Claims Locked to the Google API, Parentheses Everywhere: A Patent That Looks AI-Generated

Recently I read a set of patent claims, and the further I read, the more my brow furrowed. The amended claims not only locked the technical features onto a specific Google API, but the independent claim was also stuffed with a whole pile of parenthetical add-ons. Seeing writing like this, the first thought that came to mind was: this is probably a draft cranked out by AI.

Below I’ll roughly reconstruct what that independent claim looked like, so that even people who’ve never dealt with patent text can get a feel for its shape:

An information-processing method based on a conversational interface, executed by a server configured with at least one processor, comprising:

(a) receiving an unstructured instruction (natural-language text, speech-to-text, or a multimedia message) issued by a user through instant-messaging software (such as LINE, Messenger, Slack, or another platform with Webhook callback capability);

(b) transmitting it to a first AI model (such as a Transformer-architecture LLM) to perform intent analysis, outputting a structured instruction (such as JSON) that contains a task intent (search, summarize, integrate) and at least one first parameter (keyword, time range, file type);

(c) according to the task intent, having a routing-logic module autonomously determine and dynamically invoke at least one cloud-storage-service API (such as the Google Drive API) as an internal information source, and an Internet search-engine API (such as the Google Search API) as an external information source, respectively obtaining private-file summary information (vector-semantic summaries or key in-text fragments) and public search results (web-page titles, summary snippets, or audio/video scripts);

(d) cleaning and formatting both sets of data, integrating them into a structured comprehensive context (Context), and optimizing the ordering according to the model’s context-window limit;

(e) transmitting the comprehensive context and the structured instruction to a second AI model (which may be the same as or different from the first model) to perform synthesis, inference, and recombination, generating a final output of deeply integrated information (a structured report, an illustrated summary, or a response message), and returning it to the user through the message-push API of the instant-messaging software.

Two Things That Made Me Uneasy at a Glance

The first problem is that the claim drops wording binding it to a specific vendor’s products—“invoking the Google Drive API,” “invoking the Google Search API”—straight into the technical features that define the scope. Writing the claim scope this way amounts to shrinking the boundary of the right so that it covers only implementations using those particular APIs. The day someone swaps in an equivalent cloud-storage or search service—or even when Google itself changes the API’s name and behavior—whether this scope still has them covered becomes a giant question mark.

The second problem is that the independent claim is stuffed full of parentheses. Parentheses are meant for giving examples or adding detail, but when a single independent claim carries this many layers of add-ons piled on top of one another, it reads like someone dumped every possibility they could think of into it all at once, without ever really thinking through which ones are actually the technical features indispensable to defining the invention.

This Is Really a Sign of the Times

What I want to talk about isn’t singling out any one particular case, but the general phenomenon this style of writing lays bare.

Once generating patent drafts with AI becomes easy, “packing the words in and giving every example” takes almost no effort. And so we increasingly see claim scopes written at exhaustive length, superficially smooth in their logic, that on closer inspection lock the right onto specific products—or are hemmed in so tightly by an excess of parentheses that they can barely move.

To an outsider, a claim written long and complete looks like “thorough protection”; but the real problem often hides precisely in this “looking complete.” A patent scope isn’t worth more the more words it has; rather, every technical feature you write down becomes a rope tied around yourself when the time comes to assert your rights.

AI will help lay the bricks fast and neat, but it doesn’t necessarily know which wall shouldn’t be built at all. And this is exactly the part that most needs a human to stand guard.