When an emergency happens beyond cellular or Wi-Fi coverage, getting the right information to emergency services can be difficult. Satellite-based emergency services can connect devices when conventional networks are unavailable, but lower bandwidth and higher latency can make communication challenging.
Existing satellite emergency systems can address this challenge by using structured questionnaires to gather critical information from users. However, requiring users to complete multiple steps before transmitting an emergency message can delay communication and increase the data payload.
Google’s new patent application describes a different approach that uses a large language model (LLM) to process users’ free-form descriptions of an emergency, extract the most relevant information, and automatically generate a structured message for emergency services.
Users can provide information about an emergency through the interface in their own words, either by typing, speaking, or otherwise communicating in natural language. The LLM then analyzes the input to identify key details about the situation, such as the user’s location, injury status, psychological state, lost equipment, number of people involved, and resources needed. It can also use device information, such as GPS data and network connection status, to supplement the information provided by the user. The LLM uses these extracted details to generate a concise, structured emergency message that can be transmitted to emergency services.
A key feature of the invention is that the LLM can be deployed at least partially on the user’s device rather than relying entirely on cloud processing. This allows the system to process emergency information locally, which could be useful when a user is outside terrestrial network coverage. The filing describes a lightweight LLM that can operate on devices including smartphones, smartwatches, earbuds, smart glasses, vehicles, and other connected hardware.
The system can also adapt its questions when important information is missing. Rather than requiring users to complete a fixed questionnaire, the LLM can generate follow-up questions based on the initial emergency description. In the application’s example, a user reporting that a boat is sinking is asked how many people are onboard, with the response added to the emergency message before transmission.
Why it matters
The patent application aligns with Google’s wider effort to extend Gemini beyond smartphones and into a growing ecosystem of connected devices. Google has been expanding Gemini to watches, cars, earbuds, headsets, and glasses, while Android XR provides a platform for bringing Gemini-powered experiences to the next generation of headsets and glasses.
Additionally, the proposed system could build on Google’s existing Satellite SOS capability on Pixel phones, which allows users to contact emergency services when mobile and Wi-Fi networks are unavailable. The LLM-based approach could complement this capability by allowing users to describe an emergency in their own words, identify missing information through follow-up questions, and compress the resulting message for transmission over bandwidth-constrained networks
U.S. Pat. App. Pub. No. 2026/0259905, titled “Large Language Model Augmented User Emergency Interface,” was filed on April 23, 2026, and was published on September 3, 2026. The listed inventors are Aishwarya Mallampati, and Sooraj Sasindran.