India's AI Guidelines for Patent Examination

Published 2 October 20267 minute read

The Indian Patent Office now has written rules on how its examiners and Controllers may use artificial intelligence. The AI guidelines for patent examination in India allow AI as an assistant. They do not allow it to decide.

In August 2026 the Office of the Controller General of Patents, Designs and Trade Marks (CGPDTM) released the Guidelines for the Use of Artificial Intelligence in Patent Examination Procedure. The document governs AI use in screening, classification, search, translation, drafting support, technical comparison and knowledge retrieval. Its stated aim is to support efficiency and quality while preserving confidentiality, accountability, consistency and the independent application of mind.

These guidelines are about how the Office examines. They are not the 2025 Guidelines for Examination of Computer Related Inventions, which deal with when an AI or software invention is patentable. The two documents should not be confused.

Key points

AI may assist examiners and Controllers. The guidelines say it is meant to assist, and not replace, their functions.
The officer remains fully responsible for every official act that involves AI, and must personally verify AI output before using it.
Unpublished application material must not be entered into public AI tools. Only approved internal or secure tools may be used for it.
Material use of AI is to be recorded, and the guidelines contemplate that this record may be made available to stakeholders.
Applicants and agents should read every FER citation and quotation with care and ask for the basis of any objection that is not reasoned.

What the guidelines permit

The guidelines list a set of permitted uses. Each comes with its own safeguard. The main ones are these.

Classification. AI may suggest IPC or CPC codes. The officer must check them against the official classification scheme and against the claims and specification.
Search terms and prior art search. AI may propose keywords and run searches in subscribed tools. The officer frames the query, tests different formulations and selects the final results.
Claim feature extraction. AI may break a claim into features. The officer must reread the claim and check numerical ranges and dependencies before any analysis.
Translation. Machine translation may be used. Reliance on it must be recorded in the reasoning, and the text must be read with the drawings.
Novelty and inventive step. AI mapping may serve only as preliminary input. Each mapped feature must be checked in the actual prior art document. The final conclusion must rest on the officer’s own analysis.
Clarity and sufficiency. AI may flag possible issues. The officer must read the full specification and decide independently whether an objection arises under the Act.
Drafting and case law. AI may polish the language of a draft the officer has already written, and may help with orientation on case law. Every citation and quotation must be checked in the original source, and discarded if that is not possible.

What the guidelines prohibit

Section 5 lists the prohibited uses. An officer must not enter unpublished application content, confidential office records or internal deliberations into a public AI tool. AI must not substitute for the officer’s own judgment on novelty, inventive step, industrial applicability, sufficiency, clarity or unity of invention. No FER, hearing notice, decision or other official communication may issue on AI output alone without adequate human oversight. Case law, prior art and literature must not be cited without independent verification. AI must not be the sole basis of a decision that affects the rights of an applicant, a patentee or a third party, particularly in opposition proceedings.

The guidelines are frank about the risks. They list hallucination, omission of critical detail, false pattern matching, drift in classification and search, black box opacity, bias and difficulty with patent language. Annexure I gives worked examples. One shows different AI tools suggesting inconsistent classifications for the same applications. Another shows a generative AI tool producing quotations that it attributed to a 2020 IPAB judgment in Pharmacyclics LLC v Controller General of Patents. The tool admitted the fabrication only after repeated prompting. The lesson the Office draws is simple. Check the primary source.

Accountability, records and oversight

The administrative measures are as important as the rules on use. Material use of AI in specified functions is to be recorded. The record may include the tool, the nature and date of use and other details needed for audit or quality review. The guidelines state that this disclosure may be made available to stakeholders. Annexure II adds a checklist and a declaration for officers.

An AI governance committee drawn from the examination, IT and quality divisions will approve tools, classify uses, set safeguards, run pilots, review errors and complaints, and revise the guidelines. Officers are to be trained in the limits of AI, confidentiality, prompt discipline and the detection of fabricated citations.

What the guidelines do not decide

The guidelines bind the Office, not applicants. They do not regulate how applicants or patent agents use AI in drafting or in replies. They do not change the Patents Act, 1970 or the Patents Rules, 2003. An examination report is still issued under sections 12 and 13. The Controller must still offer a hearing under section 14 before refusing an application, and a refusal can still be appealed to the High Court under section 117A. The guidelines also do not name the approved tools, and they do not say when the recorded AI use will actually be shared with an applicant.

Who benefits, and the limits

Applicants gain most from the confidentiality rule. Before publication under section 11A, an application is not open to the public. The ban on public AI tools protects that period. Applicants also gain from the insistence on reasoned, verified objections. An objection built on an unverified AI mapping should not survive a careful reply.

Examiners gain speed in classification, search and translation, which are the most time consuming parts of the work. The Office gains consistency and an audit trail. The limits are practical. Tools can miss documents, and a search shaped by AI suggestions may drift. The guidelines answer this with human verification, but their value depends on how strictly they are followed in daily practice.

How other offices approach AI in examination

United States

The USPTO has gone further in sharing AI output with applicants. Under its Automated Search Pilot Program, announced in the Federal Register on 8 October 2025, eligible applicants receive an Automated Search Results Notice before examination. It lists up to ten documents found by an AI tool. Examiners consider them like other documents in the search file. The applicant need not respond. Separately, the USPTO’s April 2024 guidance reminds practitioners that their existing duties, including candour and confidentiality, apply when they use AI tools themselves. Our note on US AI patent inventorship guidance covers the related question of human contribution to inventions.

Europe and the United Kingdom

The EPO has used its ANSERA search system, which applies AI driven concept searching, since 2016. In 2023 the UK IPO launched its own SEARCH tool built on ANSERA. It ranks results by similarity and relevance. In both offices the examiner still reviews the prior art and takes the decision. We have not found a published EPO or UK IPO code on examiner AI use as detailed as the Indian guidelines, with a list of prohibited uses, recording duties and officer declarations.

The difference matters for applicants who file one family in several offices. Each office may now reach the art by a different AI assisted route. Inconsistent search results across the family are more likely, and should be compared with care.

Practical steps for applicants and patent agents

Read every FER citation in full. Confirm that each cited document exists, is prior art on its dates and discloses the feature it is said to disclose.
Check every quotation from a judgment against the official text. If a quotation cannot be found, say so in the reply and ask the Controller to withdraw reliance on it.
Where an objection maps features without reasons, ask for the basis of the mapping. The guidelines require the final view to rest on the officer’s own analysis.
Where translated prior art is cited, check whether machine translation was relied on and whether the translation is accurate. Raise errors with the drawings and original text.
At hearings, keep arguments human and verifiable. Rely on the claims, the specification and the cited documents, not on AI summaries.
Apply the same discipline to your own work. Do not put unpublished drafts or confidential invention details into public AI tools, and verify every authority before filing.

Our step by step guide to the Indian patent filing procedure explains where the FER and hearing fit in the timeline. Applicants who want support with FER replies and hearings can see our patent drafting and prosecution services.

Frequently asked questions

Does the Indian Patent Office use AI to examine patent applications?

Yes, as an assistant. Under the 2026 guidelines, examiners and Controllers may use AI for tasks such as classification, search and translation. The officer must verify the output and remains responsible for every objection and decision.

Can an FER be issued by AI alone?

No. The guidelines prohibit issuing an FER or any other official communication solely on AI output without adequate human oversight. Substantive findings must rest on the officer’s own analysis.

Are these the same as the CRI Guidelines on AI inventions?

No. The CRI Guidelines of 2025 explain when computer related and AI inventions are patentable. The 2026 guidelines govern how the Office itself uses AI during examination.

India’s AI guidelines for patent examination. What examiners may do and what applicants should check.