IP Author vs XLSCOUT: how they compare in 2026
IP Author and XLSCOUT both sell AI patent drafting and prior art search to the same practitioner, and the difference between them is almost entirely documentary. XLSCOUT publishes a master agreement. Clause 6.4 undertakes that customer inputs and outputs are not used to train, retain or improve its models or those of its suppliers. Clause 7.3 requires prompt written notice before a compelled disclosure so the customer can seek an order, and clauses 5 and 8 set caps, named carve outs and an indemnity with a remedy path. XLSCOUT also names Prescient Assurance as its SOC 2 Type II auditor and publishes its embedding model openly. IP Author publishes little of that, and what it does publish is aimed at the work: office action support enumerated by statutory ground, prior art search built into the drafting flow, and masked exports that withhold filing dates and raw excerpts so a finding can reach a client without oversharing.
At a glance
All 15 axes, side by side
The same grid applied to every vendor in the index, graded from public sources. Hover a grade to see what the letter means on that axis.
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.
Every IP Author module is generative, and nothing underneath works without the models. Drafting produces claims, specification and figures from a disclosure. Prior art search runs semantically from the disclosure text rather than from Boolean queries. Office action responses are generated with arguments and claim mapping from the application history. Classification is model driven, and the Evidence of Use assistant is an analysis product. The platform reads images, tables and diagrams inside Word files, PDFs and slide decks, and generates flowcharts and block diagrams where a disclosure lacks them. Without the models there is no product, not even a database, since the patent literature is searched rather than owned.
The models are the entire product, and the naming says so: seven of the nine modules have LLM in the product name. Novelty Checker LLM, Invalidator LLM, ClaimChart LLM, PatDigger LLM, Standigger LLM, TechScaper LLM and Drafting LLM each exist only as generative or retrieval model applications, and Ideacue generates invention concepts outright. Underneath sits ParaEmbed, an embedding model XLSCOUT fine tuned on patent and technology data and published on Hugging Face, with in house RAG models and Azure endpoints. There is no document management, docketing, renewals engine or services layer beneath the AI that would survive its removal. Remove the models and nothing usable remains.
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.
Search results come with an account of how they relate to the invention, alongside one published accuracy claim without a method. Each reference comes with a similarity score, a generated novelty summary, an explanation of overlaps and differences, highlighted passages and side by side comparison, so a practitioner can test a result rather than accept it. The vendor states that, compared with a human searcher, the tool returns similar or the same results 19 out of 20 times. No sample size, evaluator, definition of similar, date or method is given. The vendor also publishes its own comparisons against named competitors, which are self published.
Grounding is documented and the retrieval method is described, and figures are published without a test set behind them. Output maps to real patent documents a reader can open. Novelty reports map key features against the references, Invalidator LLM uses ParaEmbed to map prior art against the claims and surfaces the fifty most relevant references, and ClaimChart LLM maps product features to claims. The retrieval mechanism is named, and the embedding model is open source on Hugging Face, so an outsider can evaluate part of the system directly. Four figures are published, including a candid 60 percent accuracy figure for PatDigger LLM in finding top licensing opportunities and a 95 percent noise reduction claim for TechScaper. None has a described test set, method or date. The master agreement is franker than the marketing. It calls automated patentability and invalidity searches first pass results delivered as is, which shall in no manner be a comprehensive or full search, and clause 4.5 disclaims any warranty of accuracy, completeness or correctness.
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.
The oversight commitment is stated twice. Product material says users keep full review, editing and decision authority at every stage of drafting, and the FAQ says attorneys and agents stay in full control of the final work product. Output is framed throughout as a first draft or first pass response. What is missing are operating limits: nothing says what the system does unattended, what confidence threshold applies, or what happens when its tools run inside external agent frameworks, where full user control is hardest to keep.
Automation is the whole pitch, and no control structure is published to balance it. The modules are described as automated drafting, automated novelty reports and automated invalidity reports, with Ideacue said to speed ideation tenfold. Nothing says what runs unattended, where a reviewer comes in, what the system does at low confidence, or what happens when it is wrong. The nearest thing to a control is in the contract, not the product: the master agreement calls search output a first pass that is in no manner comprehensive. That tells a buyer the work is unfinished without describing any way to finish it inside the product. One module names a human step, the AI assisted source code review described as validated by experts, but that is a service design, not an oversight model for the platform. Searched the home page, security pages, SOC 2 page, master agreement and module descriptions. This matters because Drafting LLM produces patent applications that a practitioner will file.
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.
No customer, firm, case study, usage figure, funding announcement, analyst placement, award or third party review is published. The vendor's content library, including competitive comparisons and benchmarks, is self authored, so all the evidence is the vendor describing its own product. Checked the home page, the product pages, the FAQ, the blog library and the site navigation. This records an absence on those pages, not a judgment about adoption.
Logos and unattributed testimonials stand in for evidence, and one contract clause makes that worth reading carefully. Four customer logos appear, Mitsubishi, NTT, IIT Madras and IDAIR, against a claim of more than 200 clients worldwide with no basis stated. Five testimonials are published and every one is anonymous, attributed only by role and sector: a patent attorney at a US law firm, an IP manager at a machinery manufacturer, a senior IP manager in industrial automation, a senior innovation manager at an engineering company and a partner at a US law firm. No organization is named against any quote, no quote carries a date, and the published figures for time saved and efficiency gained are product claims rather than measured customer outcomes. Clause 10.5 of the master agreement is relevant to how much weight this evidence carries: customers agree in advance to participate in marketing activities including writing references and case studies, speaking at XLSCOUT events and appearing in webinars. A case studies library exists in the navigation and was not opened, so a named dated outcome was neither found nor excluded.
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.
There is one thoughtful confidentiality feature and no professional confidentiality position. Prior art results can be exported in masked versions that withhold filing dates and raw excerpts, so findings can be shared with clients or outside counsel without oversharing. Nothing located addresses attorney client privilege, work product, or the patent specific risk that exposing an unpublished application can affect patent rights. The Trust Center and privacy policy are published and were not read.
The core commitments are in the contract, not marketing, and two parts are still missing. Clause 6.4 of the master agreement says customer data is not used to train, retain or improve the models of XLSCOUT or its third party suppliers, and clause 6.3 confirms the customer keeps all right, title and interest in its data. Clause 7 is a mutual confidentiality regime with a defined category of Confidential Information, limiting disclosure to employees, contractors and suppliers who are themselves bound. Tenant separation is documented on the SOC 2 page: each customer environment sits in a dedicated trust zone, data is separated with unique encryption keys, and role based access with least privilege applies. Missing is separation within an account, so nothing addresses walls between users, teams or matters, which matters because the published testimonials come from US law firms. Privilege and work product are not mentioned on any page read, and the retention picture is muddled, as the retention signal sets out.
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. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
No professional responsibility position is published. The product drafts full patent applications and office action responses under sections 101, 102, 103 and 112, which are filed under a registered practitioner's signature, and its self serve trial means the user may not be an attorney. The vendor says attorneys and agents keep full control of the final work, which is a product statement rather than a professional responsibility framework. No reference to the USPTO Rules of Professional Conduct, 37 CFR, the duty of competence, or bar or patent office guidance is published.
A real limit on what the product is sits in the agreement, not a footer, but the professional side is untouched. The master agreement defines Automated Patentability search and Automated Invalidity as first pass results from running the proprietary algorithm on the database as is, which shall in no manner be a comprehensive or full search. That is a contract statement of what the tool does not do, binding on the vendor and readable before signing. Clause 4.5 adds an express disclaimer of any warranty of accuracy, completeness or correctness. But nothing says XLSCOUT is not a law firm or does not give legal advice, no jurisdiction limit is named, and nothing engages a practitioner's supervision or competence duties. The gap is sharpest for Drafting LLM, which produces patent applications, and Ideacue, which is sold to R&D and innovation teams who are not practitioners.
AI Governance and Bias Disclosure
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
Nothing is published on how the models are governed, tested or monitored. No AI policy, model card, bias testing, evaluation method, drift monitoring, governance body, ISO 42001 or EU AI Act position is published. The vendor's blog comments on AI industry news and on IP questions raised by AI, which is about the field rather than its own models.
No governance position of any kind is published for a platform whose entire output comes from models. There is no responsible AI page, AI policy, ethics statement, ISO 42001 or EU AI Act position, named accountable owner, testing regime before release, or disclosure about uneven output across technology areas or document types. The full site navigation was read. It covers the modules, company pages, pricing, policies, security pages, resources, terms, privacy, content security policy, research collaborations and data coverage, and it has no governance or responsible AI entry. Two items come close but are not governance. A founder's video on global AI governance at the United Nations is commentary on the field, not a statement about this product. The research collaborations with Mitacs, Hugging Face and four universities are development partnerships. The no training promise in clause 6.4 is a data commitment, covered under confidentiality.
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.
No position on data handling is published on the pages read. Nothing says whether invention disclosures, drafts, search history or generated work are used to train models, how long they are kept, or how they are deleted. The content is unpublished invention disclosures uploaded before filing, and the vendor's own masked export feature shows it understands that disclosure control matters. The Trust Center and privacy policy are published and were not read, and either may answer these questions, so this reflects only the home page, the product pages, the FAQ and the blog library, checked.
Most of the ground is covered in specific terms, with two gaps. Published: encryption at rest and in transit with unique keys for each customer, role based access control with least privilege and regular review, and penetration testing by outside firms and internal engineers. Also published are static and dynamic application security testing including open source libraries, OWASP Top 10 development practice, background checks and NDAs for all staff and contractors, continuous monitoring, and a statement that customers can delete data at any time. The SOC 2 Type II audit gives independent assurance over those controls. Missing is a named subprocessor list. Clause 6.2 of the master agreement says inputs are shared with trusted third party suppliers without naming them, and the security page names AWS and Microsoft Azure followed by and others. Also missing is any incident or breach practice: nothing on notice timing, contents or route. No retention period is published for customer content, and the Content Security Policy, which the agreement makes the governing document for storage, was not opened.
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.
No published position on liability for AI output is published: no warranty, service level or remedy. The exposure is real, and the speed claim sharpens it. A complete first draft in fifteen minutes shortens the time in which errors would surface, and claim scope lost at grant cannot be recovered. A wrong traversal argument in an office action response can create prosecution history estoppel. The vendor's answer is that attorneys keep full control, which puts responsibility on the user by design rather than by contract, and no contract is public.
A real liability position is published. The master agreement includes an indemnity with defined scope. Clause 8.1 has XLSCOUT defend and indemnify the customer against third party claims arising from breach of the warranties at 4.1 and 4.3, which cover its right to grant the licenses and its authority to contract. Clause 8.3 sets out the remedy if an injunction is sought: procure, modify or terminate with a pro rata refund. Clause 5.2 sets caps: the fees paid for the service in question, or, for matters unrelated to a service, the total fees paid in the prior twelve months. Clause 5.3 carves out the indemnity obligations and the customer's IP infringement, and 5.1 excludes indirect and consequential loss for both sides. But the indemnity covers title and authority, not output. Clause 4.5 expressly disclaims any warranty of accuracy or correctness, and no insurance position is published. A buyer knows exactly who bears the loss when a draft or search is wrong: the buyer.
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 strong inside the platform and thin outside it. Prior art search is built into drafting, so references carry into claims and specification without retyping, and results export to Word or Excel. Externally, the vendor says its search, claim charting, whitespace analysis and office action tools connect to agent frameworks, the only external connection described. No docketing system, patent management platform, document management system, Microsoft Word integration or API documentation is published. For a prosecution tool, docketing is the main gap.
No integration into the systems where legal work already lives was found, and the product stands alone. The full site navigation, read, has no integrations page, API or developer entry, documentation index or partner integration list; the Authorized Partners page, by its title, lists resellers and channels. Nothing names a connection to an IP management or docketing system, a document management system such as iManage or NetDocuments, a patent office filing system, Word or Outlook. The modules are described as a connected suite, but that links XLSCOUT's own parts, not a firm's systems, and the agreement's references to a Delivery Method and Documentation describe no interface. A customer moves work into the platform and exports results out. Checked the home page, the module entries, the security pages, the master agreement and the full navigation.
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.
Nothing is published on where the platform runs: no hosting provider, region, residency commitment, or single tenant or on premises option. That matters in patent work, because foreign filing license rules and export controls can limit where an unpublished application may be processed, and the platform serves users across more than 100 jurisdictions. The Trust Center, where such a statement would usually sit, was not read.
Both the tenancy model and the residency choice are stated, and processing location is left open. The SOC 2 page describes a multitenant cloud with complete customer isolation. Each customer's environment sits in a dedicated trust zone, using what the company calls a patented isolation approach, with unique encryption keys separating customers from each other and from XLSCOUT. The security page says clients can choose to store data on United States or European servers, and names AWS and Microsoft Azure among its providers, followed by and others, which leaves the list open. Residency is offered for storage, but processing location is not addressed separately. That matters here because the same page says inference runs partly through Microsoft Azure endpoints, and nothing says whether a customer who picks EU storage also gets EU inference.
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.
IP Author publishes a Trust Center and a privacy policy, and the Trust Center was not read. No certification is published on the pages that were read: no SOC 2, ISO 27001, ISO 42001, auditor, penetration testing partner or audit date. A published trust surface is not an absence, and the Trust Center may hold certifications that answer this.
The attestation is real, specific and independently named, but it cannot be obtained. A dedicated page says XLSCOUT completed an AICPA SOC 2 Type II audit, names the auditor, Prescient Assurance, a registered public accounting firm in the US and Canada, and reports an unqualified opinion. It also details the control areas covered across personnel, development, testing and cloud security. What is missing is access and currency. No examination period or report date appears, so a buyer cannot tell whether the attestation is current, and no route to request the report, portal or questionnaire pack was found. The ISO 27001 claim on the security page carries an asterisk saying it applies to the data center partners, not to XLSCOUT. The SOC 2 page also includes boilerplate about digital identity offerings and managing billions of digital identities, which does not describe this business.
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.
IP Author named a provider once, in passing. An April 2024 blog post comparing it with general purpose tools says IP Author also used ChatGPT, and that the difference lies in prompt engineering, its patent search engine and user experience. That statement is not repeated in current product or trust material and gives no model, version or scope, so a buyer cannot tell whether it still holds. The vendor also says its tools connect to agent frameworks, which is about orchestration rather than models. No subprocessor list is published.
The architecture is described, and the third party model is left unnamed. The security page says XLSCOUT uses its own fine tuned LLMs, embedding models and RAG models together with Microsoft Azure endpoints to draft patents, identify infringements and search prior art. That separates what it built from what it calls out to. Its own component is published, not just named. ParaEmbed, an embedding model tuned on patent and technology data, is open source on Hugging Face and has reached a second version, so a customer can inspect the actual retrieval model. Two things are missing. The generative model reached through the Azure endpoints is never named, so whose model reads an unfiled invention disclosure is unknown; naming the cloud identifies infrastructure, not the model. And no commitment to notify customers when the models change was found on any page read.
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.
IP Author has a pricing page in its navigation and a 14 day free trial that, per the FAQ, starts without going through sales. That is a more open buying path than a demo request. The pricing page itself was not read, so no price, range, tier or unit of charge is recorded here.
A page titled XLSCOUT Pricing exists and contains no pricing at any level. It says pricing models are flexible and tailored, and offers a free trial request, a demo booking and a lead form whose text says the sales team will give detailed pricing tailored to the organization. No rate, band, floor, currency, unit of charge, tier, package or implementation cost appears. The modules are named separately across the site, but no packaging, bundling or price structure by module is published, so even the shape of the offer is hidden. The vendor built a pricing page and used it only as a form.
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.
The office action grounds IP Author handles are listed, which is what a prosecution practitioner checks first: rejections under sections 102 and 103, 101 subject matter eligibility, 112 written description, enablement and definiteness, and double patenting, always with a traversal strategy. Search covers worldwide patent and non patent literature and technical standards across more than 100 jurisdictions, filterable by jurisdiction, CPC and IPC class and assignee. Drafting is stated to follow USPTO, EPO and other office requirements, and life sciences and chemistry work includes sequence handling. What is missing: the drafting jurisdiction list names only two offices, no corpus date range or update lag is published, and the full technical domain list was not stated in the material read.
Practice coverage is documented module by module, and segment coverage is thin. Nine distinct workflows are named: prior art and novelty search, invalidity search, infringement analysis through claim charts, portfolio licensing and monetization, standard essential patents, technology and competitor monitoring, patent drafting, ideation and continuation strategy, and source code review. Two analytics dashboards cover sustainability, mapping the Sustainable Development Goals and green technologies to patent literature. Languages are English, Japanese, Chinese and German, and the company operates from five countries. Who it is for is answered in one FAQ sentence listing patent attorneys, legal teams, inventors, R&D departments and businesses protecting IP. There are no segment pages, and nothing on firm size, portfolio size or fit for in house teams versus firms. The one limit stated concerns capability: the agreement's definition of the searches as first pass.
The 12 legal signals, side by side
Recorded rather than graded. These are the questions a practitioner has to answer before a tool touches a client matter, and the answers are taken from public material only.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
No statement either way was located on whether invention disclosures, drafts, search history or generated work are used to train or improve models. The quoted phrase is the vendor's clearest statement about the user and the system, and it concerns control over work product, not data use. The content is unpublished invention disclosures uploaded before filing. The Trust Center and privacy policy are published and were not read, and either is where such a statement would usually sit.
Checked the home page, the product pages, the FAQ and the blog library on 29 Aug 2026. A buyer should ask for the training position in writing.
The prohibition is in the master agreement, and it covers suppliers as well as the vendor. Clause 6.4 of the XLSCOUT General Services Terms and Conditions says the data, defined in clause 6.2 to include both input and generated output, is not used to train, retain or improve the models of XLSCOUT or its third party suppliers, and that any sharing is under confidentiality obligations. The marketing agrees: the security page says XLSCOUT does not use customer data to train its LLMs, and the FAQ says it does not train its AI models on customer inputs.
Clause 6.1 licenses XLSCOUT to use inputs and outputs only to perform the services, so training falls outside the permitted purpose.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
No retention period was located for uploaded disclosures, drafts, search results or office action responses, and no deletion right is described. The Invention Management module implies disclosures are stored, and nothing says for how long or on what terms. The Trust Center and privacy policy were not read, and either may answer this. Checked the home page, the product pages, the FAQ and the blog library on 29 Aug 2026.
The marketing and the agreement point in different directions and the agreement governs. The security page runs a heading reading Zero Data Retention or Training and states that the model does not retain any user data and that the customer can delete data at any time. The master agreement grants XLSCOUT a worldwide sub-licensable license at clause 6.1 to collect, process, store and otherwise use the customer's inputs and the outputs generated from them, and clause 6.2 spells out that this includes the right to store that data as per the content security policy.
Both can be true, since a model not retaining a prompt is a different thing from a platform storing a project, but a buyer reading the headline would not expect the storage license. No retention period is stated in either place. The Content Security Policy, which the agreement makes the governing document for storage, was not opened on 1 September 2026 and is where a period would sit if one is published.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
No permission model, matter level access limit or tenant separation was located. Conflicts are concrete in patent practice, where a firm may prosecute in the same field for competing clients, and nothing says what separates one matter's disclosure from another or what the Invention Management module shows across a firm. Masked export controls what leaves the platform, not who can see what inside it. Checked the home page, the product pages, the FAQ and the site navigation on 29 Aug 2026.
XLSCOUT documents its own permission and isolation model rather than inheriting one, and it operates at the tenant rather than the matter. The SOC 2 page states that all customer cloud environments and data are isolated using a patented isolation approach with each customer environment held in a dedicated trust zone to prevent accidental or malicious co-mingling, that each customer's data is separated from other customers and from XLSCOUT using unique encryption keys, and that role-based access controls and least privilege are applied with access reviewed and revoked as needed.
The product does not index a firm's own document estate, so the classic wall question bites less here than on a document-based tool. What is not addressed is separation inside a single customer account, between users, teams or matters, which is the level a law firm would need and the published testimonials include two US law firms.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
No clause on government or law enforcement requests, no commitment to notify a customer before producing its data, and no transparency report were located. The stakes are real: the platform holds unpublished disclosures and pre filing drafts, and disclosing an unpublished application to a third party can bear on novelty. The privacy policy was not read. Checked the home page, the product pages, the FAQ and the site navigation on 29 Aug 2026.
The commitment is in the confidentiality section of the master agreement, not a privacy policy. Clause 7.3 says that where the receiving party is legally compelled, or asked by a regulator, to disclose confidential information, it will give the other party prompt written notice unless legally barred. The purpose is to let the other party seek an order preventing disclosure or another remedy, and the receiving party will cooperate if such an order is sought.
If no order is obtained, disclosure is limited to what counsel reasonably considers legally required. The duty is mutual, and because customer inputs and outputs are within the defined Confidential Information, it covers the material a firm cares about, not just personal data. No transparency report was found on 1 September 2026.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
The search corpus is described by type and reach: worldwide patent literature, non patent literature and technical standards across more than 100 jurisdictions, filterable by jurisdiction, CPC and IPC class, assignee and publication details. Naming non patent literature and standards matters, because prior art goes beyond patents and standards documents often carry publisher licensing. What is missing: no database, publisher or standards body is named, no license basis is given, and no date range or refresh frequency is published, though results are said to arrive in near real time. A practitioner cannot tell what was searched or how current it was.
Coverage is described by scale and geography, with no sources named. The home page claims patents from more than 100 countries and non patent literature, alongside a jurisdiction count, but the numbers are animated counters that could not be read. No database, aggregator or patent office is named as a source, no license or rights basis is stated, and no update schedule is published. What is identified is the training, not the sources: ParaEmbed is described as an embedding model tuned on patent and technology data and is published on Hugging Face. A Data Coverage page in the site navigation was not opened on 1 September 2026 and may name sources.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
Nothing says whether case law cited in generated traversal arguments under sections 101, 102, 103 and 112 is checked for current treatment before it reaches a filing. The patent version of the question also applies to search: whether a surfaced reference is granted and in force, lapsed, expired, amended or invalidated. Search filters cover publication details, which is bibliographic rather than legal status.
No citator applies. The platform retrieves and analyzes patent documents and non patent literature; it does not return legal authority whose later history could be checked. The nearest equivalent would be the legal status of a retrieved patent, and nothing says whether status or validity events are shown against references in a novelty or invalidity report. Searched the module descriptions, the home page, the FAQs and the master agreement on 1 September 2026.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
Search results show similarity scores and novelty summaries, a useful relevance signal for each returned reference. That is not a confidence signal about the system's own output. Nothing says whether a weak generated claim is flagged, whether drafting declines when a disclosure is too thin, or what happens when the office action module cannot build a traversal. The vendor says a traversal strategy is always produced, which is the opposite of an abstention path. Checked the home page, the product pages, the FAQ and the blog library on 29 Aug 2026.
Nothing found says what any module does when it cannot support a result. Relevance ranking exists, with Invalidator LLM showing the fifty most relevant references, but ranking is part of retrieval, and no confidence or grounding score is shown to the user. No way to abstain or decline is documented for the drafting, ideation or claim charting modules. The FAQ acknowledges that no system can guarantee 100 percent accuracy, and the master agreement calls search output first pass and as is, but both put the burden on the user rather than describing how the system behaves.
Searched the home page, the nine module entries, the FAQs, the security pages and the master agreement on 1 September 2026.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
General web searches for the vendor and product names with court, order, sanction, fabricated citation and patent terms found nothing on 29 Aug 2026. No docket database, patent office record system or court record tracker was searched, so this is what that search found, not a clearance. The likely failure would be a wrong prior art characterization or a bad case citation in an office action response, which would show in a prosecution file rather than a published opinion. No customer of the vendor is publicly identified.
The AI Hallucination Cases database maintained by Damien Charlotin at HEC Paris, 2026 sanctions trackers and trade coverage were searched on 1 September 2026 for the company name and module names, including Novelty Checker LLM and Drafting LLM. No court order, opinion or disciplinary record naming XLSCOUT or any module was found. This describes the public record, not the product. The risk fits only in part. The search modules return real patent documents rather than generating citations, but Drafting LLM produces patent application text and ClaimChart LLM produces claim charts, and both can reach a patent office or a dispute.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
No ethics opinion, USPTO Rules of Professional Conduct, 37 CFR, duty of competence or bar guidance was located. The vendor's large practitioner content library covers drafting workflows, tool comparisons and AI news, not the professional rules governing the practitioner who signs the filing. Checked the blog library, the home page, the product pages, the FAQ and the site navigation on 29 Aug 2026.
No engagement with professional responsibility guidance of any kind was found. Nothing addresses ABA Formal Opinion 512, USPTO guidance for practitioners on AI, EPO or other national patent attorney conduct rules, or any state bar opinion. There is not even a general statement that the user's professional obligations still apply when using the tool. Searched the full site navigation on 1 September 2026, including the home page, the nine module entries, the FAQs, the master agreement, the privacy and security pages and the company pages.
This matters for a platform that drafts patent applications and is sold to patent attorneys and law firms in five countries.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
IP Author claims time savings: a complete patent draft in fifteen minutes against weeks by hand, and prior art results in near real time against days of manual searching. No method, sample or definition of a complete draft is given. Nothing addresses the client's bill: no guidance on recording AI assisted drafting time, and no record of how much of an application or response was machine generated, though firms often bill prosecution hourly. Checked the home page, the product pages, the blog library and the FAQ on 29 Aug 2026.
Time and efficiency savings are the central marketing claim and nothing addresses what happens to the invoice. Drafting LLM is said to cut patent drafting time by up to 50 percent, ClaimChart LLM to achieve 75 percent greater efficiency than traditional manual methods, and Ideacue to accelerate ideation and brainstorming tenfold. No guidance accompanies any of it on how a firm should bill for or disclose AI-assisted drafting, searching or charting to the client paying for the work, and no per-matter record of AI-assisted work is described for any module that a firm could interrogate.
The audience for the savings claims is explicitly the buyer's own cost base, with the monetization modules framed around revenue rather than fees.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
IP Author publishes a Trust Center and a privacy policy in its site navigation, which gives a firm a place to point a client. The Trust Center's contents were not verified. Nothing was located outside it: no named certification, subprocessor list, auditor, data processing agreement or downloadable summary. A firm should open the Trust Center and confirm what is there before relying on it.
The material exists behind the signed agreement, not in public. Clause 6.2 of the master agreement says customer inputs may be shared with trusted third party suppliers who may work on the services and access all or part of the data. Clause 6.4 says such sharing happens under confidentiality obligations agreed between the customer and XLSCOUT, or under the content security policy. No supplier is named. The security page names AWS and Microsoft Azure as cloud providers, followed by and others, and Azure endpoints as where generative inference runs.
That identifies infrastructure, not the model providers a client's AI clause asks about. A firm can get the disclosure by contracting, but cannot forward anything to its client today.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
On search, similarity scores, novelty summaries, highlighted passages and side by side comparison let a practitioner show what a novelty position rests on, and exports go to Word or Excel. On drafting, renaming a figure updates every reference across claims, specification and legends, which prevents a common cause of procedural rejection. Two gaps: output does not record which model produced it, and no record of human review is kept, so a practitioner cannot evidence that they reviewed a generated claim set or argument. The forum here is usually the patent office file, not a court.
Nothing located addresses producing an AI-use disclosure or a verification certification. No export tying a passage to the model that produced it, the references it drew on, or the person who reviewed it is described for any module, and no AI-generated content marking or labeling convention is published. The automated novelty and invalidation reports do record which references were surfaced against which claim features, which is a retrieval record of a kind, but nothing states that it identifies the model or captures human review.
Searched the module entries, the FAQs, the security pages and the master agreement on 1 September 2026. The obligation fits prosecution work imperfectly, though Drafting LLM output can reach a patent office and ClaimChart output can reach a dispute.
The questions both sides leave open
Derived from the records above rather than written, so it cannot favor either vendor. Take these into both conversations and ask each side the same question.
- AI Governance and Bias Disclosure
- Good Law Verification
- Refusal and Uncertainty Behavior
- Bar Guidance Alignment
Which one fits
Choose IP Author if
- Your office action work spans more than prior art rejections. IP Author enumerates the statutory grounds it handles, covering 102 and 103 prior art rejections, 101 subject matter eligibility, 112 written description, enablement and definiteness, and double patenting, always producing a traversal strategy with structured arguments and claim mapping generated from the application history, with workflows stated to align with USPTO and EPO practice.
- You need to send a search result to a client without sending everything. IP Author's prior art search runs from the disclosure inside the drafting flow and returns similarity scores, novelty summaries and explanations of overlaps and differences, and exports as client ready Word or Excel reports including masked versions that withhold filing dates and raw excerpts.
- You want to try it before you talk to anyone. IP Author publishes a pricing page in its own navigation and offers a fourteen day free trial startable directly from the website, with its FAQ confirming the trial does not run through sales, which is a more open commercial posture than XLSCOUT's pricing page, which ends in a sales form.
Choose XLSCOUT if
- You want the training and disclosure positions in a contract. XLSCOUT's master agreement undertakes at clause 6.4 that customer inputs and outputs are not used to train, retain or improve its models or those of its third party suppliers, and at clause 7.3 requires prompt written notice before a compelled disclosure, unless legally prohibited, expressly so the customer can seek an order preventing it.
- Your security review wants a named auditor and an isolation model. XLSCOUT publishes completion of an AICPA SOC 2 Type II audit naming Prescient Assurance as the auditor with an unqualified opinion, describes each customer environment as sitting in a dedicated trust zone with unique encryption keys separating customers from each other and from XLSCOUT, and lets a customer choose United States or European storage.
- You want to inspect part of the stack yourself. XLSCOUT built ParaEmbed, the embedding model behind its retrieval, fine tuned it on patent and technology data and published it openly on Hugging Face, so a technical buyer can evaluate the retrieval model directly rather than take a description of it, alongside nine separately named modules spanning search, invalidity, claim charting, standard essential patent mapping and drafting.
In summary
IP Author
IP Author is an AI assisted patent platform for practitioners covering drafting, prior art search, office action response, claim drafting, invention management and classification. It generates a complete first draft from an invention disclosure, with claims, specification and figures kept consistent, and reads images, tables and diagrams inside uploaded disclosures rather than the text alone. According to the AI Legal Index, AI is the product itself, and its office action support is enumerated by statutory ground across sections 101, 102, 103 and 112 and double patenting. It names no customer, and it states no training or retention position or security certification outside its trust center and privacy policy, which this page does not draw on.
XLSCOUT
XLSCOUT is a large language model platform for patent work, sold to IP departments, patent attorneys and R&D teams as nine separately marketed modules. They span novelty and invalidity search, claim charting, licensing and standard essential patent mapping, technology monitoring, drafting and ideation. Retrieval is built on ParaEmbed, an embedding model the company fine tuned on patent data and published openly. According to the AI Legal Index, XLSCOUT's case rests on its contract and its audit: the master agreement undertakes that customer inputs and outputs are not used to train, retain or improve its models or its suppliers', and it names Prescient Assurance as its SOC 2 Type II auditor. It publishes no AI governance position, no integration into a firm's systems and no price.
Questions buyers ask
IP Author vs XLSCOUT: which is better for patent drafting?
The difference between them is documentary rather than functional. XLSCOUT publishes a master agreement carrying training, notice, indemnity and cap terms, and names its SOC 2 auditor. IP Author publishes very little of that but is more specific about the work, enumerating the statutory grounds its office action module handles and offering a self serve trial. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
Do either say whether invention disclosures train the models?
XLSCOUT answers in its agreement. Clause 6.4 undertakes that the data, defined to include both inputs and generated outputs, is not used to train, retain or improve the models of XLSCOUT or its third party suppliers, and its security page and FAQ say the same. IP Author makes no statement either way outside its trust center and privacy policy, which this page does not draw on, so the question is open rather than answered unfavorably. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
What can you buy without talking to sales?
IP Author. It publishes a pricing page in its navigation and a fourteen day free trial that starts from the website without a sales conversation. XLSCOUT also publishes a page titled pricing, and that page contains no rate, price range, floor, currency, unit of charge or tier name, offering instead a form whose own text directs the reader to submit it so the sales team can provide details. Neither vendor publishes a figure a buyer could budget against. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
Which one publishes a security attestation?
XLSCOUT does. It states completion of an AICPA SOC 2 Type II audit, names Prescient Assurance as the auditor and reports an unqualified opinion, which is more than a badge. No examination period or report date is published, so currency cannot be established, and its ISO 27001 reference is qualified as applying to its data center partners rather than to XLSCOUT. IP Author links a trust center from its site navigation, which this page does not draw on, and names no certification elsewhere. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
What do IP Author and XLSCOUT both leave unpublished?
Neither publishes an AI governance position: no policy, no accountable owner, no pre release testing regime and nothing on uneven output. Neither names an integration into a docketing system, an IP management platform or a document management system, so both stand apart from the systems a prosecution practice runs on. Neither states whether the legal status of a retrieved reference is checked. Neither documents what happens when the system cannot support a result. And neither names a bar or patent office conduct authority. From the AI Legal Index, based on each vendor's own published materials as of September 3, 2026. No vendor pays for placement.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything on it comes from public material on the dates shown. How the index grades.
IP Author publishes a trust center and a privacy policy, and this page does not draw on their contents. Its training, retention and residency positions may be answered there, so they are open questions here rather than established absences. On XLSCOUT, two published surfaces describe retention differently. The security page runs a heading reading zero data retention or training and states that the model retains no user data. Clause 6.1 of the master agreement grants a license to collect, process and store inputs and outputs in line with the content security policy. Both can hold, since a model not retaining a prompt differs from a platform storing a project, and the agreement is the binding document. No retention period appears in either. Neither vendor reviewed this page.