Up: AI Office Action response software

AI Office Action analysis that verifies the cited passages

An Office Action is the examiner’s case against the claims as they read on the day it was mailed. Analyzing it means three things a careful practitioner has always done by hand: account for every ground, check that each cited passage says what the examiner says it says, and remember what the file already says about the same reference or the same scope. An AI can do each of those well or badly, and the difference is almost entirely what it is reading from. This guide describes how PatentReply does the analysis from the compiled file wrapper, and what it hands to the strategy step.

Scope: This guide explains the analysis a practitioner can expect from PatentReply inside Claude, ChatGPT, Grok, or another MCP client. The commands and tool names are in the documentation, under Reviewing an Office Action. Nothing here is legal advice.

What an Office Action analysis has to cover

Every ground in the action, with the statute it rests on and the claims it reaches. The grounds fall under a handful of statutes: subject-matter eligibility under 35 U.S.C. § 101, anticipation under § 102, obviousness under § 103, and written description, enablement, and indefiniteness under § 112, plus objections to the specification, the drawings, or the form of the claims. A rejection of claims 1–5 and 8 under § 103 over two references is one ground; the same references against claim 9 under § 102 is another. An analysis that merges them, or loses the dependent claims a ground reaches, is already wrong.

The status of every claim: pending, withdrawn, canceled, allowed, or objected to as allowable if rewritten. The examiner’s mapping: for each claim element the ground addresses, which reference and which passage the examiner relied on, and, in a § 103 ground, the stated reason to combine. And the posture of the action itself: non-final or final, which decides what the reply may contain under 37 CFR 1.111 and 1.116.

How PatentReply parses the Office Action into structured grounds

When a matter is compiled, each Office Action in the wrapper is read from the USPTO’s own text and parsed into issues: objections first, then rejections, each with its statute, the claims it touches, and the examiner’s reasoning as written. The reasoning is kept verbatim and checked against the stored action, so what your AI quotes as the examiner’s position is the examiner’s position, not a paraphrase. Each ground is tied to the claim set that was live on the action’s mailing date, which matters when claims were amended between actions: a rejection of “claim 1” binds to the claim 1 the examiner was reading.

The parse also records what it could not do. A paper the parser could not read, claim text that two USPTO papers disagree about, and a reference the record does not hold are flagged as exceptions rather than silently guessed. The analysis says which claim set or action the defect sits in: a defect in the operative set or the outstanding action can block drafting; one in a superseded set qualifies only the reads as of that date.

Verifying each cited passage against the reference

This is the step that separates an analysis from a summary. For each citation the examiner made, PatentReply resolves the citation as written to an address in the stored copy of the reference, returns the passage’s actual text, and sets the examiner’s assertion about it beside that text. If the examiner quoted the reference, the quotation is matched against the stored text and reported as confirmed or not. If the citation has no text to resolve, because it points only at a figure, or because PatentReply does not hold the reference, the analysis says so and tells the model not to assess the characterization.

Every passage the AI reads carries a stable address in the record, so a conclusion can always be traced back to the paragraph it rests on, by the AI and by you. On patentreply.ai, the same work is laid out for the attorney’s own eyes in the OA Reading Room: the action’s text unaltered, every citation a chip that opens the cited paragraph with its stored reading, and the deterministic findings counted beside it: a quotation attributed to one reference that is verbatim another, a quoted passage not found where cited, claims listed in a rejection that were canceled before the action. “No findings” means every pin cite, claim list, and reference name checked out.

Reading examiner reasoning and interview history across the file

A rejection rarely arrives alone. The same reference may have been applied in an earlier action and argued around in an earlier response; a claim may have been narrowed to get past it; an interview summary may record what the examiner said would be persuasive. The analysis reads the whole wrapper for these history constraints: what was argued before against the same reference, what the applicant said the claims mean, and which narrowing amendments bear on scope. Those statements shape what the next response can say without contradicting the record, and they are the raw material of prosecution history estoppel later.

Because the matter is compiled from the public record, this works for an application you inherited as well as one you have prosecuted from the start. The analysis does not depend on what anyone remembers about the case.

What the analysis hands to the strategy step

The output of the analysis is a review the strategy can be built on: every issue with its verified reasoning and resolved citations; the evidence gaps, named; the history constraints; and a readiness note that says whether the matter is ready for traversal drafting and amendment drafting, and what blocks either. From there the attorney and the AI develop options per issue (traverse, amend, both, concede, request an interview, preserve for appeal), each anchored to the passages the analysis found. When the analysis shows that a cited passage really does teach an element, the question becomes whether and how to amend, which is the subject of the guide on claim amendments in an Office Action response.

To see the actions on a real public application before compiling one, use the Office Action lookup: it shows every action the USPTO exposes for an application, with the statutory grounds, the claims each ground reaches, and the references relied on.

Up: AI Office Action response software