Adaptive.
It should sound like you wrote it.

"I can always tell which firm drafted this."

I heard that from an examiner at a conference. Not complaining; observing. She had been at the USPTO for eleven years and had examined thousands of applications across semiconductor technology. She said that after a while, you start recognising practitioners. Not by name; by style. The way certain attorneys structure independent claims. The way they deploy functional language. The way the background section sets up the technical problem before anyone mentions the invention.

That style is not an accident. It is the accumulated output of hundreds of matters, dozens of prosecution histories, years of learning what works and what doesn't in a specific technology space, before a specific examiner population, under a specific legal standard. It is, in a real sense, the most valuable thing an experienced patent attorney has.

And most AI drafting tools ignore it entirely.

Voice in patent drafting is not a stylistic preference

When I say an attorney has a voice, I do not mean they write beautifully or have a distinctive sentence rhythm. I mean they have made hundreds of strategic decisions; about when to use functional claiming versus structural claiming, about how broadly to write the independent claim before the first office action, about when to add a dependent claim that narrows elegantly versus one that narrows unnecessarily; and those decisions, accumulated across years of work, form a coherent and defensible philosophy.

That philosophy is reflected in the drafts they produce. And when a client's patent is prosecuted by that attorney across multiple applications, the consistency between those applications matters. Examiners notice it. Licensing counsel reading the portfolio notice it. District courts, if it ever comes to that, will notice it.

A generic draft does not have that philosophy. It has whatever the language model extracted from the aggregate of millions of patents; which is to say, it has everybody's approach and therefore nobody's.

What generic patent language actually costs

The problem with generic patent language is not that it is wrong. It is that it is average; and average is a very specific risk in patent drafting. When a claim is drafted in the average way, it makes the average assumptions. The average assumption about how broadly an independent claim can go before it becomes obvious. The average assumption about how many dependent claims to add and what they should cover. The average assumption about the level of detail required in the specification to support a claim under enablement.

For an attorney who has practised in this technology area for fifteen years, "average" is almost certainly wrong in multiple specific ways. They have learned through experience exactly where the average assumptions fall short. A draft that ignores that learning is not a neutral starting point; it is a draft that asks the attorney to go through the entire thing and fix the places where it diverged from their judgment. Which, depending on the technology and the attorney, could be most of the document.

What Adaptive means in eety

Adaptability in eety works at three levels, each building on the one before.

• Style library from prior work. The attorney uploads a set of their prior granted applications and prosecution histories. eety analyses the linguistic patterns; claim structure, use of functional versus structural language, the architecture of dependent claim sets, the way the specification relates to the claims. It extracts the attorney's characteristic approach and stores it as a style model.
• Applied on every new matter. When eety drafts for that attorney, the style model shapes the output. Not by filling in template sentences; by influencing the structural and strategic decisions: how the independent claim is scoped, how the dependent claims are layered, how the specification builds its evidentiary record. The draft arrives already calibrated to that attorney's approach.
• Refined through use. When the attorney makes direct edits to a draft; changes a claim scope, restructures a dependent chain, rewrites a section of the specification; eety records those as decisions, not corrections. Over time, the style model becomes more accurate. The tool learns not just from what the attorney said before, but from what they chose to do differently this time.

The output the attorney actually recognises

The test I use for whether the Adaptive pillar is working is simple. When an attorney reads the draft eety produced, do they recognise it as consistent with how they would have approached it? Not identical; they will always have things to change. But consistent in its underlying philosophy.

If the answer is yes, the tool is saving them the work of re-orienting a generic draft toward their approach. They are starting from a place that is already close to where they want to be; and making the specific changes that reflect the specifics of this invention and this client.

If the answer is no; if the draft feels like it was written for a different attorney with a different philosophy; then the tool is not actually saving work. It is creating it in a slightly different form.

Generic is not neutral. It is just someone else's judgment, applied to your matter, with your name going on the result. :)