DorothyAI vs IPRally: how they compare in 2026
DorothyAI and IPRally both run AI prior art searches from a plain description of an invention rather than Boolean queries, and both extend into freedom to operate and invalidity work. IPRally sits in the top two bands on eleven of fifteen axes and DorothyAI on two of fifteen, identical on four. The gap is almost all published commitment. IPRally's generative AI terms name Anthropic, OpenAI and Google, require each to keep customer input out of training, and keep content inside its own cloud unless the customer approves. It hosts in the EU, lists its subprocessors, states ISO 27001, and names customers, including a technology transfer office reporting searches cut by about half. DorothyAI publishes nothing on what happens to a search, although a novelty query describes an invention before it is filed. Its counterweight is breadth of data and method. It queries several search models at once and keeps results more than one agrees on, and its reports add trademarks, SEC filings, litigation and market data.
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.
The search models are the product. DorothyAI's own description is a suite built on artificial-intelligence search: Activ8 Novelty takes a plain-English description of an invention instead of keywords and Boolean operators, Activ8 Freedom reads the claims of granted patents for a product and its components, AskDot is a generative assistant over the patent database, and Adaptive Research Technology runs combinations of natural language processing, machine learning and legacy search models at once and surfaces results that more than one model agrees on. Take those away and what is left is a patent collection with no way into it that the vendor sells. Verified 20 September 2026.
The machine learning is the mechanism the buyer pays for. IPRally's proprietary Graph AI, described on the features page as trained by millions of patent examiner citations, is what executes the search: a user enters a free-text description, a publication number, an image or a hand-drawn search graph, and the model returns ranked prior art without Boolean construction. The company's own positioning is that this replaces query building rather than accelerating it, and the Agent page states the case directly against alternatives, that generic large language models cannot search the patent index and that an agent is only as good as the search technology underneath it. Remove the models and what remains is a full-text patent database with the Boolean search and filters IPRally ships as a secondary, industry-standard option, which is not the product anyone buys. This is the AI-native counterpart to Alt Legal and Anaqua in the same lane, both of which graded C, and the contrast is the axis discriminating rather than an unusually generous read.
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.
The search method is described and nothing about it is measured. Adaptive Research Technology is set out clearly on the products page: several models optimized for different subject matter are queried simultaneously and results confirmed by more than one model are treated as the good ones, which the vendor contrasts with tuning a single model. Results are patent documents a reader can open. What is claimed and never tested is that this returns the best results every time and removes inconsistency across subject matter: no recall or precision figure, no benchmark, no test set and no error analysis is published. Nothing describes how AskDot's generated answers tie back to the patents behind them, which is the part of the suite where a wrong answer is hardest to see. Verified 20 September 2026.
Grounding is real, documented and inspectable, and the measurement stops short of a test an outsider could run. The Agent page states that every search case can be opened and inspected, that AI reasoning and citations are given at the feature level, and that every claim assessment, feature mapping and relevance score is traceable to a source document in IPRally's patent index. The retrieval method is described rather than asserted: semantic search over graph representations, patent families, and following the examiner citation trail outward from the strongest documents. One measured figure is published, a 14 percent improvement in search recall against a full patent text single search, but the comparison is to IPRally's own baseline, no test set is described, and no failure mode is named. The Generative AI Additional Terms of 14 March 2025 disclaim any warranty that outputs are correct and recommend checking them against a primary source, which is an honest position rather than an accuracy claim. Two published corpus figures also disagree, 120 million patents on the pricing page against 130 million on the Agent page. The citator limb does not bite here, since the product retrieves patent documents rather than legal authority whose subsequent history could be checked.
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.
Human involvement is the pitch and the control structure is not described. The People+AI page puts more than 200 advisors alongside the tools and says the expert team runs the search and delivers the result, which describes a staffed service rather than a review step inside the platform. For the software a firm licenses, nothing published says what runs unattended, what a user must check before relying on a novelty or freedom-to-operate result, or what happens when the models disagree beyond the statement that agreement between models is how good results are picked. CrossCite is said to give control over references and intelligent filtering when building an Information Disclosure Statement, which is the nearest thing to a described review point. Verified 20 September 2026.
The control structure is unusually well described for this corpus and two limbs are missing. What runs unattended is stated: Agent analyses an uploaded disclosure, identifies claims and features, searches and builds the feature chart. Where it stops is also stated, since the page says Agent verifies its findings with the user before searching, and the depth is a user setting described as quick triage or litigation-grade digging. The review surface is concrete, with every search case openable, reasoning and citations at feature level, and output handing off into a full IPRally project so a professional continues the analysis, framed as designed for professional review, not blind trust. The Generative AI Additional Terms add a contractual layer of control: generative features are always marked with an identifier before use, an administrator enables them for the organization, and the customer can disable them at any time. What is not published is any statement of what happens after the system is wrong, and no abstention path is described for Ask AI, which is why this sits below the top band rather than at it.
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.
Three case studies are published and not one names anybody. The case studies page describes a leading technology firm that used an Activ8 Alignment report to design around a freedom-to-operate risk and then moved to quarterly monitoring, a prominent law firm that built an international filing strategy for an oil and gas client, and a mid-sized law firm that added patent landscapes and competitive intelligence to its practice. No customer is identified, no date is given, and no figure appears in any of them; the homepage testimonials carry photographs with no names or firms. What a buyer can take from the record is a description of how the reports are used rather than evidence of what they achieved. Verified 20 September 2026.
There is a named customer with a figure, and no date and no method behind it. The TLB case study identifies TLB GmbH, the central technology transfer office for universities in Baden-Wurttemberg, quotes Dr Frank Schlotter, Authorized Officer, by name and title, and reports an approximate 50 percent reduction in time spent on patent searches, with the specific features credited, Smart search for query preparation and Ask AI during review. It is downloadable as a PDF and describes what the organization did before, extensive keyword planning and manual Boolean queries plus a different AI tool they were not satisfied with. Five further case studies are published under named organizations including Unilever, RPX Corporation, Ossur, Metsa Group and Perl IP Consulting, and a testimonial on the Agent page names Josh Walling, Senior Patent Agent at Milwaukee Tool. What holds this below the top band is that no case study carries a date, and the headline figure is the customer's own estimate of time saved rather than a measurement a reader could assess. Only the TLB study was opened on 1 September 2026.
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.
Nothing published says what happens to a search. This matters more here than for most products: a freedom-to-operate query describes an unreleased product, and a novelty search describes an invention before it is filed, so the queries themselves are among the most sensitive things an IP practice holds. The Terms of Service, which govern the platform, address accounts, subscriptions, prohibited uses and intellectual property in the site's own content, and say nothing about the confidentiality of what a user enters or the results returned. The Privacy Policy covers personal information collected through the websites and does not reach search content. No statement on training, segregation between customers, privilege or work product was located. Checked the homepage, products, People+AI and case studies pages, the Terms of Service and the Privacy Policy on 20 September 2026. Verified 20 September 2026.
The commitments a buyer most wants are in the agreement, and the segregation question is unanswered. The Terms of Service of 5 April 2024 state that IPRally will not gain ownership of user-generated content, will not share, distribute or disclose it to any third party, and will not use it for any purpose other than performing the functions of the Service, and give the customer a right to permanently delete all of it. The Generative AI Additional Terms go further than anything else located in this pull on the third-party model question, requiring that any external provider must not use input or output to train its models and must not claim ownership, and stating that user-generated content is never sent outside IPRally's own cloud environment unless the customer specifically approves it. Against that, privilege and work product are never mentioned on any surface read, and nothing addresses segregation between users or between matters inside a customer account beyond company-specific encryption keys and user-level feature rights. For an in-house IP department that is close to sufficient under the band as amended, but IPRally also sells to law firms, and for that segment the matter-level question is not answered.
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.
The audience is left open at exactly the point where it matters. The products page addresses patent attorneys and their clients in the same breath and frames Activ8 Novelty around the first question any inventor, scientist, entrepreneur or business person asks, is an invention patentable, which is a legal conclusion when a non-lawyer acts on it. CrossCite is said to be tailored to meet the latest USPTO guidance on AI tools for Information Disclosure Statement submissions, which is the one place the estate engages with a practitioner duty. No statement that the output is not legal advice, no description of the user's own review duty, and no jurisdiction limit was located on the homepage, the products, People+AI or case studies pages, the Terms of Service or the Privacy Policy on 20 September 2026. Verified 20 September 2026.
What is published is a position on AI output reliability, not a position on the professional responsibility line, and the two are being counted separately here. The Generative AI Additional Terms tell users they are solely responsible for content generated, that they should independently verify it, and that they assume the entire risk, and the Agent page repeats that Agent does the searching while the user makes the decisions. Nothing located states that IPRally is not a law firm or does not provide legal advice, names a jurisdiction limit, addresses a practitioner's supervision or competence duties, or engages with ABA Formal Opinion 512, USPTO practitioner guidance or any EPO or national equivalent. This matters more than usual because the Agent page markets a should-we-file signal to R&D and innovation teams explicitly described as needing no patent expertise, which puts a patentability assessment in front of non-practitioners with only a general verification disclaimer behind it. Searched the home page, the features page, the Agent page, the pricing page, the legal hub carrying all four agreements, the about page and the PR page on 1 September 2026.
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.
No governance position is published for a system whose results decide whether an invention is filed or a product is launched. Adaptive Research Technology is described as a design choice about how search models are combined, not as a governance regime: nobody inside the company is named as accountable for model behavior, nothing describes what is tested before a change ships, no evaluation is published, and nothing addresses whether results vary in quality by technology area, language or jurisdiction, which the vendor itself says is the weakness of single-model platforms. Checked the homepage, the products page, People+AI, the case studies, the blog index, the Terms of Service and the Privacy Policy on 20 September 2026. Verified 20 September 2026.
Principles are published and no mechanism sits behind them. The about page states three company values, transparency defined as clear, explainable AI solutions so a user always knows how decisions are made, simplicity and reliability, and the features page repeats explainability as a product property. That is an AI-specific principle rather than a generic value set, which keeps this off the bottom band. What is absent is everything that would make it auditable: no responsible AI or AI governance page exists, no ISO 42001 or EU AI Act position was located, nobody inside IPRally is named as accountable for model behavior, no pre-release testing regime is described, and no evaluation of uneven output has been published for a system that classifies technology and scores relevance. Searched the home page, features, about, the legal hub and the help center collection list on 1 September 2026. The identifier marking and customer opt-out for generative features are real governance mechanisms but they are spent on the oversight row and are not counted again here.
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.
Nothing published states how long searches, uploads or reports are kept, who inside the vendor can reach them, or what is deleted when a subscription ends. The Privacy Policy is written for the websites: it lists cookies, log files and analytics in detail, and its headings for international transfers, access and correction and commitment to security carry no text at all on the page as served. The Terms of Service address billing, accounts and prohibited uses. No subprocessor list, encryption statement or incident practice was located. Checked the homepage, products, People+AI and case studies pages, the Terms of Service and the Privacy Policy on 20 September 2026. Verified 20 September 2026.
Every limb this band names is published and specific. Retention is stated with a number in the privacy policy updated 12 December 2025, which keeps personal data for a maximum of twelve months plus an ordinary backup period after the reason for holding it ends. Deletion is contractual on both sides, with the Terms of Service giving the customer a right to permanently delete all user-generated content and the Data Processing Agreement requiring deletion or return of personal data on termination. Access control is described as restricted to authorized personnel bound by confidentiality obligations, with SSO, optional adaptive multi-factor authentication, user-level feature rights and company-specific encryption keys. The subprocessor list published as DPA Annex B names nine processors, Okta, Google Cloud EMEA, HubSpot, Intercom, Planhat, Mailgun, Zapier, Slack and Pendo, each with its address, purpose, location of processing and the data categories involved. Incident practice is covered in DPA clause 9, requiring notice without undue delay and specifying the four contents of that notice, alongside a published Vulnerability Disclosure Policy last modified 17 February 2026 and a public system status page. The one soft edge is that the retention position is framed around personal data and no separate window is given for patent content or AI inputs; the subprocessor list is dated January 2025, which is noted rather than graded.
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.
The published position is a standard disclaimer and a cap set at what the customer paid. The Terms of Service provide the services as is and as available, disclaim all warranties including accuracy, reliability and fitness, and state that if liability is found it is limited to the amount paid for the products or services, with no consequential or punitive damages. No indemnity runs from the vendor, no warranty attaches to search results, and no insurance position appears. For a product whose output supports a filing or launch decision, nothing addresses what happens if a material reference is missed. Verified 20 September 2026.
The allocation of loss is published, readable before signing, and runs entirely against the buyer. The Terms of Service cap liability at the amount paid in the twelve months before the event, and state expressly that IPRally is not responsible for damage caused by either correct or potentially incorrect data in the Service, which is the exposure this product actually creates. The Generative AI Additional Terms reinforce it, providing outputs as is with no warranty of correctness, placing the entire risk on the user, and disclaiming liability for the output of external model services and for security failures at third-party providers. The Data Processing Agreement carries the same cap and does name carve-outs, excluding wilful misconduct and gross negligence from the limitation. No indemnity of any kind runs to the customer, no warranty attaches to output, and no insurance position is stated. This is a C rather than a D because a buyer can read the whole allocation before signing and the carve-outs are named, which is more than the two ip-and-patents vendors already in the index offer, neither of which publishes a customer agreement at all.
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.
No integration into the systems a patent practice runs was located. Nothing names a docketing or IP management system, a document management system, an office add-in or an API, and CrossCite, which prepares Information Disclosure Statements, is described as generating the statements without saying how they reach a prosecution system or the USPTO. The platform is reached through its own web application. Checked the homepage, the products page, People+AI, the case studies, the Terms of Service and the Privacy Policy on 20 September 2026. Verified 20 September 2026.
The product largely stands alone and the connective tissue that exists is thin. What is documented is export rather than integration: custom exporting to Excel, PDF and Word, sharing of searches, collections and monitorings inside the account, single sign-on as an optional module, and a reference on the features page to unspecified tool interoperability. No connection to any IP management or docketing system, document management system, or patent office filing system was located, and there is no integrations page anywhere in the site navigation. A first-party blog post dated 23 July 2026 describes a new IPRally API as shipped and MCP support as upcoming, and the future-tense half of that is not evidence of anything today. The help center carries a collection titled product news, data coverage, interoperability, holding three articles, which was not opened on 1 September 2026, so this grade is rebuttable if that collection contains a usable integration reference. Checked the home page, features, pricing, the full product navigation, the legal hub and the help center collection list.
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 published states where the software runs or where a customer's searches are stored. The platform is delivered over the web at search.dorothyai.com; no tenancy model is described, no hosting provider or region is named, and the Privacy Policy's international transfers heading carries no text on the page as served. A practice with client-imposed residency requirements would have to establish all of it in a sales conversation. Checked the homepage, products, People+AI and case studies pages, the Terms of Service and the Privacy Policy on 20 September 2026. Verified 20 September 2026.
Residency is answered plainly and the tenancy model is not. The privacy policy states that the servers directly used by IPRally are located in the European Union, and the subprocessor annex names Google Cloud EMEA Limited of Dublin as the main cloud provider with location of processing given as the EU, and gives a processing location for each of the other eight subprocessors individually. The Data Processing Agreement clause 4 prohibits transfer of personal data outside the EU or EEA without the customer's prior written authorization and requires standard contractual clauses where a transfer occurs. Unusually, the processing question is answered for the AI specifically: the Generative AI Additional Terms distinguish In-Cloud Services running inside IPRally's own cloud environment from External Services reached by API outside it, state that core models are hosted by IPRally itself, and commit that user-generated content is not sent outside that environment without customer approval. That is the precise limb that held Anaqua at B, answered here. What is missing is any statement of the tenancy model, single or multi-tenant, and there is no choice of region, since the EU is the only location offered.
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.
No independent security attestation was located. There is no security page and no trust center, no SOC 2, ISO 27001 or comparable standard is claimed anywhere, no auditor or penetration test is mentioned, and the Privacy Policy's own commitment to security heading carries no text on the page as served. Checked the homepage, the products page, People+AI, the case studies, the blog index, the Terms of Service and the Privacy Policy on 20 September 2026. Verified 20 September 2026.
The certification is real and stated twice on the property, on the home page and the features page, as ISO 27001 certification of the information security management system. What is not published anywhere readable is the scope of that certification, the certificate date or period, or the name of the auditor, so a buyer in 2026 cannot confirm the attestation is current. A trust center exists and is linked from both the home page and the site footer at trust.iprally.com; it is hosted on Vanta, which its own page metadata confirms, and it returned a page frame with no body across the attempts made on 1 September 2026, with a search built on the portal's own terms failing to surface its contents. That block is a retrieval limit and nothing is graded against the vendor for it: this row rests on what is published on the readable surfaces, where a named standard appears without scope, date or a route to the report. It is what the portal might additionally offer that could not be confirmed, so B is a floor here rather than a ceiling.
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 and the models are not identified. Adaptive Research Technology is presented as the vendor's proprietary framework for querying combinations of natural language processing, machine learning and legacy search models optimized for particular subject matter, and AskDot is called a generative AI tool over the patent database. No model, version or provider is named, nothing states whether the generative layer runs on a third party's service or the vendor's own, and no commitment to notify customers when any of it changes was located. Verified 20 September 2026.
This is the most complete model supply chain disclosure located in the pull, and it sits in a contract rather than on a marketing page. The Generative AI Additional Terms of 14 March 2025 name the third-party providers IPRally may use, Anthropic PBC and OpenAI LLC as External Services reached through their APIs, and Google LLC as an In-Cloud Service, specifying Gemini models and all models in Model Garden meeting stated requirements, with links to each provider's terms. Where they run is stated and the distinction is defined: External Services host models outside IPRally's cloud environment, In-Cloud Services run within it, and the features page adds that IPRally hosts its core AI models itself. The document also sets the minimum requirements a provider must meet to be used at all, no training on input or output and no ownership claim over either. The change route is the limb a strict reader may want to test: providers may be added or removed at IPRally's sole discretion, but the list itself sits inside a published agreement whose amendments are posted, with material changes notified by email or at login under the Terms of Service, so a customer can see the list move rather than being told separately that it has.
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.
The shape of the commercial offer is visible and the numbers are not. The Terms of Service state that parts of the service are sold on subscription, billed in advance on a monthly cycle and renewing automatically, that a free trial may be offered and converts to a paid subscription unless canceled, that fees are non-refundable, and that fee changes take effect at the end of a billing cycle. The products are separately named and split by function, the products page invites inquiries about Alignment Bundles, and the People+AI page offers an instant quote form for patentability, freedom-to-operate and invalidity searches. No rate, tier price or unit of charge appears anywhere, and every route ends at a demo booking or a quote request. Verified 20 September 2026.
The unit and the structure are published without the figure, which is the same shape graded B on Juro and SpotDraft. Two plans are named, Individual for a single user and Team for two to unlimited users, so the unit of charge is explicit, and the comparison table below them is the most detailed in this pull, running to roughly sixty line items across search, Boolean, review, AI assistants, monitoring, collections, collaboration, classification, coverage, security and support, each marked as included, optional or coming soon. Which components are chargeable extras is named rather than implied, covering custom taxonomy AI classifiers, AI classification of monitoring results, simplified R&D licenses, SSO and adaptive multi-factor authentication. A three-day free trial is self-serve. No number appears anywhere: there is no rate, no band, no floor and no currency, the Team plan routes to Contact Sales, and the Agent product is described separately as value-based and agreed per customer. This sits at the lower end of the band, since unlike Juro no volume bands or currencies are given and nothing states what implementation adds.
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 data behind the product is described with unusual specificity and the limits are not. The published answer to what sources the search tools use names thirteen kinds: global patents, trademarks, SEC filings, licensing, agreements, assignments, litigation, financial data, market data by product category and supply chain information. A Jurisdictional Realities Database is described as weighing more than thirty factors per jurisdiction, covering innovation ecosystems, government policy and judicial systems, to inform filing strategy. The buyers addressed run from patent attorneys and law firms to in-house teams, entrepreneurs and inventors. What is not stated is coverage by jurisdiction or date range for the patent collection itself, how current it is, or which technology areas the models handle less well. Verified 20 September 2026.
Coverage is described with real substance and the outer boundary is left open. Five role-based solution pages address the patent searcher, patent analyst, patent manager, patent attorney and head of IP, and the Agent page segments its audience into IP leaders, patent professionals, R&D and innovation teams, and law firms, with a paragraph explaining what each gets. Five use cases are documented as supported, novelty and patentability, invalidity, state of the art, freedom to operate and classification. Corpus coverage is quantified as 58 jurisdictions and 120 million or more patents with machine translations, and the customer roster demonstrates the range, from corporate IP departments including Unilever, Bosch and Nestle through a public authority in the Danish Patent Office to tech transfer offices such as TLB and firms such as Perl IP Consulting and Laine IP. Some limits are stated at feature level, with Agent supporting novelty and invalidity while freedom to operate is on the roadmap and Boolean monitoring marked coming soon. What is not stated is where the product stops as a matter of coverage, with no statement on team size, portfolio size, or the fact that the platform addresses patents rather than the wider IP estate.
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?
The published Terms of Service govern the platform and never address training either way. They reserve the vendor's rights in its own content while excluding content provided by users from that reservation, and they take a broad license over feedback only. No improvement or model-training right over user searches is granted anywhere, and no policy page states a position, so a firm has nothing to hold the vendor to in either direction.
The prohibition sits in the agreements rather than on a policy page, and it runs in two directions. The Terms of Service of 5 April 2024 limit IPRally itself to using user-generated content only as required to perform the functions of the Service, and state it will not gain ownership of that content or disclose it to any third party, which forecloses training without using the word. The Generative AI Additional Terms of 14 March 2025 address third-party models expressly, setting it as a minimum requirement that any external or in-cloud provider does not use input content, prompts or output to train its machine learning models and does not claim ownership of either.
The home page adds that IPRally's own AI is trained exclusively on existing patent data. The quote recorded is the provider-side requirement, which is the explicit training language; the IPRally-side commitment is the purpose limitation in the Terms.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Checked the homepage, the products page, People+AI, the case studies, the Terms of Service and the Privacy Policy on 20 September 2026. No retention period is published for search queries, invention descriptions, uploaded material or generated reports, and no deletion commitment applies when a subscription ends; the Privacy Policy covers personal information collected through the websites and does not reach search content.
No retention period for prompts or outputs is published, and no zero-retention option is stated, but the customer holds a contractual deletion right over the material. The Terms of Service allow the customer to permanently delete all user-generated content, subject to ordinary backup storage periods, and the Generative AI Additional Terms define prompts as built from user input forming part of that same user-generated content, so the right reaches what a user typed into Ask AI or a search field.
The privacy policy's twelve-month retention figure governs personal data rather than patent content or AI inputs. The control is a delete-on-demand right in the agreement rather than a configurable retention window in the product.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
Checked the same pages on 20 September 2026. Nothing addresses separation between users, matters or customers, and nothing describes whether one practitioner's searches are visible to colleagues sharing an account. The question has weight for a product where a single query can reveal a client's unfiled invention.
IPRally operates its own permission model rather than inheriting one, and the documented level is the user and the company rather than the matter. The pricing comparison names user-level feature rights management, single sign-on, optional adaptive multi-factor authentication and company-specific encryption keys, and the Terms of Service describe admin users setting license counts and enabling or disabling features for some or all users.
Sharing of searches, collections and monitorings is an explicit act rather than a default. The classic ethical wall question bites less here than on a document-based product, because retrieval runs over a public patent corpus rather than over the firm's own files; what is not addressed is segregation between matters or teams inside a single customer account.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
The Privacy Policy says information may be disclosed in response to subpoenas, court orders and other lawful requests by regulators or law enforcement, and makes no commitment to notify anyone. It is written for the websites rather than for search content, and the Terms of Service do not address legal process at all.
The commitment is present but narrower than the question. Clause 3(A) of the Data Processing Agreement requires IPRally to process personal data only on the customer's written instructions unless compelled by law, and in that case to inform the customer of the legal requirement before processing, unless the law prohibits telling them. That is a notice-before-disclosure commitment in the standard GDPR form. Its scope is personal data under the DPA; no equivalent clause covering user-generated content or search history was located in the Terms of Service, and the privacy policy separately reserves the right to disclose personal data to authorities under court order or subpoena without stating a notice duty. No transparency report was located 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 published answer to what the search tools draw on lists thirteen kinds of data: global patents, trademarks, SEC filings, licensing, agreements, assignments, litigation, financial data, market data by product category and supply chain information. They are described by class rather than by provider, no license or rights basis is given for any of them, and no update cadence for the patent collection is published.
The product does not retrieve case law; its corpus is patent literature and it is identified by scale and by jurisdiction rather than by license. The pricing page states coverage of 58 jurisdictions and 120 million or more patents with machine translations of non-English documents and downloadable PDFs, while the Agent page gives 130 million or more for the same index, a discrepancy noted rather than resolved. The training signal behind the ranking is named specifically as millions of patent examiner citations.
The Terms of Service name Google Patents and Espacenet as third-party resources the Service may link to. No license, rights basis or update cadence for the underlying patent data is stated anywhere located on 1 September 2026.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
For a patent product this asks whether the tool tells a reader that a patent is still in force. Checked the homepage, the products page, People+AI, the case studies, the Terms of Service and the Privacy Policy on 20 September 2026. Nothing describes a legal status, lapse or expiry determination on the patents returned, although the alignment reports cover litigation, assignment and licensing data.
No citator applies and the row is recorded rather than skipped. The product retrieves and analyses patent documents rather than legal authority whose subsequent history could be checked, so there is no treatment signal to surface and no good-law question in the sense this signal asks. The nearest analog is registration currency: the pricing comparison lists bibliographic data, legal statuses and full specifications among the review features, so a user can see the status of a patent, which is a record of the register rather than a judgment about whether an authority still stands. Searched the features page, the pricing comparison, the Agent page and the use-case pages on 1 September 2026.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
Checked the homepage, the products page, People+AI, the case studies and the Terms of Service on 20 September 2026. Nothing describes what the tools do when a search finds nothing relevant or AskDot cannot answer from the patent database. The nearest published mechanism is that results confirmed by more than one search model are treated as the good ones, which ranks what was found rather than signaling that the answer may not be there.
A relevance signal is exposed and an abstention path is not. The pricing comparison names AI relevancy scores and Boolean relevancy scores as shipped features, and relevance-feedback search lets a user flag strong results to pull similar ones, so the product surfaces a graded confidence in each retrieved document. That is a retrieval relevance score rather than a measure of confidence in a generated answer. Nothing located describes what Ask AI or Multi-patent Ask AI does when it cannot support an answer from the documents, and no no-answer or low-certainty path is documented.
The Generative AI Additional Terms instruct the user to verify outputs against a primary source, which places the duty on the reader rather than describing a behavior of the system.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
Searched the AI Hallucination Cases database maintained by Damien Charlotin on 20 September 2026 on the product names DorothyAI and AskDot and the corporate name Three10 Legal. No court order, opinion or disciplinary record naming the product was located. This is a statement about the public record rather than a finding about the product.
Searched the AI Hallucination Cases database maintained by Damien Charlotin at HEC Paris, together with 2026 sanctions trackers and trade coverage, on 1 September 2026, on both the company name and the product name. No court order, opinion or disciplinary record naming IPRally or IPRally Technologies was located. This is a statement about the public record rather than a finding about the product. The failure mode this signal tracks also fits the product poorly: IPRally retrieves real patent documents from an index and its generative features summarize and analyze those documents, rather than producing citations to legal authority for filing.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
The products page says CrossCite is tailored to meet the latest USPTO guidance on AI tools for Information Disclosure Statement submissions, which is a reference to a practitioner duty without naming, dating or quoting the guidance, and the blog index lists a post on AI in patent matters before the USPTO. No ethics opinion, bar guidance or USPTO notice is identified anywhere the index could read.
No engagement with any professional responsibility guidance was located. Nothing addresses ABA Formal Opinion 512, USPTO guidance for practitioners on artificial intelligence, EPO or national patent attorney guidance, or any state bar opinion, and no ethics or professional responsibility page exists on the property. This is a European vendor selling to patent attorneys in multiple jurisdictions as well as to US firms, and none of the relevant regimes is named.
Searched the home page, the features page, the Agent page, the five solutions pages listed in the navigation, the legal hub carrying all four agreements, the about page and the PR and media page on 1 September 2026.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
The published case is speed and substitution: reports delivered in minutes rather than weeks, some within 24 hours, and the claim that the reports give associates the equivalent expertise of hiring additional analysts overnight. The buyers include law firms that bill this work to clients. Nothing addresses how the saving is reflected on a client's bill, or whether a report charge is passed through as a disbursement.
Time savings are the central marketing claim and the client side of that equation is never addressed. The TLB case study reports an approximate 50 percent reduction in search time, the Agent page is built on the promise of minutes rather than weeks, and it positions directly against the cost of outsourced search, contrasting reports that cost thousands with a predictable subscription. No guidance accompanies any of it on how a firm or a tech transfer office should bill for AI-assisted search work or disclose it to the client paying for the assessment. No per-matter record of AI-assisted work is described that a firm could interrogate for billing purposes.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
Checked the homepage, the products page, People+AI, the case studies, the Terms of Service and the Privacy Policy on 20 September 2026. No subprocessor list, model provider list or client-facing disclosure material was located, and there is no trust page or documented route to request any of it.
Both artifacts a client-side AI clause asks for are published, and neither requires an agreement to obtain. DPA Annex B lists nine named subprocessors with addresses, purposes, processing locations and data categories, dated 24 January 2025. Separately, the Generative AI Additional Terms name the model providers themselves, Anthropic PBC, OpenAI LLC and Google LLC, distinguish which run inside IPRally's cloud from which are reached externally, and state the minimum requirements each must satisfy, no training on input or output and no ownership claim.
The same document is forwardable to a client as disclosure material: it describes exactly what leaves the environment on an Ask AI query, gives a worked example, and records that an administrator can disable generative features entirely. This is the highest value on this signal recorded in the pull to date and the reasoning is set out in the build log for comparison against Litify and Anaqua.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Checked the homepage, the products page, People+AI, the case studies, the Terms of Service and the Privacy Policy on 20 September 2026. CrossCite produces primary and supplemental Information Disclosure Statements, which is a disclosure of references to the patent office rather than a record of AI use, and nothing records which model produced a result or who checked it.
Some of the record exists and no document-level export of it is described. Generative outputs are always marked in the interface with a specific identifier under the Generative AI Additional Terms, so a user can tell which material was AI-produced, and the Agent page states that every search case can be opened and inspected with AI reasoning and citations attached at feature level, which covers what was retrieved and why.
Relevance marking and commenting record who reviewed a document. What is not described is any export tying a particular passage to the model that produced it, since the identifier marks that generative AI was used rather than which of the three named providers ran, and no certification artifact is offered. The obligation fits prosecution and portfolio work imperfectly, but the row is recorded rather than skipped.
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.
- Good Law Verification
Which one fits
Choose DorothyAI if
- You want patent search tied to the business picture. DorothyAI's Activ8 Alignment combines prior art with market, regulatory and court data into a shareable report that refreshes on a schedule or on demand, drawing on trademarks, SEC filings, licensing, assignments and litigation data.
- You want results that more than one model agrees on. DorothyAI's Adaptive Research Technology queries several natural language, machine learning and legacy search models at once and treats results confirmed by more than one as the strongest.
- You prepare Information Disclosure Statements or want a search done for you. DorothyAI's CrossCite prepares primary and supplemental IDS filings, and its People+AI service offers patentability, freedom to operate and invalidity searches through a network of more than 200 advisors.
Choose IPRally if
- You must tell a client which models see its invention. IPRally's generative AI terms name Anthropic, OpenAI and Google, state which run inside its own cloud, require each to keep input and output out of training, and let an administrator switch generative features off.
- You want an agent that checks its reading before it searches. IPRally Agent reads an invention disclosure, identifies the claims and features, confirms them with you, searches, and returns a claim chart and report whose every mapping traces to a source document for professional review.
- Your security review needs residency and a processor list. IPRally runs its servers in the European Union, publishes nine named subprocessors with purpose and location, states ISO 27001, and commits to specified breach notice content in its data processing agreement.
In summary
DorothyAI
DorothyAI, published by Three10 Legal, Inc. of Pennsylvania, is a patent search and IP analysis platform for patent attorneys and the businesses they advise. Activ8 Novelty searches from a plain English invention description, Activ8 Freedom checks granted claims against a product, AskDot is a generative assistant over its patent database, Activ8 Alignment builds reports combining prior art with market and court data, and CrossCite prepares Information Disclosure Statements. The AI Legal Index grades it in the top two bands on two of fifteen capability axes, with an A on AI centrality. It also sells search reports staffed by advisors. As of 20 September 2026 the index located no confidentiality position for searches, security attestation, named model or price.
IPRally
IPRally, from IPRally Technologies Oy of Helsinki, is a patent search and analysis platform whose graph based AI, trained on millions of examiner citations, runs prior art searches across 120 million or more patents in 58 jurisdictions from text, a patent number or an image. IPRally Agent turns an invention disclosure into a claim chart and novelty or invalidity report for professional review. The AI Legal Index grades it in the top two bands on eleven of fifteen capability axes, with A grades on AI centrality, data stewardship and model supply chain disclosure. Its terms name Anthropic, OpenAI and Google and bar them from training on customer content, and it hosts in the EU. As of 1 September 2026 the index located no published price figure.
Questions buyers ask
DorothyAI vs IPRally: which is better for prior art search?
On published evidence IPRally sits in the top two bands on eleven of fifteen AI Legal Index capability axes and DorothyAI on two of fifteen, identical on four, because IPRally publishes its model providers, training terms, hosting, subprocessors and customers. DorothyAI publishes a broader mix of business data around its searches and offers staffed search reports. Firms that must vet a vendor before entering unfiled inventions have far more to read from IPRally.
Which AI models does IPRally use?
IPRally hosts its core graph based models itself. Its generative AI additional terms of 14 March 2025 name Anthropic and OpenAI as external services reached by API and Google's Gemini models as a service inside its own cloud, and require each provider not to train on input or output. Content is not sent outside IPRally's cloud without customer approval. DorothyAI names no model or provider. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 27, 2026. No vendor pays for placement.
What happens to a search entered into DorothyAI?
Nothing published says. DorothyAI's terms of service address accounts, billing and prohibited uses, and its privacy policy covers personal information collected through its websites, so neither reaches the invention descriptions or product details entered as searches. No training, retention, deletion or segregation position was located. IPRally's terms limit use of such content to running the service and let customers delete it. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 27, 2026. No vendor pays for placement.
How are DorothyAI and IPRally priced?
Neither publishes a figure. IPRally sells an Individual plan and a Team plan with unlimited users, lists about sixty features by plan with chargeable extras named, and offers a three day trial; its Agent is priced per customer. DorothyAI's terms describe monthly subscriptions billed in advance with a free trial, and its search reports are quoted. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 27, 2026. No vendor pays for placement.
What do DorothyAI and IPRally both leave unpublished?
An advice line and any stake in a wrong result. Neither states that its output is not legal advice, although both reach users outside patent practice, and neither indemnifies customers or warrants search results. Neither names a specific ethics opinion or patent office notice on AI, and neither describes what its generative assistant does when it cannot answer from the documents. Graded by AI Legal Index against 15 capability axes and 12 legal signals, including privilege handling and citation accuracy, from each vendor's own published materials, verified September 27, 2026. No vendor pays for placement.
Three readings to weigh. DorothyAI's privacy policy headings on international transfers and security carry no text on the page as served, and its terms say nothing about the confidentiality of searches, so its low grades record what is published. IPRally's terms disclaim responsibility for correct or incorrect data in its service and cap liability at a year's fees; those are published terms. IPRally gives its corpus as 120 million patents on one page and 130 million on another. DorothyAI was verified on 20 September 2026 and IPRally on 1 September 2026. Neither vendor reviewed this page.
Neither vendor paid for inclusion, placement or a grade, and neither reviewed this page before it published. Everything above comes from public material on the dates shown. How the index grades.