sep 13, 2026Evidence

Court Disclosure Support

If a judge’s standing order requires an AI disclosure, can the product produce one?

Not addressed

No located public material addresses court disclosure or verification certification.

No located public material addresses disclosure of AI involvement in the work the product supports, which is the floor, and the forum where it would matter here is the patent office rather than a court. A prior art search underpins a patentability opinion, a freedom-to-operate opinion and, in prosecution, the applicant's duty to disclose material references. Where a search was performed with AI assistance, a firm may need to describe how it was conducted, whether to a client questioning the search's adequacy, to an examiner, or to a court in later invalidity litigation testing whether a reference should have been found.

Nothing published helps. No search history export is described as an evidentiary record, nothing states that a result set is marked as AI Search-derived rather than produced by keyword or classification searching, no search report template is offered, and no guidance addresses whether or how the use of AI Search should be recorded or disclosed. Two adjacent capabilities are recorded and not credited because they document the art rather than the method: the Derwent Patent Citation Index provides forward, backward, literature and examiner citations at family level, and search history cards and search management tools exist within the interface. Both help a searcher work; neither is offered as a record a third party could rely on.

Source: Vendor PublishedAs of Sep 13, 2026Evidence
Contact

Correct a record, or ask how something was graded

Every grade and every signal on this index is drawn from public sources and dated. If a record is wrong, out of date, or missing an artifact the index did not locate, send the source and it will be reviewed and the record redated. Vendors are welcome to submit documentation. Nothing on this index is for sale, including a listing, a placement, or a grade.

AI Legal Index

The AI Legal Index is an independent index that tracks changes to AI vendors in legal. It holds 61 vendors across 9 categories, each graded on the same 15 capability axes and recorded against 12 legal signals, from public sources. No vendor pays for inclusion, placement, or rating.

Index Status
Last index update
September 13, 2026
The AI Legal Index is an editorial reference. It is not a regulatory body, not a law firm, and nothing published here is legal advice or a recommendation to retain or avoid a vendor. Records are verified against published sources, bar guidance and public court records. Where a record reads not addressed, the material was not located in public sources on the date shown. See the Methodology page for evaluation standards and limitations.
© 2026 AI Legal Index
3801 N Capital of Texas Hwy, Ste E240 · Austin, TX 78746