AI Use for Patent Drafting & Prosecution – Client FAQ.
We've prepared this FAQ for Aurora clients to share our current understanding of how AI is impacting the patent drafting and prosecution world. It's our goal to be able to collaborate as effectively and responsibly as possible with our clients as workflows rapidly evolve around advancements in generative AI.
Our conclusions are based on lessons learned from client interactions, extensive research we conducted for a white paper published in the Thomson Reuters Practical Law Journal, and our ongoing evaluation of both general-purpose AI tools and those specifically crafted for patent practitioners.
Our conclusions are based on lessons learned from client interactions, extensive research we conducted for a white paper published in the Thomson Reuters Practical Law Journal, and our ongoing evaluation of both general-purpose AI tools and those specifically crafted for patent practitioners.
Can I use AI to draft my own patent application?
You can, but there are significant risks. AI tools can generate text that looks like a patent application, but they often lack:
- Strategic claim structure
- Proper support for future amendments and support for arguments in prosecution
- Awareness of enforcement and design-around risks
- Alignment with USPTO examination practices
A patent application is not just a document; it is a legal asset that must be carefully constructed from the outset. Mistakes made early are often difficult, or impossible, to fix later.
If AI produces a “complete” draft, isn’t that enough?
Not necessarily. AI-generated drafts often:
- Include overly broad or unsupported claims
- Miss key embodiments needed for flexibility during prosecution
- Use inconsistent terminology that weakens claim interpretation
- Fail to anticipate likely prior art or examiner objections
These issues may not be obvious initially, but they can significantly reduce the patent's long-term value or enforceability.
Can AI help me respond to USPTO Office Actions?
This is particularly risky. AI frequently:
- Misinterprets examiner rejections
- Applies incorrect legal standards
- Suggests arguments that are weak, irrelevant, or inconsistent with the record
- Fails to appreciate long-term prosecution strategy
A poorly handled response can narrow your claims unnecessarily or create prosecution history that limits future enforcement.
Why can’t AI replace a patent practitioner?
Patent practitioners develop expertise over many years (often decades), including:
- Deep knowledge of USPTO procedures and examiner tendencies
- Strategic decision-making (when to argue vs. amend vs. pivot)
- Field-specific experience (e.g., software vs. life sciences)
- An “arsenal” of approaches to overcome rejections
- Case law strategy reasoning
This judgment is built through experience and cannot be reliably replicated by AI tools.
What are the biggest risks of relying on AI alone?
Common risks include:
- Irreversible drafting errors that limit claim scope
- Inadequate disclosure that prevents future claim amendments
- Inconsistent terminology that weakens interpretation
- Strategic missteps during prosecution
- Reduced enforceability or commercial value
- Inadvertent public disclosure if you use the wrong AI, losing US and/or foreign rights
In many cases, these issues only become apparent years later, when it is too late to correct them.
If I use AI, can a practitioner fix it later?
Sometimes—but not always. A practitioner can often improve:
- Clarity and structure
- Claim strategy
- Consistency and support
However, they cannot add new subject matter after filing. If key concepts or variations were not included in the original application, they may be permanently lost.
Are there confidentiality risks when I use AI tools?
Yes, depending on the tool. Some AI platforms:
- Retain your data
- Use your inputs to train their models
- Do not offer enterprise-level confidentiality protections
Before using any AI tool, you should confirm:
- No data retention
- No training on your inputs
- Appropriate security safeguards
If these are not clearly stated, you should assume your information may not be fully protected.
Is there a safe way to use AI in the patent process?
AI can be useful when used carefully, for example:
- Organizing technical ideas before working with counsel
- Brainstorming variations or alternative implementations
- Preparing background materials
However, AI should be treated as a preliminary tool, not a replacement for legal expertise.
What approach do you recommend?
The most effective approach is typically:
- Use AI (if desired) for early-stage idea development
- Engage a practitioner to develop claims, strategy, and filings
- Rely on experienced counsel for prosecution and decision-making
This balances efficiency with quality and long-term value.
What are the cost implications of using AI in the patent process?
AI can both reduce and increase costs, depending on how it is used. In some cases, AI can improve efficiency. For example, when used appropriately by a practitioner, AI may help streamline the drafting of background sections or organize technical content. This can allow more time and budget to be directed toward higher-value work such as claim strategy, prosecution planning, and long-term portfolio positioning, resulting in even stronger, more robust patents for the same cost.
However, when clients provide large volumes of AI-generated material, costs often increase rather than decrease. Practitioners must carefully:
However, when clients provide large volumes of AI-generated material, costs often increase rather than decrease. Practitioners must carefully:
- Review and verify all content for accuracy
- Reconcile inconsistencies in terminology and concepts
- Evaluate whether the material is legally sufficient and strategically sound
- Heavily edit non-patent language and inject patent-ease language
This review process can take more time than working from a clean, concise invention disclosure.
The most cost-effective approach is typically:
The most cost-effective approach is typically:
- Provide a well-organized, curated, and succinct invention disclosure
- Use AI, if desired, to help structure ideas—not to produce full drafts
- Allow the practitioner to build the application strategically from the ground up
In short, AI delivers the most value when it reduces noise, not when it adds to it.
If I use a public AI tool like ChatGPT or Claude to enter my invention details, can my confidential information lose legal protection?
In Trinidad v. OpenAI, a court dismissed trade secret claims because the plaintiff had voluntarily uploaded her proprietary frameworks to ChatGPT without any confidentiality protections in place. In United States v. Heppner, a court held that documents created using Claude were not protected by attorney-client privilege because Anthropic's standard privacy policy permits data collection, use for model training, and disclosure to third parties.
The core lesson from both cases: sharing confidential information with a public AI platform, without appropriate safeguards, is legally equivalent to disclosing it to the world before securing your patent protection.
The core lesson from both cases: sharing confidential information with a public AI platform, without appropriate safeguards, is legally equivalent to disclosing it to the world before securing your patent protection.
Can't I just keep quiet about having used AI?
That does not solve the underlying problem. Publicly disclosing your invention before you file a patent application can cause you to lose your patent filing rights.
The legal exposure arises at the moment of disclosure to the AI platform, not at the moment you tell anyone about it. Once confidential information enters a public AI platform's environment, it may be:
The legal exposure arises at the moment of disclosure to the AI platform, not at the moment you tell anyone about it. Once confidential information enters a public AI platform's environment, it may be:
- Used to train the AI model
- Subject to disclosure to third parties under the platform's terms
- Treated by courts as having lost its protected status
Simply put, 35 U.S.C. §102 states that if your invention is known to the public before the filing date of your application, no patent can be had.
Does it matter that I didn't intend to waive protection?
No, intent is not the standard.
The Trinidad court applied the principle that when a party discloses trade secret information to others who are under no obligation to protect its confidentiality, the property right is extinguished. The plaintiff's consent to OpenAI's Terms of Service (Trinidad v. OpenAI), and the voluntary use of ChatGPT to develop the frameworks were sufficient to preclude trade secret protection, regardless of the party’s subjective intent.
The Trinidad court applied the principle that when a party discloses trade secret information to others who are under no obligation to protect its confidentiality, the property right is extinguished. The plaintiff's consent to OpenAI's Terms of Service (Trinidad v. OpenAI), and the voluntary use of ChatGPT to develop the frameworks were sufficient to preclude trade secret protection, regardless of the party’s subjective intent.
Key Takeaways: What should I keep in mind?
AI can help you move faster, but patents reward getting it right, not just getting it done.
The quality of your initial filing and prosecution strategy will directly impact:
The quality of your initial filing and prosecution strategy will directly impact:
- Scope of protection
- Ability to enforce your rights
- Overall business value of your patent
For an even deeper dive, check out our Prompting and Patents: What to Know Before You Ask AI podcast episode that explores where AI is genuinely helping in patent drafting and prosecution, where they’re seeing it go wrong, the surprising ways it can actually increase legal costs, what recent court decisions may mean for confidential information, and some practical guardrails for inventors who want the benefits of AI without inadvertently undermining the very intellectual property they’re trying to protect.