AI Tools explained: this guide covers what it means, who it applies to, the step-by-step process, documents required, fees, due dates and penalties in India — so you can stay compliant with confidence and avoid costly mistakes.
After describing the risks, the Patent Office's AI guidelines turn to practice. Paragraph 4 lists twelve typical uses of AI in examination, each with the benefit, the risk and the safeguards; paragraph 5 lists what is not permitted at all; paragraph 6 sets out administrative measures. This article summarises all three in plain terms and explains what each means for an applicant. The guidelines concern how examiners and Controllers use AI tools, not whether AI inventions are patentable.
The Guidelines, as published by the Patent Office, are the Office's guidance and do not have the force of law; the Patents Act, 1970 and the Patents Rules, 2003 as now in force prevail. The Office revises its guidelines, so check the current version on ipindia.gov.in.
The Office may use AI for preliminary tasks such as suggesting classifications, search terms, claim features, translations and issue-spotting, but each use carries safeguards, and a verified human decision must follow. Unpublished application content may not be entered into public AI tools, and AI may not substitute for the officer's application of mind on novelty, inventive step, industrial applicability, sufficiency, clarity or unity.
The safeguards pattern
Before the table, the Guidelines note that the patent search databases subscribed by the Office have built-in AI tools designed for patent search, and these private tools offer better data security and a better grasp of patent prompts. The twelve uses in paragraph 4 then follow a common pattern: AI gives a first suggestion, the officer verifies it against the original, and the officer's own analysis decides. The Guidelines refer to an Annexure-II containing a checklist and declaration; its text is not reproduced in the Guidelines as available and is not described here.
The twelve uses: TaxClue's table
| Use | Main risk named | Safeguard in short |
|---|---|---|
| 1. Suggesting IPC/CPC classifications | Wrong or overbroad classes that drift the search | Check against an authenticated classification source; verify against claims and specification; adjust manually |
| 2. Generating search terms and concept clusters | Terms too broad or too narrow | Officer selects terms, separating close, merely adjacent, exploratory and noisy terms |
| 3. Extracting claim features | Wrong splitting, loss of relations between features | Re-read the claim; verify ranges and dependencies; no analysis on an unverified feature list |
| 4. Translation support | Loss of technical nuance | Read with drawings and surrounding text; record machine translation if relied on |
| 5. Polishing a draft office communication | Unsupported statements, imprecise language | Use only after the officer's own draft; limit to grammar and readability; check every statement |
| 6. Finding legal or technical citations | Fabricated citations, wrong attribution | Do not use unless verified in the original source; discard if verification is not feasible |
| 7. Public AI tool with unpublished material | Disclosure of unpublished or internal material | Prohibited; only approved secure tools; if in doubt, do not enter |
| 8. Preliminary novelty or inventive-step mapping | Misreading, hindsight, persuasive but unsound reasoning | Treat as preliminary; verify each mapped feature in the document; examine amendments and submissions independently; final conclusion on the officer's own analysis |
| 9. Spotting clarity issues | Overstated objections, foreign standards | Preliminary only; compare with the full specification; reformulate in the Office's style with statutory provisions; no objection unless personally satisfied |
| 10. Prior-art search with officially subscribed AI tools | Query dependence, over-reliance on ranking | Frame and refine queries; assess results manually; try several formulations; select results by independent judgment |
| 11. Preliminary check of sufficiency | Over-reading or under-reading gaps | Examine the full specification, drawings and general knowledge manually; independent technical and legal assessment before objecting |
| 12. Understanding case law and concepts | Fabricated quotations, misattributed paragraphs | Orientation only; check the primary source; read in the facts of the case; discard fabricated material |
Prohibited uses (paragraph 5)
The Guidelines list six uses that are not permitted:
- Entering unpublished application contents, confidential office records or internal deliberative material into public AI tools.
- Using AI as a substitute for the officer's application of mind on substantive matters such as novelty, inventive step, industrial applicability, sufficiency, clarity or unity.
- Issuing office actions, FERs, hearing notices, decisions or other official communications solely on AI output without adequate human oversight.
- Citing case law, prior art, scientific literature or other references suggested by AI without verifying them in authentic sources.
- Using AI-generated content in official communications without review, correction and adoption by the officer.
- Relying solely on AI for decisions affecting the rights of applicants, patentees or third parties, especially in cases with opposing parties, for instance opposition proceedings.
The first item links directly to confidentiality. Unpublished applications are confidential until publication, and examiner reports are not open to the public; see our article on confidential examiner reports. The sixth item is the one that matters most for a contested hearing: the officer's decision in an opposition must be the officer's own.
If you are preparing for an opposition or a hearing and want the evidence and submissions built to be tested independently, our patent objection reply service can assist.
Administrative measures (paragraph 6)
The Guidelines list steps the competent authority may take:
- Records of use. It may prescribe that material use of AI in specified functions be recorded, with the tool name, the nature of the use, the date and other supervisory details, and the disclosure may be made available to stakeholders to build confidence in examination.
- A governance committee. It may set up a committee to approve tools, divide uses into permitted and prohibited classes, prescribe safeguards, run pilots, review errors and complaints, revise the Guidelines and consult experts. The Guidelines describe an AI Governance Committee in the office of the Controller General, made up of examiners and controllers from the examination division and officers from the IT and quality management divisions, spread across technology groups.
- Training. It may train officers on AI limitations, confidentiality and prompt discipline, verification, responsible use and spotting hallucinations and fabricated citations.
- Audit and feedback. It may provide independent audit, impact assessment, feedback and incident reporting.
The word used throughout is "may": these are enabling measures, not commitments an applicant can enforce.
What it means for applicants
- Your unpublished specification should not be sitting in a public AI tool at the Office, and you should not put it in one either while preparing a reply; the same confidentiality logic applies on your side.
- Where an objection looks unsupported, the Guidelines give you the vocabulary: unverified citation, omitted limitation, preliminary mapping presented as final.
- The decision remains the officer's; an applicant can ask at a hearing for the reasoning on novelty or inventive step in the officer's own words.
- Do not overstate. These Guidelines do not make an office action invalid because AI was used. They set conduct rules for the Office.
A worked example
Orchid Photonics Pvt Ltd receives an FER citing a foreign publication for every feature of claim 1 and a decision-style paragraph on obviousness. The agent notices that the mapping treats a wavelength range in the publication as the claimed range, though the document gives a different range for a different embodiment. The reply sets out the difference, relies on the officer's duty to verify each mapped feature, and requests a hearing. At the hearing the Controller reviews the publication directly and narrows the objection to one claim.
Common lapses
- Pasting unpublished claims into a public AI tool while drafting a response.
- Treating the Guidelines as a ground to void an office action.
- Replying only that "AI was used" with no substantive points.
- Overlooking that classification and search suggestions are preliminary and do not bind the Office.
- Expecting the administrative measures to be mandatory.
Need help with objections and hearings?
A well-built reply tests every citation, mapping and conclusion in an office action. If you would like a review of an objection and its cited documents, speak to our patent objection reply team.
Key takeaways
- AI is allowed only as a preliminary aid, with a stated safeguard for each of twelve uses.
- Public AI tools may not be given unpublished application material.
- Six uses are prohibited, including decisions solely on AI output and unverified citations.
- Administrative measures such as use records, a governance committee and training are enabling.
- Applicants can test an objection on verification and reasoning, not on the mere use of AI.
Read next
- AI in patent examination: scope, terms, risks and human oversight
- AI in patent examination: Annexure I illustrative examples
- Chapter 9: examiner's report, FER, reply and hearing
- Chapter 10: post-grant opposition, Board and hearing
Disclaimer: Based on the manuals and guidelines published by the Office of the Controller General of Patents, Designs and Trade Marks that are named in the article, as consulted on 4 October 2026. They are guidance and do not have the force of law; the Patents Act, 1970 and the Patents Rules, 2003 as amended (including the 2024 amendment rules) prevail, and the current versions on ipindia.gov.in should be checked. This article is general information, not legal advice; check the official text before acting.