DeepIP vs Patent Bots: how they compare in 2026
DeepIP and Patent Bots meet in Microsoft Word but take opposite views of how much a model should write. DeepIP generates complete applications from an invention disclosure, with claims, specification and abstract ready for review. It drafts office action responses with case law citations and searches more than 120 million patents for prior art by meaning. It runs on Microsoft Azure in the United States or the EU and shapes drafts to the formal rules of major offices, including the USPTO and EPO. Patent Bots keeps generative drafting to sections and edits, through a chat assistant and a prompt library. The rest of its suite is built on rules and data: proofreading, examiner statistics, an art unit predictor, IDS and USPTO forms. Its generative features stay off until an administrator enables them, and its published prices start at $27 a month. DeepIP suits teams that want first drafts of whole applications across patent offices. Patent Bots suits United States prosecution teams that want checks, forms and examiner data around a lighter drafting assistant.
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 DeepIP module is model driven. Drafting turns an invention disclosure into claims, specification and abstract, and the review module checks claims for antecedent basis, dependency gaps and section 101 risk. The prosecution module analyzes rejections and drafts responses with case law citations. Patentability, prior art retrieval, freedom to operate and invalidity analysis map claims against more than 120 million patents and non patent literature, and drawing generation turns sketches and method steps into formal figures. DeepIP combines an OpenAI model hosted on Azure with proprietary models built for patent work. Without the models, the Word add in has nothing to do.
Machine learning drives two parts of the suite. Art Unit Predictor is a neural network trained on public USPTO data. It reads claims and ranks the five art units and three groups most likely to receive an application. It also highlights the words pulling it toward each one, weighted by how often they appear in that art unit's patents compared with patents generally. Generative AI drafts figure descriptions, claim summaries, claim clauses and flowcharts and answers questions in a chat that takes the application and supporting documents as context. Most of the suite runs without either. Proofreading checks follow rules, examiner statistics come from USPTO records, and the office action tools fill forms and shells from templates. The terms say most tools use no generative AI at all, and every generative feature is off until an administrator turns it on.
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.
DeepIP's office action module drafts responses with case law citations from its citation engine, and prior art retrieval ranks references by relevance with the citation shown for each. Freedom to operate and invalidity conclusions link to the claim or prior art text that supports them, and Modern Times Legal's founder describes quoted citations linked to their sources. Retrieval is semantic, across more than 120 million patents and non patent literature. DeepIP says its models go through testing for factual accuracy with patent expert feedback, but publishes no accuracy figure, test set or error rate.
Generative drafting and chat answer from the user's own application, claims and supporting documents, which are passed in as context, and the patent chat in PatentPlex draws on the full prosecution record and family claims. Statute and MPEP citations in Office Action Explorer link to their sources at the USPTO. No accuracy figure is given for drafted text, chat answers or art unit predictions, and drafted sections carry no pointer back to the passage they rest on. The terms state that the service is not warranted to be accurate or error free.
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.
DeepIP works as an assistant inside Microsoft Word. Drafts, review flags, office action arguments and freedom to operate opinions arrive in the document, where attorneys comment, edit and track changes. The review module marks issues inside the claim text, and freedom to operate and invalidity conclusions link to the claim or prior art text behind them. DeepIP offers trials only to practitioners who draft and prosecute patents, and its terms leave the customer solely responsible for reviewing and validating every output. Nothing states what a module may complete without an attorney, or where its agentic prior art retrieval stops and hands back to a person.
Generative drafting works section by section inside the attorney's own Word document, as a chat assistant and prompt library, rather than producing a whole application. Every generative feature is off by default and must be switched on by a subscription administrator. Proofreading returns findings for the attorney to fix, and the claim markup check compares amended claims against a prior version before filing. Patent Bots frames the drafting tools as support for attorney judgment. No limit is stated on what drafted text may be used for, and nothing describes what happens when a draft is wrong.
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.
Nixon & Vanderhye's Corey Hawse reports tasks that took 20 to 30 minutes now taking under five, and gains of 50 percent or more on the tasks where savings are sharpest. Schwegman, Lundberg & Woessner, with more than 100 practitioners, reports focused work rising from four to six hours a day to seven or eight. Philips saw about 20 percent better drafting and prosecution efficiency in a trial and now uses DeepIP for invention analysis, prior art, drafting and office actions. Dilworth IP chose DeepIP as its primary drafting system after two years of evaluating tools, and Wood IP and Greenberg Traurig are quoted by name. The figures are the firms' own estimates, and the case studies carry no dates. DeepIP reports more than 25,000 applications drafted and 40,000 matters assisted, and $40 million raised through a $25 million Series B announced on 3 March 2026.
Five firms appear in case studies, among them Young Basile Hanlon & MacFarlane, Edell, Shapiro & Finnan, Brake Hughes Bellermann, GTC Law Group and Pierce Atwood. Young Basile ranked third among large patent firms in the 2026 Patent Bots quality rankings of 709 firms, and Edell, Shapiro & Finnan's ranking in Tech Center 2400 rose from eighth in 2020 to second. Those rankings come from Patent Bots' own count of errors in issued patents, so the measure and the product share an owner. According to the vendor, 45 of the top 100 patent law firms use it and drafting tools save 6 to 12 hours per patent, without a stated method.
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.
All customer data counts as confidential information under DeepIP's terms, which recognize that it may include material covered by attorney client privilege and keep the confidentiality duty on it indefinitely. The terms bar training on customer data, require model providers to keep it confidential and apart from other customers, and set out its destruction after termination. DeepIP lists section 203 of the German Criminal Code, the professional secrecy provision, beside ISO 27001, SOC 2 Type II and GDPR. Users and organizations are kept separate, and no matter level wall within a firm is described.
Section 2 of the terms commits that neither Patent Bots nor AWS Bedrock uses customer data for model training or improvement, and that neither stores or logs prompts or completions. Prep & Pros Pro discards patent documents immediately after returning results. Section 9 treats any nonpublic information uploaded as confidential, used only to provide the service, with notice of misuse and prompt notice before any legally required disclosure. Section 8 says use of the service creates no attorney client relationship or privilege, and the treatment of privileged material is not otherwise addressed. Accounts may not be shared, but nothing describes separation between matters or clients inside a firm's subscription.
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.
Not a law firm and not an attorney, DeepIP gives no legal advice, its terms state. Outputs and templates are tools for operational support, not guaranteed accurate, and the customer stays solely responsible for reviewing them and for advising its own clients. No attorney client relationship arises with DeepIP or its staff, including staff with patent backgrounds and the experts who run product sessions. DeepIP offers trials and demos only to practitioners who draft and prosecute patents, and tells inventors to consult patent attorneys. Duties of competence and supervision, and limits by jurisdiction, are not addressed.
Registered patent practitioners are the buyers: individual plans are open only to practitioners at law firms, law firm prices follow the count of registered attorneys and agents, and companies buy for their in house IP teams. The drafting tools are described as support for attorney judgment, assisting with sections and edits rather than writing a whole application. Section 8 of the terms states that no attorney client relationship arises, and section 3 puts the content of patent documents on the user. Competence and supervision duties are not discussed, and no ethics guidance is engaged.
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.
DeepIP holds ISO/IEC 42001:2023 certification for its AI management system, listed on its trust center with a 2026 certificate available through an access request. The trust center's controls include board oversight briefings, control self assessments and tested continuity plans. DeepIP says its models go through testing and validation for factual accuracy, with a feedback loop to patent experts, but publishes no test method or result. No accountable owner for the models is named, and nothing is disclosed about whether output quality differs across technical fields or patent offices.
No AI policy, principles, testing method or accountable owner is published for the generative features or the art unit model. Controls are product facts rather than governance: generative features are off by default and switched on by an administrator, and model providers may not train on customer data. Nothing addresses uneven output across technology areas or art units.
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.
DeepIP encrypts data in transit with TLS 1.2 or later and at rest with AES 256, restricts access to production systems and databases, and requires multifactor authentication for remote access. Its trust center lists ten subprocessors with their roles and regions. The terms require zero data retention at the model provider and an opt out from provider abuse monitoring wherever the provider offers them. Customer data is stored only in the territories named in the order form and can be exported on request during the term. It is destroyed within 30 days of a final export, or after termination when no export is asked for within 8 days, with a destruction certificate on request. No incident notification commitment or breach timeline is published.
Patent documents are processed and discarded at once, with no content stored or logged, and generative calls run on AWS Bedrock with zero data retention. PatentLink stores only public patent data, on Google Cloud. Data is encrypted in transit and at rest, production access needs two factor authentication, employees pass background checks, and penetration tests run yearly with weekly vulnerability scans. The sub processor list, last updated 3 April 2026, names nine providers with location and purpose: Google, AWS, Cloudflare, Stripe, HubSpot, PostHog, Zoom, Calendly and Anthropic. Under the data processing addendum, a new sub processor gets 30 days' notice before it receives personal data, and a personal data breach is reported within 72 hours with set contents. Personal data is deleted on request, with stated limits for logs, backups and disputes.
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.
Under its April 2026 terms, DeepIP warrants professional performance and a service free from reproducible errors against its documentation, with repair, replacement or termination as the remedy. It defends and indemnifies customers against third party claims that the service infringes intellectual property or trade secrets. DeepIP may then procure rights, modify or replace the service, or end the agreement and refund unused prepaid fees. Liability is capped at the fees paid in the prior 12 months, with confidentiality, intellectual property and indemnity claims outside the cap and consequential damages excluded. DeepIP warrants cyber liability and errors and omissions insurance of at least $5,000,000 per occurrence. The terms disclaim the accuracy of AI output and leave its review to the customer, and free trials carry a $1,000 cap.
Section 3 of the terms has the user assume all risk and says Patent Bots holds no liability for errors or inaccuracies in processing patent documents or in its statistics. Section 15 excludes direct, indirect, consequential and other damages, and section 14 provides the service as is, disclaiming accuracy and reliability. No cap figure, indemnity, warranty or insurance runs to the customer. A firm with a separate signed contract may have other terms, which are not published.
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.
Microsoft Word is where DeepIP runs, so drafting, review, office action, freedom to operate and invalidity work happen in the document with comments and tracked changes. DeepIP integrates with docketing and IP management systems and offers a REST API for custom integrations. The prosecution module retrieves USPTO and EPO file documents in one click and indexes them with OCR. No docketing or IP management system is named, and what moves between DeepIP and those systems is not described. DeepIP reports 20 percent higher adoption and 40 percent higher usage for its embedded approach than for standalone tools.
Prep & Pros Pro runs as a Microsoft Word add in as well as on the web. It converts Word tracked changes into claim markup, generates assignments and declarations from firm Word templates, writes client reporting letters from firm templates, and produces flowcharts in PowerPoint and Visio. Office Action Explorer pulls the latest office action, pending claims and cited prior art from USPTO records by application number. IDS forms come out as PDF on the USPTO form or as Word and Excel files. Family trees draw on USPTO and EPO data. No docketing, document management or matter management connection is documented.
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.
DeepIP runs as SaaS by default, with an on premises data deployment available through the order form. Customer data is stored and processed only in the territories the order form names, and does not leave them without the customer's written consent. Its cloud hosts, Microsoft Azure and Google Cloud Platform, and its model provider, OpenAI, each run in the United States or the EU. Schwegman, Lundberg & Woessner chose DeepIP after a security review of its Azure deployment in the United States. No architecture or support terms are published for the on premises option.
Delivery is from the cloud, and section 2 of the terms places all processing and storage in the United States, with every server in the country. For Prep & Pros Pro the customer can choose to have documents processed on Google Cloud Platform or Amazon Web Services, under each provider's own terms. Generative features run on AWS Bedrock, and PatentLink data is stored on Google Cloud. No region outside the United States, private tenancy or on premises option is offered, and the tenancy model is not described.
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.
DeepIP's trust center lists ISO 27001:2022, SOC 2, ISO/IEC 42001:2023 and GDPR. It offers the 2026 ISO 27001 and ISO 42001 certificates, the 2026 SOC 2 Type 2 report and a penetration test through an access request, and shows an engagement letter openly. Its control list covers production access, multifactor authentication, network segmentation, firewall review, background checks and tested continuity plans. DeepIP's security program follows NIST 800-53 Moderate and the NIST Cybersecurity Framework, with outside penetration tests and vulnerability assessments. Auditor names and examination periods sit inside the reports, which open after the access request.
A SOC 2 Type II report from Prescient Assurance, issued under AICPA standards with an unqualified opinion, covers security, availability, processing integrity, confidentiality and privacy. Audit reports, policies and the full control list sit in a trust center behind a bot check that did not load for a scripted browser, so how a buyer obtains the report is unknown. No audit period is stated.
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.
DeepIP uses an OpenAI model hosted on Microsoft Azure, sends no data to OpenAI's own infrastructure, and has Microsoft's abuse monitoring switched off under its Azure agreement. Its terms name OpenAI, Azure OpenAI, Google Cloud Vertex AI, Amazon Bedrock and Azure AI Foundry among the providers it may use. Each must contractually bar training on customer data and keep that data confidential. OpenAI, the language model provider, runs in the United States or the EU. No model or version is named for any module, and the terms commit to no notice when providers change.
Generative features run through AWS Bedrock, and the sub processor list names Anthropic's large language models for processing document content, so the provider and the platform are both named. Which Anthropic models are used is not stated. The data processing addendum promises 30 days' notice before a new sub processor receives personal data. The art unit model is Patent Bots' own neural network trained on public USPTO data.
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.
DeepIP sells subscriptions by the seat, on monthly or annual terms that renew automatically, and its terms put the fee on the subscription page or an order form. A self service trial opens with a card on file and no charge before it ends. Price rises at renewal need 45 days' notice on annual plans and 20 days on monthly ones, and prepaid fees are not refunded on early exit. No seat price, tier or volume rate is published.
Prices are published for every buyer type. An individual practitioner at a law firm pays $702 a year for Prep & Pros Pro, without the office action admin tools. Law firm prices follow the count of registered practitioners, on monthly or annual terms. For a solo firm the same product costs $65 a month, rising to $1,240 a month for 31 to 50 practitioners and $3,700 a month above 200, with Art Unit Predictor, PatentPlex and premium examiner statistics priced alongside. Company packages follow the number of patents issued to the company in the United States, with yearly minimums of $2,999 for Basic, $5,999 for Core and $9,999 for Premium. Onboarding needs carry their own minimums: $9,999 for a security review, $19,999 for single sign on and $39,999 for a custom contract. A 14 day free trial needs no onboarding, prices change with at least a month's notice, and payments are not refundable.
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.
DeepIP serves law firms and in house patent teams, and limits trials and demos to practitioners who draft and prosecute patents; inventors are not offered access. It shapes drafts to the formal rules of the USPTO, EPO, CIPO, CNIPA, DPMA, INPI, IPO, JPO, KIPO and UKIPO, and to PCT filings. Modules run from invention disclosure and patentability through drafting, drawings, review and office actions to freedom to operate, invalidity and portfolio work. Chemistry and life sciences drafting handles Markush structures, stereochemistry, experimental data and ST.26 sequence listings, with further pages for mechanical, ICT and biology work. DeepIP reports more than 500 law firms and in house teams as customers, across 25 jurisdictions.
The suite covers United States patent preparation and prosecution: drafting, proofreading, office action responses, IDS and USPTO forms, examiner and art unit research, family and timeline lookup, portfolio lists and firm rankings. Buyers are patent law firms sized by registered practitioners, solo practitioners at firms, and in house IP teams sized by issued patents. Family trees draw on EPO data, but drafting and forms follow USPTO practice. Generative drafting is built for sections and edits rather than whole applications, and trademark work, litigation and prosecution outside the United States are not described.
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?
DeepIP's April 2026 terms bar it from using customer data to train, retrain or improve any AI model unless both parties agree in writing. Model providers it selects must contractually commit not to train their own models on that data. A separate clause lets DeepIP use aggregated, anonymized usage data, which can include prompts, to analyze and improve the service. DeepIP's security page and trust center state the same no training position.
Section 2 of the Patent Bots terms states that neither Patent Bots nor AWS Bedrock, which runs the generative features, uses customer data for model training or improvement. Section 9 limits use of uploaded nonpublic information to providing the service, and the security page repeats that customer data is used only for the requested service.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Prompts sent to the model provider are kept only as long as processing takes, under the zero data retention settings DeepIP's terms require where a provider offers them. Inside DeepIP, customer data is stored encrypted during the subscription, and no period is stated for prompts and outputs while it runs. After termination, DeepIP destroys customer data within 30 days of the final export, or once 8 days pass without an export request. DeepIP's trust center says customer data is not stored.
Once proofreading results are returned, the patent documents are discarded at once, with no copy and no log of their content. Generative prompts and completions are not stored or logged by Patent Bots, and AWS Bedrock runs them with zero data retention. PatentLink lists, which hold public patent data and user notes, are stored on Google Cloud, and personal data in logs and backups follows unstated retention periods.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
DeepIP states that users and organizations are kept completely separate, and its trust center lists network segmentation among its controls. How separation works inside a firm is not described, including whether one attorney's matters are walled from another's or a matter can be limited to named users. Patent firms often prosecute for competitors in the same technical field, and no matter level restriction is described.
Each person needs an account of their own, and subscription content may be shared only within the subscribing company and with its clients or outside counsel. Nothing describes separating matters, clients or teams inside a firm's subscription.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
When a court order or administrative process compels disclosure, DeepIP's terms commit it to notify the customer promptly in writing. DeepIP also takes reasonable steps the customer asks for to resist, and lets the customer control the proceeding with counsel of its choice. No transparency report is published. Contracts run through DeepIP Incorporated in New York for North American customers and DeepIP SAS in Paris for everyone else.
Under section 9 of the terms, Patent Bots may disclose confidential information when legally required but promptly informs the customer so the disclosure can be contested or narrowed. It also commits to notify the customer of any misappropriation or misuse. No transparency report is published.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
DeepIP's prior art retrieval covers more than 120 million patents worldwide and non patent literature. Its prosecution module pulls USPTO and EPO file documents, including claims, examiner notes and cited art. No database or data provider is named for the patent collection, and no licensing basis, date range or update cadence is stated. DeepIP describes its product as combining third party models, proprietary models and unique data access, without naming the data.
Examiner statistics, office actions, prosecution histories and art unit training data come from USPTO records, and family trees combine USPTO and EPO data, including PCT and foreign filings. Statute and MPEP citations link to the USPTO. The terms note that patent office data may be temporarily unavailable, and no update cadence or rights basis for the data is stated.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
Office action responses drafted by DeepIP carry case law citations, and its freedom to operate module identifies active patent rights across jurisdictions. Nothing describes a check of a cited decision's later history, and no citator is named.
Family trees mark issued patents apart from pending and other applications, and PatentPlex timelines list prosecution and post issuance events from USPTO and EPO data. No determination of whether a patent remains in force is described, and drafted text cites no authority to check.
Refusal and Uncertainty Behavior
What does the product do when the answer is not in the corpus?
DeepIP's patentability module returns a structured novelty and obviousness assessment with confidence levels. The review module gives a probability of a section 101 rejection, and freedom to operate analysis rates the risk from each blocking patent. What the system does when a disclosure cannot support a claim, or no prior art supports an answer, is not described.
Nothing describes how the generative chat or drafting tools respond when the supplied documents do not support an answer. Ranked art unit predictions come without a stated confidence or abstention.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
The AI Hallucination Cases database, Damien Charlotin's record of court decisions on hallucinated AI content, has no entry for DeepIP. DeepIP drafts patent applications and office action responses with case law citations.
The AI Hallucination Cases database maintained by Damien Charlotin, which records court decisions worldwide that address hallucinated AI content and the tool involved where known, has no entry naming Patent Bots. Its output goes into patent applications and USPTO responses rather than court filings.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
DeepIP tells practitioners to validate everything the system produces and to apply their own professional judgment. Its terms state that it is not a law firm and gives no legal advice. No bar opinion, USPTO rule of professional conduct or patent office guidance on AI is named.
No ethics opinion, bar guidance or USPTO guidance on AI use by practitioners is engaged. The drafting tools are described as support for attorney judgment, without tying that to a practitioner's duties.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
DeepIP claims drafting time cut by up to 70 percent on a full application and more than two hours saved per attorney each day. Its ROI calculator puts typical savings at 40 to 55 percent of drafting and office action time, roughly $25,000 to $90,000 a year per attorney. The law firm page promises higher margins per case, and Philips reported about 20 percent better efficiency in a trial. Nothing addresses how AI assisted work appears on a client invoice.
Patent law firms are the main buyers, and the pitch is higher realization rates and fewer written off hours, especially on fixed fee work, with drafting tools said to save 6 to 12 hours per patent. Nothing addresses how AI assisted drafting or proofreading time should appear on a client's bill.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
Ten subprocessors appear on DeepIP's trust center, with roles and regions. Microsoft Azure and Google Cloud Platform host the service, OpenAI provides the language model and Datadog monitors it, each in the United States or the EU. Auth0, Mixpanel, Intercom, Hyperline, HubSpot and PandaDoc cover identity, analytics, support, billing, marketing and documents. The 2026 SOC 2 Type 2 report, ISO 27001 and ISO 42001 certificates and a penetration test are available through an access request. No data processing agreement is published.
A sub processor list dated 3 April 2026 names nine providers with location and purpose, including AWS for hosting and language model services and Anthropic for processing document content. The data processing addendum gives 30 days' notice of new sub processors. Audit reports and policies sit in a trust center behind a bot check, and no client facing disclosure pack is offered on the open site.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
Drafting, review, freedom to operate and invalidity work happen inside Microsoft Word, so comments and tracked changes record the editing. Freedom to operate and invalidity conclusions link to the underlying claim or prior art text, and DeepIP describes those outputs as audit ready. Nothing marks which passages a model wrote, and no record shows that a practitioner adopted generated claim language before filing.
Output goes into applications and office action responses filed with the USPTO, and nothing records which AI features or models contributed to a filing or how it was reviewed.
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 DeepIP if
- You want a full first draft from a disclosure. DeepIP turns an invention disclosure into claims, specification and abstract in one pass, ready for attorney review.
- You file beyond the USPTO. DeepIP shapes drafts to the formal requirements of major patent offices, including the EPO.
- Prior art search belongs in the drafting tool. DeepIP searches more than 120 million patents by meaning and surfaces references while you draft.
Choose Patent Bots if
- Error checking matters more than drafting speed. Patent Bots proofreads antecedent basis, claim numbering, word support and reference labels in applications and responses.
- You want to know your examiner before responding. Patent Bots shows grant rates, interview effect and appeal outcomes for every USPTO examiner.
- Budget approval needs a number. Patent Bots publishes per practitioner firm pricing and company packages with yearly minimums.
In summary
DeepIP
DeepIP is an AI patent assistant that works inside Microsoft Word, founded in New York and Paris. It drafts complete applications from invention disclosures in a firm's own templates and style, drafts office action responses with case law citations, and retrieves prior art from more than 120 million patents. Its trust center offers SOC 2 Type 2, ISO 27001 and ISO 42001 reports on request, and its terms bar training on customer data. It runs on Microsoft Azure, with an on premises option, and sells seats without a published price.
Patent Bots
Patent Bots, founded in 2018, sells Prep & Pros Pro for proofreading and drafting in Word and the browser, along with office action and IDS tools, examiner statistics, Art Unit Predictor and portfolio lists. Its generative features run on Anthropic models through AWS Bedrock and are switched on by an administrator. According to the AI Legal Index, Patent Bots processes and stores all data in the United States and discards patent documents once results are returned.
Questions buyers ask
Can DeepIP or Patent Bots draft a whole patent application?
DeepIP generates complete applications from invention disclosures. Patent Bots drafts sections such as figure descriptions, claim summaries and claim clauses, and leaves the full application to the attorney. From the AI Legal Index, based on each vendor's own published materials as of October 13, 2026. No vendor pays for placement.
Which patent offices do DeepIP and Patent Bots support?
DeepIP adapts drafts to the requirements of major offices, including the USPTO and EPO. Patent Bots builds its forms, statistics and office action tools around the USPTO and uses EPO data for family trees. From the AI Legal Index, based on each vendor's own published materials as of October 13, 2026. No vendor pays for placement.
Where do DeepIP and Patent Bots run?
DeepIP runs on Microsoft Azure and Google Cloud in the United States or the EU, in the territory the order form names, and also offers an on premises deployment. Patent Bots runs on Google Cloud and AWS in the United States and discards patent documents after processing. From the AI Legal Index, based on each vendor's own published materials as of October 13, 2026. No vendor pays for placement.
How do DeepIP and Patent Bots handle office actions?
DeepIP pulls the USPTO or EPO file, summarizes and ranks the examiner's objections, and drafts structured responses with case law citations. Patent Bots generates response shells, summarizes claim status and lines up the office action, pending claims and cited prior art in Office Action Explorer. From the AI Legal Index, based on each vendor's own published materials as of October 13, 2026. No vendor pays for placement.
Is pricing published for DeepIP and Patent Bots?
DeepIP sells seats on monthly or annual subscriptions and publishes no price. Patent Bots lists firm, company and individual prices, and its free trial runs for 14 days. From the AI Legal Index, based on each vendor's own published materials as of October 13, 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.
DeepIP was founded in New York and Paris. Patent Bots is made by Patent Bots, Inc., which publishes its own comparison of patent drafting tools that includes DeepIP. Neither vendor reviewed this page.