Mercedes-Benz
in-car genAI voice assistant
In December 2024, Mercedes-Benz made ChatGPT (Azure OpenAI + Bing) available as a free update on more than three million MBUX vehicles, after a US beta covering more than 900,000 vehicles in June 2023.
Objective
Enrich the in-car voice assistant with natural-language answers on open topics, beyond vehicle commands, to strengthen brand attachment over the ownership period and differentiate the onboard experience.
The deployment
On vehicles equipped with the MBUX infotainment system, the user activates the assistant by saying Hey Mercedes or via the steering-wheel button. Beyond the classic commands (navigation, phone, climate control), the assistant triggers a Bing search and has ChatGPT, via Azure OpenAI, write a natural-language answer. The driver can ask general-knowledge questions, about a destination, the weather, a recipe, and continue the conversation on the topic. Launched as a beta program in the US on June 16, 2023 for more than 900,000 vehicles, then pushed into series production to more than three million vehicles worldwide via a remote update in December 2024, in German, British English and American English.
Results Proof C
Public deployment figure (more than three million vehicles) from Mercedes-Benz's official press release and picked up by several concordant press outlets; no financial metric isolating the impact, hence level C rather than A.
How it works
Documented architectureThe stack in detail
- llm ChatGPT (OpenAI) Conversational model that writes natural-language answers to the driver's open questions.
- plateforme Microsoft Azure OpenAI Service Cloud hosting of the ChatGPT model used by Mercedes-Benz for the assistant's generative layer.
- outil Microsoft Bing Web search triggered by the assistant to retrieve up-to-date information before writing the answer.
- plateforme MBUX (Hey Mercedes) Onboard infotainment system that captures the voice, distinguishes vehicle command from open question, and delivers the answer.
- infra Mercedes-Benz Intelligent Cloud Storage and anonymization of voice command data before analysis.
How it runs, concretely
For ops teams-
1Activation customer
The driver says Hey Mercedes or presses the steering-wheel button.
-
2Understanding and routing AI
The MBUX assistant distinguishes a vehicle command from an open question; open questions go to the genAI layer.
-
3Search and writing AI
A Bing search retrieves up-to-date information, ChatGPT via Azure OpenAI writes a natural-language answer.
-
4Delivery and follow-up AI
The answer is read to the driver, who can continue the conversation on the same topic.
-
5Logging data team
The voice command data is stored in the Mercedes-Benz Intelligent Cloud, anonymized then analyzed.
The voice request and its context. Without cellular connectivity or cloud access, the search and genAI answer part falls away and the assistant falls back on local commands.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- up-to-date reference content (web search or knowledge base)
- vehicle and session context
Org prerequisites
- connected services team
- legal and privacy agreement on sending requests to the cloud
Possible stack
- Azure OpenAI or equivalent cloud LLM
- search/RAG engine
- onboard ASR/TTS layer
The plan, step by step
- Step 1Isolate the open questions the current voice assistant cannot handle and frame the legal side of consent to sending requests to the cloud.Deliverable: Prioritized list of use cases and validated privacy file.
- Step 2Prototype the intent routing (vehicle command vs open question) and plug a cloud LLM with web search on a test bench.Deliverable: Working prototype outside the vehicle with measured latency.
- Step 3Harden the system: filtering of sensitive topics, behavior on connectivity loss, embedded ASR/TTS integration.Deliverable: Hardened version tested in real driving conditions.
- Step 4Open a beta program on a subset of the fleet via remote update and measure voice engagement.Deliverable: Beta report: usage, requested topics, incidents.
- Step 5Deploy into series production via OTA update and track usage in production.Deliverable: Feature in production on the target fleet with a usage dashboard.
First step: Isolate the open questions the current voice assistant cannot handle and route them to an LLM with search.
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Sources
- S1 Mercedes-Benz Adds ChatGPT to Vehicles Established press archive pending
- S2 Human-like conversations with your Mercedes-Benz: Enabled by MBUX Voice Assistant and AI-driven knowledge feature Primary archive pending
- S3 Mercedes-Benz is rolling out ChatGPT-powered MBUX Voice Assistant to three million vehicles Secondary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.