18 August 2026
TLDR: don’t use LLMs. Your own thoughts in your own words is where the value is. LLMs just embed your invention in noise.
If you have come up with an invention and want to understand how to use intellectual property rights to protect it, the thought of briefing an IP attorney can be daunting if you have never done it before. What do you tell them? What do you not tell them? What documents must you prepare? What do you even want from them? In the face of such questions, it can be tempting to reach out to Claude, ChatGPT or other large language model- (LLM-) based tool.
Truth is, LLMs don’t know what your IP attorney needs from you and they don’t know the best way to brief a lawyer. They can come up with something that looks credible based on probabilities. But the result, typically a longwinded email, is rarely of any value to us and makes it difficult to decipher exactly what you are looking for. In reality, this is the opposite of what your patent attorney needs and usually adds to costs as significant time can be wasted in trying to figure out exactly what it is that is the invention.
Of course, LLMs offer enormous potential for very specific applications. But they can also be dangerous and incredibly wasteful of resources.
So, in response, we have created this guide on how to brief a patent attorney in the age of LLMs.
- It is ok if you feel like you don’t know what you are doing.
When instructing us on a new matter, we don’t need a lengthy, multi-point form email setting out your objectives and expected deliverables. The verbosity of LLMs means that invariably what could have been an email of a few sentences ends up being unnecessarily long and complex.
We would far rather spend time deciphering what you want from a short email in your own words than from a long one written by an LLM. In these circumstances, less really is more. And if we do need more, we will give you call.
- Your idea in your own words. Written or verbal.
It is easy to turn to LLMs to flesh out your idea of an invention into what looks like a nicely written disclosure. However, the effect of this is that your invention is embedded into a layer of LLM-generated noise. This makes it very difficult for us to distil your contribution and what it is you want to protect from the common general knowledge that the LLM adds into the mix.
Further, an invention must be kept confidential up until the first patent application has been filed for it. Inputting your invention into an LLM risks breaking this confidentiality and precluding patent protection for the invention. This is particularly so for certain models and subscription tiers which allow training on your inputs, but remains a risk even for enterprise-grade implementations that are outside of your control.
If your invention can be communicated to an LLM in a few prompts for generating the LLM-generated disclosure, then it can easily be conveyed to us in a call or a written disclosure of a few paragraphs. And, because inventions are creations of the human mind, your description is where the value is.
Any invention builds upon existing technology. In general, then, even the most complex of inventions can be captured in a few paragraphs which describe what is being added to the prior art and what the associated advantages are. That is all we need to start drafting a robust patent specification.
- LLMs are not very good at generating patent work products such as patent specifications, freedom to operate (FTO) opinions and the like.
To the untrained eye, a draft patent specification spat out by an LLM looks fantastic. In reality, they are of almost no value. A skilled patent attorney writes each sentence of a patent specification with a purpose in mind by drawing on their years of experience and taking into account the particulars of the invention in question. That is, apart from some boilerplate paragraphs here and there, each sentence is specifically crafted to serve a particular purpose. LLMs have no insight into these considerations. Yes, they may have been trained on millions of patent specifications so can adopt the turn of phrase used in patent filings in the same way they can write poems in the style of Shakespeare. But each of those patent specifications is unique, and the LLM has no sight of the individual considerations that went into each of them. I.e., the LLM does not know why they say what they say. It only sees the end-product. The result is that the LLM is really just adding text for the sake of adding text, spinning the content of your prompts into more words with the goal of having the appearance or the look and feel of a patent specification.
Further, of the millions of patent specifications, many of them aren’t very good and there is no differentiation or labelling of good examples versus bad examples. So even if an LLM could learn to draft patent specifications from these examples, there is no guarantee that the output meets professional patent-drafting standards.
As for other patent work products, such as freedom to operate opinions, patentability searches, and the like, there are very few publicly available examples of these. Most of them remain confidential and stay between the attorney and client. So, in these cases, LLM generated versions don’t even resemble the required end product. We also caution that some patents, and in particular South African patents, are generally not publicly available online. Asking an LLM what patents you should be concerned about from an infringement perspective will not give you the correct answer, and may possibly lead you down the wrong track.
- LLM-generated feedback on our work products is not helpful.
As with anything, if you ask an LLM to give feedback on a patent work product as if it were an experienced patent attorney, it will oblige and produce something that looks constructive. But in our experience, the feedback is generally superficial and overlooks the key points on which we typically want the inventor’s input and on which only the inventor can weigh. This leads to lengthy analyses of the draft specification with all manner of suggestions that we must work through, generally with no major improvement to the draft. This is compounded by the sensitivity of LLM-generated outputs to the language of the prompts by which they are instructed. Two different prompts to review the same draft patent specification will yield very different feedback.
Another challenge is when the invention has been conceived of with the help of an LLM. In these circumstances, when our work product is input into the same LLM for feedback, we have seen the LLM referencing features of the invention in its feedback that have not been disclosed to us before. This is probably because of the LLM drawing on information present elsewhere in the conversation, prior prompts, uploaded materials, or inferred assumptions, none of which have been conveyed to us. The result is constantly moving goalposts, which in turn leads to a patent attorney who is always a step behind and a client who is getting more and more frustrated.
- You won’t save on costs by starting off with an LLM-generated work product.
You might think that if you share an LLM-generated draft specification with us we can just give it a quick review and file your patent application, resulting in a massive saving in legal fees. This is not the case and, as explained above, we find that LLM-generated patent work products are of very little value.
Not only are LLM-generated specifications not very good, we find that they do not necessarily convey to us, or anyone else, what the key features of your invention may be. Therefore, for us to turn an LLM-generated draft specification into a quality specification worthy of filing will generally take more time and effort than if we prepared the application from scratch ourselves, based off your own disclosure. This increased time and effort ultimately translates into higher costs. The same is true for LLM-generated feedback on our draft patent specifications.
As a firm, we prioritise ensuring that our clients’ inventions are afforded the best protection possible by maintaining a high standard of patent drafting. We draft high quality patent specifications from the ground up, drawing on our years of experience in prosecuting patent applications and litigating patents in a wide range of territories. We use our experience to inform on what parts of the invention to invest time and emphasis. We take into account practical considerations that apply to the case and the client and draft the patent specification and formulate the filing strategy accordingly. This is not something we have seen any LLM do effectively.
Disclaimer: in the above, we are referring to general purpose LLMs that are widely available, not to any patent-specific LLM tools created for use by patent attorneys.
