Blog

AI in the Indian Patent Office: What the New Examination Guidelines Mean for Applicants

IPR Manager Team · 6 Oct 2026

AI in the Indian Patent Office: What the New Examination Guidelines Mean for Applicants

India's Patent Office now lets examiners use AI, but the decision on your application must still be a person's own. The same rule is a good one for your side of the file.

What the guidelines say: assist, never replace

On 7 August 2026, the CGPDTM issued Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures. Their core principle is that AI may assist examiners and Controllers but cannot replace their independent application of mind.

The guidelines allow AI for preliminary, supporting work, each with a safeguard attached:

Task AI may help with What the officer must still do
Suggesting IPC/CPC classification Verify against the official schemes
Generating search terms and running prior-art searches Judge relevance and select the final citations
Extracting claim features Re-read the actual claims
Translating foreign-language prior art Read it against the drawings and context
Preliminary novelty and inventive-step analysis Reach an independent conclusion
Finding case law and legal citations Verify each one from the primary source
Improving the wording of draft communications Prepare the reasoning first and verify every statement

The guidelines also set up an AI Governance Committee to approve tools, run pilots and review errors. They call for material AI use to be recorded, including the tool, the nature of the use and the date.

Note that these are different from the 2025 Guidelines for Computer-Related Inventions. Those govern whether AI inventions are patentable. These govern how examiners work.

The red lines

The guidelines name six things examiners and Controllers must not do:

  1. Enter unpublished applications or confidential records into public AI tools.
  2. Use AI in place of their own assessment on substantive patent questions.
  3. Issue an official communication or decision based solely on AI output, without adequate human oversight.
  4. Cite case law or prior art that AI suggested without verifying it from the authentic source.
  5. Put AI-generated text into an official communication without reviewing and adopting it.
  6. Rely solely on AI for decisions that affect an applicant's or patentee's rights, especially in oppositions.

The guidelines are candid about why. They list failure modes such as hallucinated citations, dropped claim limitations, false pattern matching and outdated training data. They even give an example of a tool inventing case-law quotations, complete with citations and bench details, all of them false.

What this means when you read an FER

If examiners use AI for searching, classification and claim mapping, the errors AI tends to make can show up in examination reports. Those are the errors to look for when an FER arrives.

  • Check every citation. Confirm each prior-art paragraph and each case quoted actually says what the FER claims. The guidelines themselves treat unverified citations as a known risk.
  • Check the search direction. If the cited art looks non-analogous, ask whether a broad or drifted classification led the search astray, and say so in the reply.
  • Check the claim reading. Automated feature extraction can drop relational words such as "wherein" or "operably connected". Map the objection back to every limitation of the claim.
  • Check the translation. Machine translation can blur the difference between "attached", "bonded" and "coupled".
  • Answer the full statutory test. A fluent objection is not necessarily a sound one. Respond on novelty and inventive step in full, not just to the wording of the FER.

None of this assumes bad faith. It simply means a careful reply now includes checking the analysis, not only the conclusions.

Using AI responsibly on your side of the file

The guidelines bind the Patent Office, not applicants. But the principles carry over well to drafting and prosecution. Five habits follow directly from them.

  1. Protect the invention before you prompt. An unpublished specification is as confidential to your client as it is to the office. Only use AI tools whose terms say your data is not used for training and is handled securely. Keep consumer chatbots away from disclosures, drafts and strategy notes.
  2. Verify everything that will be filed. Every case citation, prior-art passage and statutory reference that an AI suggests is a lead, not a fact. Check it against the source before it goes into a reply.
  3. Keep the judgement yours. Let AI produce a first draft of claims, a support map or an objection summary. The scope of the claims, the arguments and the amendments remain the practitioner's decisions.
  4. Draft for careful reading by machines and people. Clear definitions, intact relational language, layered fallback positions and worked examples all make a specification harder to misread, whoever is reading it.
  5. Keep a record. Note which tool helped with which part of a draft or response and who reviewed it. It costs little and answers the client's questions later.

The Patent Office has drawn its line between assistance and judgement. Firms that draw the same line in their own work will find it easier to explain their AI use to clients.

AI that keeps the practitioner in charge

IPR Manager is built on the same principle as the Patent Office's guidelines: AI does the groundwork and a practitioner makes the call.

  • Our AI data policy sets out where your clients' inventions go and how they are protected.
  • Deadlines read from Patent Office documents show the source line for every date, and nothing is saved until you confirm it.
  • Every AI output, from an FER analysis to a draft claim, is kept with its review state, so you can see what was checked and by whom.

Read our AI data policy · Start a free trial of IPR Manager

Sources