Patent prosecution is one of the few legal disciplines where the value of a single conversation can be measured in millions of dollars and decades of exclusivity. The hour an inventor spends explaining a novel architecture to outside counsel is not a meeting — it is the foundational record on which an entire patent family will be built. Get it wrong, and every continuation, divisional, and litigation strategy that follows inherits the gap.
This is the inventor disclosure gap, and it is the most expensive failure mode in patent prosecution.
The Problem
Patent attorneys conduct invention disclosure meetings under brutal conditions. The inventor speaks in dense domain language. The attorney is simultaneously listening, asking clarifying questions, and trying to capture claim-critical details on a notepad. Most disclosures last sixty to ninety minutes and cover material the attorney has never seen before.
What gets lost is rarely the headline invention. It is the dependent claim material: alternative embodiments, the inventor's offhand mention of a failed approach (which is often the strongest non-obviousness argument), specific numerical ranges, and the inventor's own characterization of why prior art does not anticipate. The independent claim is usually safe — everyone in the room knows the headline invention is the headline invention. The damage happens around the edges, where dependent claims and fallback positions live.
The cost is not theoretical. A narrowed independent claim can shave seven figures off the value of a patent family. A missed prior-art reference can tank a continuation. A mischaracterized embodiment can hand a competitor a design-around in writing. And because the loss often does not surface until office action — months or years later — it can be impossible to reconstruct what the inventor actually said in the original meeting.
What an Invention Disclosure Actually Sounds Like
To understand why generic note-taking fails, it helps to picture the linguistic density of a real disclosure. In a single ninety-minute meeting, an inventor might:
- Reference six to ten prior art systems by name, often by author and conference, sometimes only by paper title.
- Describe three to five distinct embodiments, each with its own variation in materials, geometry, or process parameters.
- Quote specific numerical ranges — "between 12 and 18 nanometers," "at temperatures above 850°C," "at concentrations of 10^16 to 10^18 dopants per cubic centimeter."
- Mention an approach they tried first that did not work, which is frequently the strongest evidence of non-obviousness.
- Use highly domain-specific vocabulary that consumer transcription tools have never seen: photolithography, eigenmode, methyltrioxorhenium, FinFET, Czochralski, monoclonal antibody, atomic layer deposition.
An attorney capturing this in real time is doing three jobs at once: listening, deciding what is claim-relevant, and writing. The cognitive load is the bottleneck — not the writing speed.
Why Current Solutions Fail
Handwritten notes
Even an experienced prosecutor captures maybe thirty percent of what an inventor says. The rest is reconstruction from memory, days later, often after the attorney has already started drafting. By the time a question comes back from the examiner, the contemporaneous record is whatever made it onto the legal pad — and the legal pad is mostly an outline, not a transcript.
Generic transcription tools
Off-the-shelf speech-to-text mangles technical vocabulary. "FinFET" becomes "fin fet." "Czochralski" becomes nothing. Specific numerical ranges get dropped or misheard. A transcript that says "between 12 and 80 nanometers" instead of "between 12 and 18 nanometers" is worse than no transcript — it is a confidently wrong record that an attorney might accidentally rely on while drafting.
Bot-based meeting recorders
Tools that join Zoom calls solve part of the problem for remote disclosures, but most invention disclosures happen in person, in a lab, or at the inventor's whiteboard. A bot in a calendar invite is not present where the work actually gets discussed. And in heavily regulated industries — defense, medical devices, certain semiconductor segments — IT policy often blocks third-party meeting bots from joining at all.
Paralegal write-ups
Even with a recording, transcription by a paralegal who is not technically trained loses the same detail the attorney did. Worse, the gap is now invisible: the attorney trusts the write-up because it looks complete, even though the technical specifics have been smoothed over into legalese.
What Actually Works
The right tool captures the disclosure verbatim, identifies who said what across multiple inventors, surfaces technical terms accurately, and lets the attorney search every prior disclosure for related claim language. Four capabilities matter most.
Domain-specific transcription accuracy
AmyNote runs transcription through OpenAI's latest Speech API, which handles patent vocabulary other tools garble: photolithography, eigenmode, methyltrioxorhenium, FinFET, Czochralski, monoclonal antibody, atomic layer deposition. Numerical ranges like "between 12 and 18 nanometers" survive intact. The model that powers ChatGPT's voice features is the same one transcribing the disclosure — and it has seen enough technical English to keep the specifics specific.
Speaker identification across inventors
Many disclosures involve two or three co-inventors plus the attorney. AmyNote tags each speaker and remembers them across future meetings, so when the same engineer is interviewed for a continuation six months later, attribution is automatic. This matters more than it sounds: in a disputed inventorship situation, knowing which inventor said which thing is the difference between an enforceable patent and an invalid one.
AI-powered claim extraction
Anthropic's Claude Opus drafts a structured summary after each disclosure: invention abstract, candidate independent claim language, listed embodiments, prior art mentioned by the inventor, and outstanding clarifying questions. The attorney edits rather than reconstructs. The first draft is no longer the attorney's blank page — it is a structured starting point that already mirrors the verbatim record.
Privacy architecture for sensitive IP
This is the dealbreaker for patent work. Both OpenAI and Anthropic contractually guarantee zero training on user data. Audio is encrypted in transit and not retained on provider servers after processing. Transcripts and recordings are stored locally on the attorney's device with end-to-end encryption. No unfiled invention sitting on a third-party server. No trade secret feeding a model training pipeline. For inventions that are not yet on file — and especially for trade secrets the client never intends to file — this is the difference between a usable tool and one that cannot pass conflict-of-interest review.
What Changes for the Prosecutor
The day-to-day shift is small but compounding. A disclosure that used to require a full afternoon of reconstruction becomes a thirty-minute review of a structured draft. The attorney spends the saved time on claim charts and prior-art analysis instead of typing. Continuations and divisionals draft faster because the original disclosure is searchable text, not a paper notebook in a banker's box.
The strategic shift is bigger. When every disclosure is captured verbatim and searchable, a firm's portfolio becomes a knowledge base. An attorney drafting a new application can search prior disclosures from the same client for related embodiments, distinguishing language, or prior-art arguments that already worked. The firm's institutional memory stops walking out the door when a senior partner retires.
Getting Started
Patent prosecution rewards precision. The cheapest hour in a patent's lifecycle is the disclosure meeting — and the most expensive one is the office action three years later when the examiner finds a prior-art reference the attorney never knew the inventor mentioned. AmyNote turns an inventor's words into a faithful technical record the attorney can mine for claim language, fight art with, and defend in litigation years later.
The trial is three days, no credit card required. Search AmyNote on the App Store.
Originally published as an X Article: The Inventor Disclosure Gap on X.
