AI Office Action response
AI Office Action response work, grounded in the file wrapper.
Connect PatentReply to Claude, ChatGPT, Grok, or any MCP client you already use. Give it a U.S. application number. It works from the organized, current, source-linked file wrapper, the Office Action, the claims as they stand, the cited art, and the prosecution history, not a pile of uploaded PDFs.
Works with Claude, ChatGPT, Grok, and any other MCP-compatible AI client.
Patent Professional: $59 per seat per month, 3 new matters included. Free tier available. Pricing
How it works
Compile, review, plan, approve, prepare, and check: the stages the product keeps with the matter, in the order the work happens.
01
Review the Office Action, claims, cited art, and prosecution history together.
PatentReply compiles the application into a single matter: every Office Action, the current claims and their amendment history, the references the examiner cited, and the prosecution timeline. Your AI reviews the pending action from that compiled record, so the review covers the whole file, not whichever PDF happened to be uploaded.
The review lists every objection and rejection in drafting order, each with its statute, the claims it touches, and the examiner’s reasoning in the examiner’s own words, checked against the stored action. It also names what the record does not hold: a reference PatentReply could not retrieve, a figures-only citation, disputed claim text.
02
Verify what the examiner actually relied on.
Each citation in the action is resolved to the passage of the cited reference it points at and set beside the examiner’s assertion. The attorney can open the exact paragraph the examiner mapped to each claim element and decide whether it says what the rejection says it says. Paragraph-level provenance, not a summary.
When the examiner quotes a reference, the quotation is matched against the stored text and reported as confirmed or not. On patentreply.ai, the OA Reading Room shows the action with every cite as a chip that opens the cited paragraph, for the attorney’s own check by eye. Whether a passage supports the rejection is never asserted; that judgment stays with you.
03
Develop an attorney-approved response strategy.
For each issue, your AI and you develop a response plan: traverse, amend, both, concede, request an interview, or preserve for appeal, with the supporting passages attached and the history constraints the review found (what was argued before against the same reference, what scope was already given up). The plan is stored with the matter, per issue and revisioned, so it survives the conversation.
Nothing proceeds until the attorney approves the plan on PatentReply’s approval page, signed in. Approval is an explicit act, recorded on the matter with the plan revision it approved. A yes typed into the chat is not approval, and the model cannot approve on your behalf.
04
Prepare amendments, Remarks, and a filing-ready response.
From the approved plan, PatentReply generates the claim listing under 37 CFR 1.121 deterministically from the proposed claim text, with the status identifiers and the underline-and-strikethrough markings the rule requires. Your AI drafts the Remarks from the approved strategy, citing the record; PatentReply assembles the response document (.docx) with the caption, the listing, the Remarks, and the signature block, for the attorney’s review and edits.
An independent pre-filing QC checks a draft against the record: the caption and posture, the listing, the amendments against the operative claims, and the evidence behind each assertion about a reference. Every finding is anchored to a passage or marked unverified. The draft can come from anywhere, including a document that never touched PatentReply.
05
Continue in Claude, ChatGPT, or another MCP client without rebuilding the matter.
The compiled matter persists. Switch clients, return weeks later for the next Office Action, or hand the matter to a colleague on their own seat; the record, the plan, the approvals, and the history are already there. Ask where you are on the response and PatentReply reports the live stage from stored state, not from the conversation.
When the next action arrives, refresh the matter: the new action, the response you filed, and any new references join the same record, and what you argued becomes a history constraint on the next response. Refreshing a matter you already hold is free.
06
What PatentReply does not do.
PatentReply does not replace legal judgment, sign papers, file papers with the USPTO, or manage deadlines. The attorney reviews every output, decides every strategy, and files every paper. The record is automated; legal judgment is not.
Approvals and legal judgment sets out what requires approval, where it happens, and what stays with the practitioner.
Built on the compiled matter, not on uploaded PDFs.
Uploading a wrapper gives a model documents. A compiled matter gives it the state of the case, and keeps it.
A persistent compiled matter
One matter per application, compiled from the USPTO record, kept current by refresh, and open from any supported AI client and at any later prosecution event.
Paragraph-level source verification
Every passage your AI reads carries a stable address in the stored record. Citations resolve to that address and render the exact stored text, so an answer can always lead back to the filed paper.
Claim-history and current-claim awareness
Every claim set ever presented is parsed, so the record can state the operative text of a claim as of any legal date, its lineage, and the arguments made around it.
Attorney approval gates
Confirming the response strategy and releasing a filing paper stop for approval on patentreply.ai, signed in as you. The model cannot approve its own plan or turn an unapproved strategy into filing-ready amendments.
Response assembly and QC against the record
The claim listing and the response document are generated on the server from the approved plan, never assembled by the model from claim text it read. A pre-filing QC checks any draft against the record.
For patent attorneys and agents who already use an AI assistant.
- Solo and small-firm prosecutors who want the record organized before the AI starts, without uploading a wrapper into each conversation.
- In-house counsel who approve strategy on their own matters and want a draft from outside counsel checked against the record before it is filed.
- Firms standardizing AI use with approval controls: every substantive step stops for a signed-in approval on the site, recorded with the matter.
Documentation: connect a client, compile a matter, work the response
Go deeper
- AI Office Action analysis: rejections, claims, and cited artHow the action is parsed into structured grounds and each examiner citation is resolved to the passage it relies on.
- Claim amendments in an Office Action responseWhen to amend versus traverse, finding written-description support, why claim history matters, and the 37 CFR 1.121 listing.
- The Office Action response workflow: review to filingWhy the work runs compile, review, strategy, approval, amendments, QC, and filing by the practitioner, in that order.
Questions
The prosecution record for your AI
Work your next Office Action from the file wrapper.
Connect PatentReply to Claude, ChatGPT, Grok, or another supported AI client and give it the application number. PatentReply compiles the record behind the conversation: the claims and their amendment history, every Office Action and response, and the cited references, each answer pointing back to the filed passage.
One new public application a month is free. The first compile takes a few minutes.
Strategy and filing changes require your approval.