GM launches an in-car AI assistant this year, because Gemini can’t access car data

General Motors will launch its own in-vehicle AI assistant later this year, and the key reason is blunt: Google’s Gemini cannot access what the car itself knows. GM is building the assistant to tap directly into vehicle data via OnStar, GM telemetry systems, and GM’s proprietary vehicle knowledge, so it can do things a general-purpose assistant cannot.

That matters because the output is not “drive safely” style generic help. GM’s planned assistant is meant to support predictive maintenance, real-time diagnostics, and auto-specific commands that depend on knowing what is happening inside the vehicle right now. In other words, it is not competing with a chatbot. It is trying to become the car’s operational brain, with the permissions to actually read the sensors and systems that determine what maintenance is needed and what is malfunctioning.

This is a classic AI product trap that the industry is still stumbling into: model capability is not the same thing as system access. A general-purpose assistant can sound smart while lacking the authorization, the integration points, and the context to act on real-time telemetry. For a vehicle, that gap is expensive. Predictive maintenance is only useful if it can reliably infer failure risk from the right signals. Real-time diagnostics require low-latency visibility into vehicle state. Auto-specific commands require tight control paths so the request maps to actions that the vehicle can actually execute.

GM’s approach, as described here, is to move the intelligence into the stack where the data is. By integrating with OnStar and GM telemetry systems, the assistant is positioned to translate raw vehicle and network signals into vehicle-relevant explanations and recommended next steps. By relying on proprietary vehicle knowledge, it can also tailor responses to the specific architecture, behaviors, and constraints of GM vehicles. That is exactly the sort of specificity that a general-purpose model will struggle to provide when it cannot directly see the underlying facts.

Zoom out and this fits the broader automotive shift toward connected services. The car is increasingly a data-producing device, not just a machine for moving people from A to B. OnStar and telemetry are the highways for that data, and AI assistants are the on-ramps that turn the data into user value. The more the industry leans on these services, the more the “assistant” becomes a gateway to revenue streams like diagnostics subscriptions, maintenance planning, and fleet-like insights even for personal vehicles.

There is also a regulatory and safety framing lurking underneath the product description. Vehicle diagnostics and maintenance guidance is not the same as telling you what the weather is. It has implications for safety, liability, and consumer trust. When an assistant provides real-time diagnostics, it effectively becomes part of the vehicle’s decision support loop. That means the implementation details matter: which systems the assistant can read, how it interprets faults, how it communicates risk, and how it ensures that recommendations are grounded in vehicle-specific data. Building this assistant around GM-controlled telemetry and knowledge is one way to keep the system bounded to authoritative inputs.

Now consider the board-level perspective. If GM can deliver predictive maintenance and diagnostics through an always-available assistant, it strengthens customer engagement and can improve service economics. It also creates a platform advantage that competitors may have to match: integration depth. A rival can add a “chat with your car” experience quickly, but matching the ability to access telemetry and convert it into diagnostic and maintenance actions is harder. It requires tighter partnerships, deeper system integration, and the willingness to treat the assistant as an embedded product, not a thin UI on top of a model.

For executives at automakers and suppliers, the second-order implication is straightforward: the next wave of vehicle AI will be won by data access and workflow integration, not by which model you picked for your demo. An in-car assistant that cannot reach the vehicle’s internal state is entertainment. An in-car assistant that can is operations. GM is betting that this year’s assistant launch will show the difference, and the stakes are customer trust, service revenue, and the long-term control of the car-to-cloud-to-service loop.