How-To Guide

How to Use AI for a Patent Law Firm in 2026: Prior-Art Search, Spec Drafting, Office Actions & PTAB Briefing

Published May 19, 2026 · 17 min read · For managing partners of 2-25 attorney patent-law firms — prosecution, post-grant, litigation, and portfolio management; utility, design, PCT, and international.

TL;DR

Two AI layers for a patent-law firm in 2026: a prosecution-and-litigation layer (prior-art search, spec and claims drafting, office-action response, PTAB briefing, claim charts) and a portfolio-and-practice layer (docket management, IDS preparation, annuity decisions, client status letters, managing-partner scorecard).
Ten prompts below: new-matter intake and conflicts, prior-art search assistant, spec and claims drafting brief, IDS management under 37 CFR 1.56, office-action response drafter, examiner-interview prep memo, claim charts and FTO memo, PTAB IPR preliminary response, docket and annuity decision memo, managing-partner monthly scorecard.
AI drafts; the registered practitioner signs. Every paper filed at the USPTO carries a 37 CFR 11.18 certification and an ABA Model Rule 3.3 candor obligation — AI does not absorb either.
Client confidential information and invention disclosures go only into BAA-covered or enterprise-tenanted AI tools — never consumer ChatGPT. ABA Formal Opinion 512 (July 2024) made this explicit for every Model Rule 1.6 jurisdiction.
ROI is real: prosecution attorneys recapture 30-50% of drafting and OA-response time; search costs drop 60-80%. The firms that write it into billing policy and QA workflow keep the margin; the ones that don't leak it to clients and to hallucinated citations.

The 2026 patent-law firm AI stack

Layer	Tool	Use
Patent drafting AI	Edge by Rowan, Solve Intelligence, PatentPal, DeepIP, Henry AI, Qatent, Patentfield, PowerPatent, Specif.io	Spec, claims, drawings from invention disclosure
Prior-art search	IPRally, Patlytics, XLSCOUT Novelty Checker, Rowan, Solve Search, LexisNexis PatentSight+, Clarivate Derwent Innovation, Questel Orbit Intelligence, Google Patents, The Lens, PatSnap	Semantic + citation + classification search
Office-action response	Edge, Solve Intelligence, DeepIP, Henry AI, Rowan OA, Patlytics Respond, LexisNexis Lex Machina, Westlaw Precision AI, CoCounsel Patent, Harvey, Paxton	§102/§103/§101/§112 response drafts + claim charts
Docketing + PAIR	Anaqua, CPA Global (Clarivate IPMS), Alt Legal, PATTSY WAVE, Foundation IP, FirstToFile, Dennemeyer, Computer Packages Inc., IPfolio, Inteum	Due-date calc, USPTO PAIR pull, annuity deadlines
Annuity decisions	Anaqua AQX, CPA Global, Dennemeyer AnnuiPlus, Questel Renewals, IPAN, Clarivate Maintenance Fees, PatentSight valuation	3.5/7.5/11.5 maintenance-fee go/no-go with valuation scoring
Claim charts + FTO	Patlytics Claim Charts, XLSCOUT Ideacue, IPRally Claim Charting, Rowan FTO, Solve Freedom-to-Operate, Cipher, LexisNexis PatentAdvisor, Clarivate CompuMark	Infringement read + FTO opinion drafts
Litigation + PTAB	Relativity aiR, Everlaw AI, DISCO Cecilia, Reveal, Logikcull, CoCounsel Patent, Harvey, Paxton, Westlaw Precision AI, Lexis+ AI, Lex Machina, Docket Navigator, Darts-IP, UnifiedPatents	IPR petitions, preliminary responses, discovery, Markman briefing
Practice management	Clio Manage + Duo, Litify, Filevine, Smokeball, MyCase IQ, PracticePanther, CosmoLex, Aderant, Elite 3E, iManage, NetDocuments	Time, billing, trust, docs, conflicts
Legal research	Westlaw Precision AI, Lexis+ AI, CoCounsel, Harvey, Paxton, Bloomberg Law AI, Fastcase, vLex Vincent, Casetext, BriefCatch	Case law + MPEP + Federal Circuit + PTAB precedent
Ambient + dictation	Jump, Zocks, Fireflies Enterprise, Otter Business, Read.ai, Supernormal, Noota, tl;dv, MeetGeek	Inventor-interview capture, client meeting notes

10 copy-paste prompts for a patent-law firm

Run each prompt in a BAA-covered, enterprise-tenanted AI (CoCounsel Patent, Harvey, Paxton, Westlaw Precision AI, Lexis+ AI, or a firm-tenant Edge/Rowan/Solve). Never paste invention disclosures, client-confidential specifications, unpublished applications, or attorney work product into a consumer chatbot. ABA Formal Op 512 is explicit.

1. New-matter intake + conflicts + engagement scoping

You are a patent-law intake associate. From the inventor interview transcript, extract: (a) real inventors and assignee (check §100(f) & §115 inventorship — natural persons only, per USPTO Feb 2024 Inventorship Guidance on AI-assisted inventions), (b) invention area (mech/EE/chem/bio/software/CS/medical device), (c) public-disclosure events with dates for §102(a)(1)/(b)(1) grace-period analysis, (d) prior patent family + priority claims, (e) funding source + government-interest under 35 USC §202 (Bayh-Dole), (f) export-control flag (ITAR 22 CFR 120-130 / EAR 15 CFR 730-774) for foreign-filing license 35 USC §184, (g) conflicts (prior representation + current competitors + litigation counsel), (h) scope (utility vs design vs PCT vs provisional), (i) budget and claim-count target. Output: intake memo with conflicts-clear / conflicts-likely / conflicts-hard flag, §102 bar-date calendar (one-year + absolute bar), foreign-filing-license checklist, engagement-letter scope items, fee estimate (flat vs hourly), and three open questions for the inventor. Do not issue a legal opinion — draft only.

2. Prior-art search assistant (augmented, not autonomous)

You are a patent-search associate. Given this invention disclosure, produce a structured prior-art search plan for the registered practitioner to review and execute: (a) independent-claim concept extraction (3-5 core concepts), (b) CPC + USPC + IPC classification candidates with rationale, (c) keyword stems + synonyms + acronyms for each concept, (d) suggested semantic-search queries for IPRally / Patlytics / XLSCOUT / Rowan / Solve Search, (e) non-patent literature targets (IEEE Xplore, ACM DL, arXiv, SSRN, Google Scholar, trade publications), (f) competitor and named-inventor focused searches, (g) earliest-plausible-prior-art date based on disclosure, (h) foreign-family review priorities (EPO, JPO, KIPO, CNIPA, CA, AU). Output: 10-row search-strategy table with query, database, classification, expected hit count. Flag: I do not make the materiality determination under 37 CFR 1.56. The registered practitioner reviews every hit and decides what goes on the IDS. Document the search methodology in the file wrapper.

3. Specification + claims drafting brief (from invention disclosure)

You are a patent-drafting associate. From this invention disclosure and closest prior art, draft: (a) Title (noun-phrase, under 500 characters, non-marketing), (b) Field of Invention, (c) Background (problem + prior-art limitation, no admissions), (d) Summary of Invention (mirroring independent claims), (e) Brief Description of Drawings with figure numbering and labels, (f) Detailed Description with embodiment(s), alternatives, and 112(a) written-description + enablement support for every claim term, (g) three-level claim set: one broad independent, one narrower independent alternative, 15-20 dependents covering fallback features, method + apparatus + CRM Beauregard claims where appropriate, (h) abstract under 150 words. 112 discipline: every claim term has antecedent basis in the spec. Every "means for" gets §112(f) corresponding structure. Avoid "wherein the X is Y" where X is unclaimed. Software/AI claims — draft with Alice step-two in mind: tie to technical improvement, computing-resource efficiency, or practical application, not just a mental process on a generic computer. Flag any claim that risks §101 rejection post-Alice/Mayo. Output in a single draft that the attorney can edit; flag every term that needs inventor confirmation.

4. IDS management under 37 CFR 1.56 + 1.97

You are an IDS-management paralegal. Given the prior-art search results, the related foreign-family search reports (EPO, JPO, WIPO ISR/IPRP, CNIPA first-office action), and the file wrapper history, produce an IDS preparation memo: (a) PTO/SB/08a form with patent-literature section (patent number, kind, date, name, class) and non-patent-literature section (AIP citation format), (b) categorization: must-cite (Federal Circuit Therasense "but-for" material + cited in related family + cited by examiner in related US case), should-cite (closest art from search), discretionary (background), (c) timing analysis: before first OA (free), before final OA (1.97(c) + fee), after final OA or allowance (1.97(d) + statement), after issue-fee (1.97(i) or supplemental exam 35 USC §257), (d) translation requirements for non-English references with a concise statement of relevance, (e) 37 CFR 1.56 duty-of-disclosure memo for the attorney's signature. Flag inequitable-conduct risk under Therasense v. Becton Dickinson (Fed Cir 2011): but-for materiality + specific intent to deceive. Every IDS decision is the practitioner's — this memo is a draft for review.

5. Office-action response drafter (§102 / §103 / §101 / §112)

You are an office-action response associate. From this non-final office action, the cited references, the current claim set, and the specification, draft: For §102 anticipation rejections: a claim-chart mapping every limitation to the reference; identify any limitation not expressly or inherently disclosed (In re Robertson inherency standard); propose claim amendment if missing limitation exists in a dependent. For §103 obviousness rejections: Graham factors analysis (scope of prior art, differences, level of ordinary skill, objective indicia); KSR rationale rebuttal if examiner used generic "common sense" without articulated reasoning; teaching-away, unexpected results, long-felt need, commercial success if supported by the record; secondary-reference combination critique if examiner used TSM in a non-analogous art (In re Clay). For §101 Alice/Mayo rejections: step 2A prong 1 (is claim directed to judicial exception), prong 2 (integrated into practical application per 2019 Revised Patent Subject Matter Eligibility Guidance); step 2B (significantly more / inventive concept); focus on technical improvement and computing-resource efficiency, not mental-process reframing. For §112(a) written description + enablement: map claim terms to specification pages and lines; Wands factors if enablement challenged. For §112(b) indefiniteness: Nautilus "reasonable certainty"; provide antecedent-basis fix or definitional amendment. Output: draft response with Remarks section (reference-by-reference), claim amendments in annotated markup (added text underlined, deleted text in strikethrough brackets), and Interview Summary if telephonic interview held. Every cited case gets KeyCited or Shepardized — do not hallucinate citations (Mata v. Avianca + Park v. Kim sanctions apply).

6. Examiner-interview prep memo

You are a prosecution associate. For this pending application with a pending office action, prepare an examiner-interview agenda and talking-points memo: (a) examiner's Art Unit, SPE, tenure, allowance rate, and docket posture from PatentAdvisor or Patexia analytics, (b) two-page case summary with the claim-at-issue, the closest reference, and the single strongest distinguishing feature, (c) proposed amendment(s) to present at the interview — include at least one narrowing amendment the examiner is likely to allow and one that preserves broadest scope, (d) three claim-construction arguments with specification cites, (e) two Federal Circuit or PTAB decisions on point (verify every cite), (f) interview-summary draft skeleton per 37 CFR 1.133 and MPEP 713.04. Output: 2-page interview brief the attorney can review the day of. Remind the attorney: §1.133 does not require written agreement for allowance, but every interview requires a complete written record; the examiner signs the Interview Summary, but the attorney may file a supplemental one.

7. Claim charts + freedom-to-operate memo

You are a patent-analysis associate. Given (a) this target product/service description, (b) this set of third-party issued US patents and pending published applications, and (c) the assignee-and-family landscape search, produce: Claim-chart matrix: one row per independent claim of each asserted/blocking patent, columns for each claim limitation, cells filled with evidence from the target product (screenshots, spec sheets, marketing materials) or "NOT FOUND — gap". Infringement read: literal + doctrine-of-equivalents (Warner-Jenkinson function-way-result + Festo prosecution-history estoppel); identify any all-elements-rule gaps. Invalidity read: §102/§103 prior art identified for each blocking claim; obviousness combination with motivation-to-combine analysis; §101 + §112 defenses where available. FTO opinion draft: non-infringement + invalidity conclusion for each patent on a Low/Medium/High risk scale; redesign options for Medium/High; licensing/indemnification recommendation. CRITICAL: this is a draft for partner review. A real FTO opinion is attorney-signed, well-documented, and protects against willful infringement (Halo v. Pulse) — the firm's work-product and the client's attorney-client privilege attach only to the signed opinion, not to this draft.

8. PTAB IPR petition + preliminary response

You are a PTAB-practice associate. For this challenged patent, prepare: (A) Petitioner side — IPR petition draft under 37 CFR 42.104: (1) real-party-in-interest + privies disclosure (Applications in Internet Time v. RPX), (2) §315(b) 1-year bar analysis from service of complaint, (3) claim-construction under Phillips (post-SAS), (4) statement of precise relief, (5) grounds — each ground a single §102 or §103 combination with motivation-to-combine and KSR rationale, (6) Graham factors, (7) expert declaration outline with level of ordinary skill, (8) Fintiv factor analysis for co-pending district-court litigation (post-Vidal memo 2022 + 2024 interim guidance). (B) Patent-owner side — preliminary response under 37 CFR 42.107: (1) RPI/privy challenge if plausible, (2) §315(b) bar challenge if applicable, (3) claim-construction pushback with intrinsic + extrinsic evidence, (4) reference-by-reference challenge (not a teaching, teaching away, not analogous art), (5) Fintiv discretionary-denial argument, (6) §325(d) argument if same or substantially same art/arguments previously considered. Output: 60-page skeleton draft with citations placeholders. Every case cite gets verified against Westlaw Precision AI or Lexis+ AI before filing. Partner signs under 37 CFR 11.18.

9. Docket + annuity-decision memo

You are a portfolio-management associate. For this client's portfolio, pull from the docket system (Anaqua / CPA Global / Alt Legal / PATTSY WAVE) and produce: (A) 90-day due-date dashboard: office-action responses, RCE deadlines, notice-of-allowance + issue-fee windows, foreign-filing-deadline PCT 30/31-month + Paris Convention 12-month, annuity deadlines 3.5/7.5/11.5 year US + foreign schedules, USPTO reply-brief 2-month + 5-month extendable-to-7-month. (B) Annuity-decision memo for each asset at 3.5/7.5/11.5 year: (1) commercial-use status (practicing / licensed / asserted / dormant), (2) claim-scope value via PatentSight or IPlytics benchmarking, (3) remaining term, (4) pending continuations that depend on parent, (5) forward-citation count and competitor-watch hits, (6) recommended action: pay / abandon / convert-to-provisional-continuation / licensing-candidate. Output: one-page-per-asset go/no-go memo for managing partner and client decision. Flag every asset with a deadline under 30 days in red. Every recommendation is a draft — the client and partner make the call.

10. Managing-partner monthly scorecard

You are the managing partner's chief of staff. From this month's billing data, docket system, USPTO PAIR data, and client-satisfaction surveys, produce a one-page scorecard covering: Portfolio health: new applications filed, applications allowed, allowance rate (firm vs. art-unit benchmark from PatentAdvisor), average pendency, continuation rate, foreign-filing rate, claim-amendment-per-OA average. Financial: billable hours by attorney, realization rate, WIP, AR aging (< 30, 30-60, 60-90, > 90 days), write-offs, new-matter revenue vs. prior month, flat-fee vs. hourly mix, alternative-fee matter profitability. Client: top 10 clients by revenue, client concentration risk (>25% flag), NPS or CSAT from quarterly survey, client departures + onboarding, RFP pipeline. AI program: prompts-used count, hours saved estimate per attorney, hallucination-catch incidents, per-matter AI-cost disclosure compliance under ABA Model Rule 1.5 and state bar guidance, Rule 1.6 violation incidents (zero tolerance), CLE hours on AI-competence per Model Rule 1.1. Risk: malpractice tail (missed deadlines = zero tolerance), IDS-completeness audit results, docket-discrepancy incidents, bar complaints, conflict-waiver compliance. Output: one-page scorecard with three trendlines + one risk flag + one action item for next month. Partner signs and distributes to partnership.

Compliance floor

USPTO April 11, 2024 Guidance on Use of AI-Based Tools in Practice Before the USPTO — duty of candor + good faith (37 CFR 1.56, 11.18) extends to AI-generated content; practitioner personally vouches for all content in signed papers; reasonable inquiry under Rule 11.18(b); pro hac vice and foreign practitioners included.
USPTO February 13, 2024 Inventorship Guidance for AI-Assisted Inventions — natural person must make a significant contribution to each claim (Thaler v. Vidal, Fed Cir 2022); Pannu v. Iolab factors apply; AI-assisted inventions are patentable but the AI is not an inventor; each claim needs at least one natural-person inventor.
37 CFR Part 1 (rules of practice), Part 11 (representation — §11.18 signature certification, §11.101-.901 USPTO Rules of Professional Conduct), Part 41 (PTAB appeals), Part 42 (AIA trial practice — IPR, PGR, CBM, derivation).
ABA Model Rules — 1.1 competence (comment [8] on technology + AI), 1.3 diligence, 1.4 communication, 1.5 fees (per-matter AI billing disclosure), 1.6 confidentiality (enterprise tenancy + BAA), 3.3 candor to tribunal, 5.1 + 5.3 supervision of AI as nonlawyer assistant, 7.1 truthful communication.
ABA Formal Opinion 512 (July 2024) — generative AI + competence, confidentiality, candor, supervision, fees, communication; binding interpretive on every jurisdiction that adopts Model Rules.
State bar AI opinions — CA Practical Guidance Nov 2023 + Apr 2024; NY Task Force April 2024; FL Ethics Opinion 24-1 Jan 2024; DC Ethics Opinion 388 April 2024; plus VA, NC, WA, TX, PA, NJ advisory opinions — check your home jurisdiction before rollout.
Citation-verification sanctions — Mata v. Avianca (SDNY 2023), Park v. Kim (2d Cir. 2024), Morgan & Morgan 2025 nationwide order; FRCP Rule 11 + 28 USC §1927; USPTO 37 CFR 11.18(b) signature certification. Every cite gets KeyCited or Shepardized before filing.
35 USC §§101, 102, 103, 112, 115, 184, 202 — subject-matter eligibility (Alice/Mayo + 2019 Revised Guidance), novelty, obviousness (KSR), written description + enablement + definiteness (Nautilus, Wands, Ariad), inventorship declaration, foreign-filing license, Bayh-Dole government interest.
Federal Circuit and PTAB precedent — Phillips v. AWH (claim construction), Therasense v. Becton Dickinson (inequitable conduct), Halo v. Pulse (willful infringement), Warner-Jenkinson + Festo (DOE), SAS v. Iancu + Fintiv + 2024 Vidal guidance (PTAB discretionary denial), Thaler v. Vidal (AI inventorship).
ITAR 22 CFR 120-130 + EAR 15 CFR 730-774 — export-control screening for foreign-filing licenses and cloud-based AI tools; defense-article + defense-service classifications.
Data handling — client confidential information (invention disclosures, draft applications, unpublished prosecution) goes only to enterprise-tenanted AI with BAA/DPA and no-training defaults; consumer chatbots are off-limits under Model Rule 1.6.

60-day rollout plan

Week 1-2 — Governance. Adopt firm AI-use policy (permitted tools, prohibited tools, per-matter disclosure, confidentiality defaults, citation-verification requirement, supervision of nonlawyer use). Map tools to data-sensitivity tiers: (1) public research, (2) firm-work-product, (3) client-confidential + PHI + SUD + privileged. Get enterprise tenants for Westlaw Precision AI, Lexis+ AI, CoCounsel Patent, Edge/Rowan/Solve, and at least one prior-art search tool.
Week 2-3 — Training. 4-hour CLE-credit training for every attorney on ABA Formal Op 512, USPTO April 2024 AI Guidance, and citation-verification discipline. Hands-on lab with two actual pending matters per attorney using prompts 2-5 and 7. Everyone signs the AI-use policy.
Week 3-4 — Prosecution workflow. Stand up prompts 1-6 (intake, prior-art search, spec drafting, IDS, OA response, interview prep) with designated review-before-filing checkpoints. Designated "AI QA" rotation — every AI-drafted paper gets second-attorney review before filing.
Week 4-6 — Portfolio + litigation workflow. Roll out prompts 7-9 (claim charts/FTO, PTAB, docket/annuity). Connect docket system (Anaqua / CPA Global / Alt Legal) to due-date dashboard. Annuity-decision memo goes to client quarterly.
Week 6-8 — Billing + scorecard. Stand up prompt 10 (managing-partner scorecard). Update engagement letters to reflect AI-use disclosure and billing model (flat vs. hourly with AI-assisted rate). Quarterly partnership review of AI program metrics + incidents.
Day 60 — First partnership review. Hours-saved by matter type, hallucination-catch incidents, client-satisfaction delta, AR-aging delta, and first written update to insurance malpractice carrier on AI-use program.

Common mistakes

Consumer chatbot for invention disclosures. Pasting an inventor interview or a draft specification into ChatGPT's default (non-enterprise) tier is an ABA Model Rule 1.6 violation and, for unpublished applications, may create §102(a) public-disclosure risk in some configurations. Always enterprise tenant, always BAA/DPA, always no-training defaults.
Unverified case citations. Every cited case, MPEP section, and PTAB decision gets KeyCited or Shepardized before filing. Mata v. Avianca, Park v. Kim, and Morgan & Morgan 2025 sanctions apply to USPTO filings under 37 CFR 11.18(b) and to PTAB under 37 CFR 42.11.
IDS autopilot. AI can draft the IDS; the registered practitioner signs and remains personally responsible for duty of disclosure under 37 CFR 1.56. Therasense inequitable-conduct risk is unchanged — but-for materiality + specific intent survive AI-assisted search.
Claim drafting without 112 discipline. AI will happily generate claims with no antecedent basis, unsupported "wherein" clauses, or §112(f) means-plus-function traps with no corresponding structure. Every claim term gets a 112(a)/(b)/(f) review before filing.
AI inventor trap. Do not list "AI" or a system name on the §115 inventor declaration. Thaler v. Vidal (Fed Cir 2022) and USPTO Feb 2024 Guidance — inventors are natural persons. Failure to name a real inventor or naming an AI is a patent-invalidation risk.
Billing ambiguity. Clients in 2026 expect per-matter AI-use disclosure. ABA Formal Op 512 and multiple state opinions require reasonable fees under Model Rule 1.5 — you cannot bill 10 hours for an AI-drafted OA response that took 3. Update the engagement letter and state the AI-rate model.
Export control blind spot. Cloud-based AI tools may route invention disclosures through foreign servers. Check the tool's data-residency before use on technology with export-control sensitivity (ITAR/EAR). Foreign-filing-license applications (35 USC §184) are still the practitioner's obligation.
No QA checkpoint. Every AI-drafted paper filed at USPTO carries a 37 CFR 11.18(b) certification. Without a named second-attorney QA checkpoint, a hallucinated citation or a missed limitation in a claim chart becomes a malpractice event and a bar complaint.

FAQ

Does USPTO allow AI use in patent prosecution?

Yes, with clear guardrails. The USPTO's April 11, 2024 Guidance on Use of AI-Based Tools in Practice Before the USPTO confirms that AI tools may be used but the registered practitioner signing a paper under 37 CFR 11.18 personally vouches for its contents. The duty of candor and good faith (37 CFR 1.56) extends to any AI-generated content — the practitioner must verify every citation, every claim-construction argument, and every representation. AI-assisted inventorship is addressed in the USPTO's separate February 2024 Inventorship Guidance: a natural person must make a significant contribution to each claim. Using AI to draft a specification is permitted; using AI to generate an invention with no human inventor is not.

Can AI write a patent application?

AI can draft a specification, generate claim sets in parallel, and produce drawings — but the signed work product belongs to the registered practitioner. The 2026 practical pattern: attorney interviews the inventor, feeds a structured invention disclosure into a drafting assistant (Edge, Rowan, Solve Intelligence, PatentPal, DeepIP, Henry AI), reviews and rewrites every independent claim, and supervises dependent claim generation. The attorney is still the drafter for ABA Model Rule 1.1 competence and USPTO duty-of-candor purposes. What AI genuinely accelerates: first-pass spec, figure-description boilerplate, claim-chart mapping, 112(b) antecedent-basis cleanup, and parallel claim-scope generation.

Is AI safe for prior-art search?

Yes for augmentation, no for substitution. Semantic prior-art tools (IPRally, Patlytics, XLSCOUT, Rowan, Solve Intelligence, LexisNexis PatentSight+, Clarivate Derwent, Questel Orbit Intelligence, Google Patents, The Lens) find non-obvious references faster than keyword search alone. But the registered practitioner still signs the IDS under 37 CFR 1.56 and 1.97. The duty to disclose material prior art known to any individual substantively involved in prosecution is personal — AI does not absolve it. The pragmatic rule: use AI to expand the search universe, then have a human review every material reference and decide what gets cited. Document your search methodology in the file.

Can AI draft an office-action response?

AI is useful for first drafts of §102 and §103 rebuttals once the attorney has decided the theory — it can produce claim-chart mappings, align the reference to the limitations, and surface Graham factor arguments. For §101 Alice/Mayo responses, AI helps with step-one abstract-idea framing and step-two significantly-more arguments but the attorney owns the strategy. For §112 written-description and enablement, AI assists with citation mapping to the specification. Every response still gets attorney-signed under 37 CFR 11.18, and every cited case gets Shepardized or KeyCited — citation hallucinations in Mata v. Avianca (SDNY 2023), Park v. Kim (2d Cir. 2024), and the Morgan & Morgan 2025 sanctions apply to patent practice as much as to litigation.

What is the ROI for a 2-25 attorney patent firm?

Real, and the gains are concentrated in three areas: (1) first-pass drafting — a prosecution attorney who used to bill 15-20 hours on a new application now bills 8-12 with equivalent or better claim scope; (2) office-action responses — typical OAs that took 6-10 hours now take 3-5 hours with claim-chart and argument drafts pre-populated; (3) prior-art search — searches that cost a $200-400 outside vendor can now run internally in 30-60 minutes. The firms pulling ahead in 2026 are not the ones using the most AI — they are the ones with written policy on AI use, per-matter billing transparency under ABA Model Rule 1.5, and a QA workflow that catches the hallucinations before they reach the USPTO.

Sources & further reading

USPTO, Guidance on Use of Artificial Intelligence-Based Tools in Practice Before the USPTO, 89 Fed. Reg. 25,609 (April 11, 2024).
USPTO, Inventorship Guidance for AI-Assisted Inventions, 89 Fed. Reg. 10,043 (February 13, 2024).
USPTO, Manual of Patent Examining Procedure (MPEP), 9th ed., Rev. July 2022 (current).
ABA, Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024).
37 CFR Parts 1, 11, 41, 42 — USPTO Rules of Practice + Rules of Professional Conduct + PTAB.
35 USC §§101, 102, 103, 112, 115, 184, 202, 257, 311-319 (IPR), 321-329 (PGR).
USPTO, 2019 Revised Patent Subject Matter Eligibility Guidance + 2024 Updates.
Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022); Mata v. Avianca, 678 F. Supp. 3d 443 (S.D.N.Y. 2023); Park v. Kim, 91 F.4th 610 (2d Cir. 2024).
ABA Model Rules of Professional Conduct (current) — Rules 1.1, 1.3, 1.4, 1.5, 1.6, 3.3, 5.1, 5.3, 7.1.

Related guides

← Back to all guides

Some links may be affiliate links that support the site at no cost to you.