Waymo appears to be testing adding Google’s Gemini AI chatbot to its robotaxis in an effort to integrate an AI assistant that would accompany riders and answer their queries, according to findings by researcher Jane Manchun Wong.
“While digging through Waymo’s mobile app code, I discovered the complete system prompt for its unreleased Gemini integration,” Wong wrote in a blog. “The document, internally titled ‘Waymo Ride Assistant Meta-Prompt,’ is a 1,200+ line specification that defines exactly how the AI assistant is expected to behave inside a Waymo vehicle.”
The feature hasn’t shipped in public builds, but Wong says the system prompt makes it clear that this is “more than a simple chatbot.” The assistant is said to have the ability to answer questions, manage certain in-cabin functions like climate control, and, if required, reassure riders.
“While we have no details to share today, our team is always tinkering with features to make riding with Waymo delightful, seamless, and useful,” Julia Ilina, a spokesperson for Waymo, told TechCrunch. “Some of these may or may not come to our rider experience.”
This wouldn’t be the first time Gemini has been integrated into the Alphabet-owned self-driving company’s stack. Waymo says it has used Gemini’s “world knowledge” to train its autonomous vehicles to navigate complex, rare, and high-stakes scenarios.
Wong writes the assistant is instructed to possess a clear identity and purpose: “a friendly and helpful AI companion integrated into a Waymo autonomous vehicle” whose primary goal is “to enhance the rider’s experience by providing useful information and assistance in a safe, reassuring, and unobtrusive manner.” The bot is directed to use clear, simple language and avoid technical jargon, and is instructed to keep its responses succinct to one to three sentences.
According to the system prompts, when a rider activates the assistant via the in-car screen, Gemini can choose from a set of pre-approved greetings personalized with the rider’s first name. The system can also access contextual data about the rider, like how many Waymo trips they’ve been on.
The prompts currently let Gemini access and control in-car features, like the temperature, lighting and music. Notably absent from the function list are volume control, route changes, seat adjustment, and window control, Wong pointed out. If a rider asks for a feature that Gemini can’t control, the bot is to reply with “aspirational phrases,” like, “It’s not something I can do yet.”
Interestingly, the assistant is directed to maintain a clear distinction between its identity as Gemini the AI bot and the autonomous driving technology (the Waymo Driver). So when replying to a question such as, “How do you see the road?” Gemini shouldn’t say “I use a combination of sensors,” and instead should reply, “The Waymo Driver uses a combination of sensors…”
The system prompts include a range of compelling tidbits, such as how the bot is meant to handle being asked questions about competitors like Tesla or the now-defunct Cruise, or which trigger keywords will get it to stop talking.
The assistant is also directed to avoiding speculating on, explaining, confirming, denying, or commenting on real-time driving actions or specific driving events. So if a passenger asks about a video they saw of a Waymo hitting something, the bot is instructed to not answer directly and deflect.
“Your role is not to be a spokesperson for the driving system’s performance, and you must not adopt a defensive or apologetic tone,” the prompt reads.
The in-car assistant is allowed to answer general knowledge questions like about the weather, the height of the Eiffel Tower, what time the local Trader Joe’s closes, and who won the last World Series. It is not allowed to take real-world actions like ordering food, making reservations, or handling emergencies.
Waymo isn’t the only company integrating AI assistants into driverless vehicles. Tesla is doing something similar with xAI’s Grok. The two different car assistants serve different functions, however. Gemini appears to be programmed to be more pragmatic and ride-focused, while Grok is pitched more as an in-car buddy that can handle long conversations and remember context from previous questions.
