P

Patent Bots

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.

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.