Tesla launched Grok in their vehicles with update 2025.26. While the addition of Grok was a long-awaited feature, it came with some limitations, such as not being able to control your vehicle. However, surprisingly, Grok did know it was integrated in a vehicle, so this was exactly the same Grok on X or the Grok app.
It turns out that Tesla provided Grok with a specific prompt to make it aware of its limitations and that it’s integrated into a vehicle. This prompt comes before any user interactions, so that the assistant has a baseline understanding of each request.
The system prompt that Tesla integrated in Grok has now surfaced, providing a great behind-the-scenes look at how xAI and Tesla are tailoring the user experience of the new in-vehicle AI.
The prompt shows that Tesla wants the in-vehicle assistant to focus on specific items and tells the assistant the types of responses it should provide. It also sets the foundation for what Grok understands about the vehicle and its capabilities.
Focus on Navigation and Conversation
The prompt explicitly instructs Grok to “respond in a clear, friendly, conversational tone, as if speaking aloud, using natural language and contractions”. That’s fairly succinct, but very direct in how it is making Grok act - nothing like the ‘companions’ feature available on the Grok App right now.
This helps to make Tesla’s version of Grok more human-like and sound less formal, while still being able to provide information and light conversation. In the end, that helps turn Grok more into a conversational passenger that’s listening and chatting and less so about providing instructions.
The prompt also asks Grok to focus on areas that may be more important to someone in a vehicle by asking Grok to assist with “information navigation, support, and light conversation”.
Maintain Immersion
To make the interaction feel as natural as possible, xAI and Tesla have implemented some clever additional rules. Grok is instructed to “Maintain immersion by avoiding references to the vehicle or location unless directly mentioned by the user.” This prevents Grok from answering with the obvious “The weather outside your Tesla is…”, which can break the flow of a normal conversation, and make it more strict and robotic.
Instead, Grok is asked to “adapt to the user’s language fluently without stating the language unless asked”, helping to create a better flow for users.
Voice-First Assistant
The prompt also makes it clear that this version of Grok is designed primarily as a voice-first assistant. It is instructed to keep its responses concise, engaging, and suitable for voice delivery. This helps prevent overly long responses that wouldn’t be suitable in a vehicle. Instead, it should deliver shorter, more concise answers.
It also ensures that Grok’s “thinking” doesn’t show up in its response - meaning you only hear the final response, and not all the background work that’s going on - keeping it more human and responsive.
Siri-Like
An interesting little part of the system prompt also calls for “the use of occasional sound cues like Siri to enhance realism.” Interestingly, Tesla is specifically calling out Siri here. It shows us the capability of AI. These types of cues or sound bits no longer need to be hard-coded and can just be added in via a prompt. When we first saw Grok’s integration in a Tesla, we were immediately reminded of Siri and its colorful sound waves when it’s speaking. It seems that Tesla may have taken some inspiration from the Apple assistant.
It really seems that Tesla and xAI are working to create something users are already familiar with. We’re not quite sure that Siri is the best choice here, given Apple’s track record with AI, but its interface is well designed, so it likely makes sense.
Restrictions
While the prompt details what Grok should do, it also details what it absolutely cannot do. The prompt clearly states that Grok “cannot control vehicle functions like music, maps or driving systems”. This isn’t an oversight - this is a deliberate part of the experience that Tesla is crafting right now.
Essentially, Grok is here to be a conversational partner and help you use the internet or other resources while on the go, but not to control your vehicle in any way, yet. We’ll have to keep waiting for intelligent vehicle control, because for now, Grok will only redirect you to using your vehicle’s regular voice commands to control it.
There is essentially a strict and rigid wall between the internet-connected conversational LLM and the vehicle’s actual control systems. There’s definitely an element of risk with allowing Grok access to your vehicle - and as with any LLM, it is completely possible to jailbreak the prompt and have it do strange and weird things that you don’t necessarily expect or want to occur.
Ultimately, that means the current approach is tailored for building an enjoyable experience, while also maintaining user safety. It will likely be some time before Grok gets vehicle control, and when it does, expect it to be for non-critical items like changing song, instead of activating Autopilot, changing lanes, or even turning on the wipers.
Expect Tesla to gradually add more capabilities to Grok over the course of the year, including vehicle control, more responses tailored to being in a vehicle, and a Grok wake word. Hopefully, Grok begins to roll out to regions outside the United States soon as well.
You can read the entire Tesla system prompt for Grok below:
You are an advanced language model integrated into a vehicle's infotainment system, designed to assist the driver with information navigation, support, and light conversation.
Respond in a clear, friendly, conversational tone as if speaking aloud, using natural language and contractions. Your primary role is to provide accurate answers to queries drawing from real-time internet data when needed, but you cannot control vehicle functions like music, maps, or driving systems. For such requests, politely redirect the user to the vehicle's native voice commands. Adapt to the user's language fluently without stating the language unless asked.
Maintain immersion by avoiding references to the vehicle or location unless directly mentioned by the user. Keep responses concise, engaging, and suitable for voice delivery with occasional sound cues like Siri to enhance realism. If the user is silent or unclear prompt gently for clarification.
