LAST REVIEWED: OCTOBER 2026
By Dr Ritchie Lewis Daniel, Chartered Patent Attorney and European Patent Attorney
Yes. Artificial intelligence (AI) can help organise invention information, explore terminology and produce a draft. Generating patent-shaped text is not the difficult part. The value lies in the quality of the protection, not the polish of the document.
Short on time?
- Use AI tools such as ChatGPT to organise technical material, explore terminology and challenge how the invention is described.
- The difficult part is deciding what deserves protection, asking the right technical questions and judging whether the claims and disclosure are commercially useful.
- For a commercially important invention, focused patent-attorney input before filing can be especially valuable on the inventive concept, independent claims, prior art and supporting disclosure.
- Before uploading unpublished invention details, check confidentiality, retention and access arrangements.
There is evidence of useful assistance. A September 2026 field experiment involving 133 patent lawyers at 11 US firms found that AI access improved expert-rated drafting performance. Google funded the direct costs. The paper is a working paper and has not been peer reviewed. The scores do not establish later commercial value or enforceability.
IPReg's current policy takes a similarly pragmatic position. AI may bring time and cost savings, but regulated practitioners remain responsible for quality, checking outputs and protecting confidential information.
My perspective comes from mechanical and electromechanical patent practice. The question I would ask is whether the application gives you a sound basis for protecting what matters to your business.
First, understand what is worth protecting
When an inventor brings me a prototype, I want to understand the story behind the product. What problem were they solving? Why did they choose this arrangement? What did they try and discard? Which benefits matter commercially?
The inventor usually understands the product far better than I do. My job involves a lot of listening. A feature they consider routine may deserve closer investigation, and the contribution may lie in how familiar components work together.
Consider a hypothetical quick-release coupling for a hand tool. The inventor might emphasise how quickly an attachment can be changed. If the information supplied to AI centres on speed, the resulting draft may keep developing that theme. A search could instead show that similar release mechanisms are already known.
Discussion may then draw attention to a separate arrangement for adjusting the coupling's clamping force. That feature could deserve greater attention depending on the prior art, commercial value and how readily competitors could work around the resulting claim.
AI can help structure that questioning, but the analysis still depends heavily on the information supplied and the questions asked. A founder may be excellent at using AI and still not know which fact matters, or which question needs to be asked.
Handling a prototype can reveal movement, resistance or an adjustment that the inventor has not thought to describe. Video can convey much of that information, but a physical mechanism can expose points that were never put into words. Patent searching then helps frame those observations against what has gone before.
Start with the claims
The claims define the protection sought, interpreted using the description and drawings. I spend substantial time on the independent claims: those that stand on their own rather than adding detail to another claim.
For a mechanical invention, I would ask whether the claim captures the important component relationships and how the mechanism operates. Does the claim include a particular material or shape simply because the prototype uses one? Is the terminology clear? Would a commercially sensible alternative fall outside the wording? Does the claim target the product or activity that matters?
Removing unnecessary restrictions can help, but breadth has to be justified by the invention and the prior art. The description must also support the claims with enough technical detail, alternatives and narrower positions if objections arise.
These foundations matter after filing. Under European added-matter rules, an amendment must be supported by the application as originally filed. UK practice applies the same underlying restriction under section 76 of the Patents Act 1977. You cannot simply add a missing technical arrangement when the feature becomes important during examination.
A technically capable founder can use AI to organise material and develop a draft. The harder task is assessing the output. A polished claim may contain an unnecessary limitation, while a convincing description may omit a fallback position needed later. Someone without patent training may have no reliable benchmark for deciding whether the draft is broad, narrow, unsupported or directed at the wrong inventive concept.
That is where focused professional input can have disproportionate value. A patent attorney knows what information to ask for, how to frame the invention against the prior art and how to stress-test the claims and disclosure. If the budget is limited, I would spend professional time on the invention and independent claims before investing in a complete AI-generated draft.
Where AI helps
In my own practice and testing, I have found AI useful for analysing published patents, reviewing patent-office objections and exploring arguments that a patent may be invalid. AI can offer a second perspective and flag something worth checking.
That checking means returning to the original document. Has the earlier patent actually disclosed the feature, or has AI made an unsupported assumption? A preliminary search can provide a steer, but does not amount to a complete patentability assessment or an opinion on whether your product infringes existing patents.
Where AI drafts can go wrong
Much of the thinking in patent drafting happens through the writing. Reviewing someone else's draft requires understanding the invention and why each feature and phrase is there. A long AI-generated draft can therefore create substantial checking, particularly where the text introduces unnecessary restrictions, inconsistent terminology or unsupported technical benefits.
CIPA's guidance notes that some AI-related weaknesses may emerge only years later during licensing or litigation.
A document can look finished on filing day while leaving a weakness that becomes apparent during examination or a competitor's design-around. I would also be cautious about entrusting amendments to AI: similar wording does not by itself establish legal support for a change.
Will AI save money?
AI may help strengthen an application before filing, but I would be slower to promise that AI will make the overall job cheaper. Time saved generating text can instead be spent understanding the invention, stress-testing the claims, considering alternatives and checking the draft. For a startup with a limited budget, focused professional input on the invention and independent claims can be a better use of funds than paying for extensive review of a long AI-generated specification.
Check confidentiality before uploading an invention
Before uploading unpublished invention information, check the service terms, settings and confidentiality arrangements. In UK and European patent practice, information made available to the public before the relevant filing or priority date can become prior art. Confidentiality therefore matters to patentability as well as privacy. See the UKIPO guidance on novelty and the EPO guidance on public availability and confidentiality.
Whether information supplied to an AI service has been made available to the public depends on the service and the circumstances. Do not assume that a no-training statement answers that question. “Not used for training” does not mean “not retained”. Consumer and business services can also have different terms.
Business services, including OpenAI's business products and Anthropic's commercial products, publish default no-training commitments. Retention, access and contractual confidentiality still need separate consideration.
Once a patent application has been published, the information in that publication is already public, so using AI to analyse that published material raises far less concern about disclosure. Unpublished improvements, commercial plans and legal analysis may still remain confidential.
Using AI alongside a patent attorney
For a startup or individual inventor, AI can make preparation more accessible. Bring a clear account of the problem, drawings, development history and alternatives. Professional involvement can vary with budget and risk, but for a commercially important invention I would want a patent attorney involved before filing.
AI can then be used as a tool within that process. The founder supplies technical knowledge and commercial priorities; the patent attorney identifies missing inputs, frames the claims against the prior art and checks whether the output provides useful protection.
General information only. This article does not constitute legal advice. The discussion focuses on mechanical and electromechanical inventions in UK and European patent practice. AI services and their terms change.