Google reported that active users of its Gemini app in Southeast Asia have more than doubled over the past year, with adoption in the region outpacing any previous Google app launch, according to the company’s inaugural “Gemini Southeast Asia Report 2026.”
The report examines six of Gemini’s largest regional user bases: Indonesia, Malaysia, the Philippines, Singapore, Thailand, and Vietnam. Google did not disclose absolute user numbers, making it difficult to gauge the scale of adoption relative to rivals such as OpenAI’s ChatGPT, Microsoft Copilot, Anthropic’s Claude, Meta AI, and China‑linked AI assistants increasingly embedded in consumer platforms.
Nevertheless, the usage patterns released provide a valuable glimpse into how generative AI is moving beyond early adopters in Southeast Asia.
The central finding is that Gemini’s growth in the region is driven less by English‑language productivity use cases and more by local‑language, mobile‑first behavior. Google said nearly 70 percent of Gemini prompts in Southeast Asia are submitted in native languages, led by Vietnam at 89 percent, Thailand at 87 percent, and Indonesia at 84 percent.
This is crucial in a region where language fragmentation has long shaped the adoption of digital products. Southeast Asia has more than 600 million people, hundreds of languages and dialects, and a young internet population that often leapfrogs desktop‑first behavior. Google noted that nearly 40 percent of the region’s population is under 25, a demographic that is helping drive heavier usage, longer conversations, and more detailed prompts on Gemini compared with older groups.
Local‑language AI becomes the adoption test
The report underscores a broader point: generative AI adoption in Southeast Asia will depend heavily on whether models can handle local languages, code‑switching, and cultural context, rather than merely translating English‑language outputs.
Google cited the Southeast Asia Holistic Evaluation of Language Models, or SEA‑HELM, as evidence of Gemini’s performance, saying the model ranked as the best‑performing large language model for Southeast Asian languages. SEA‑HELM was developed to evaluate LLMs across regional linguistic contexts, including languages that are often underrepresented in global AI benchmarks.
Dr Leslie Teo, Senior Director of AI Products at AI Singapore, said the region’s AI adoption depends on whether technology feels “native, not translated.”
“In a region as linguistically rich as Southeast Asia, models must understand local context deeply to be used effectively, whether that’s for learning, nuance‑heavy writing, or complex business brainstorming,” Teo said. “Gemini’s strong standing on our SEA‑HELM evaluation framework demonstrates its ability to navigate the complex linguistic realities of the region.”
The emphasis on language is not merely academic. For startups and enterprises across Indonesia, Vietnam, and Thailand, AI tools that perform poorly in local languages are less useful for customer service, marketing, education, document processing, and internal knowledge work. That leaves room for both global platforms and regional initiatives, including AI Singapore’s SEA‑LION model, national AI programmes, and localised enterprise AI vendors.
Mobile‑first and multimodal by default
Google’s data also suggests that Southeast Asian users are not treating AI assistants primarily as desktop chat boxes. The company said 75 percent of Gemini requests in the region come from mobile devices. More than 40 percent of prompts involve voice commands, photos, or video uploads, while voice‑only prompts account for 10 percent of usage.
That behavior fits the region’s wider internet economy. Southeast Asia’s digital economy reached US$263 billion in gross merchandise value in 2024, according to the Google, Temasek, and Bain e‑Conomy SEA report, with mobile commerce, digital payments, short‑form video, and super‑app behavior shaping how consumers interact with online services.
For AI companies, the implication is clear: assistants that require typed, structured English prompts may struggle to become mainstream in markets where users are more comfortable speaking, uploading images, or mixing languages in a single query.
This also raises the competitive bar. OpenAI has been pushing multimodal features through ChatGPT, Meta has embedded AI into WhatsApp, Instagram, and Facebook, and Microsoft has integrated Copilot into Windows and Office. Google’s advantage lies in distribution through Android, Search, Workspace, and Gmail, but it must show that Gemini can become a daily utility rather than a feature users test once and abandon.
Creativity, research, and daily assistance
The report shows Gemini usage spreading across creative, academic, and practical tasks. Google said around 40 percent of queries ask the app to generate new outputs, including images, music, videos, and documents.
Across Southeast Asia, users have generated five billion images with Nano Banana, Google’s image‑generation model, over the past year. They have also created almost one million songs since Google introduced Lyria 3, its music‑generation model, in the region.
Beyond content generation, users are relying on Gemini to summarise dense documents, structure messy data, seek recommendations, plan travel, troubleshoot problems, and generate ideas. These are familiar use cases across generative AI, but the regional difference lies in how users access them: through mobile, voice, images, and native‑language prompts.
For Southeast Asian startups, this creates both opportunity and pressure. AI‑native consumer applications will need to compete with increasingly capable platform‑level assistants. Enterprise software companies, meanwhile, must decide whether to build their own AI layers, partner with model providers, or rely on integrations from Google, Microsoft, and OpenAI.
Gemini Spark pushes Google into agentic AI
Google is now extending Gemini further into agentic AI with Gemini Spark, a feature designed to complete tasks on behalf of users rather than simply answer questions. The company said Spark is integrated with Workspace tools such as Gmail, Docs, and Slides, and can work in the background even when a user’s laptop is closed or phone is locked.
After an initial English‑language launch, Google is rolling out Gemini Spark in local languages to Gemini Advanced Ultra subscribers in Southeast Asia this week. The paid‑subscriber framing is important: it suggests Google is testing whether regional users will pay for more proactive AI assistance, not just free chatbot access.
For now, the bigger question is trust. Agentic AI that interacts with email, documents, and workplace tools must handle privacy, permissions, accuracy, and security more carefully than a chatbot producing text on demand. In Southeast Asia, where data‑protection regimes vary widely—from Singapore’s more mature Personal Data Protection Act framework to newer or still‑evolving rules in markets such as Indonesia and Vietnam—deployment will not be uniform.
Google’s Gemini report makes a strong case that Southeast Asia is becoming a serious testbed for consumer AI adoption. But it also leaves important questions unanswered: how many active users Gemini has, how often they return, how usage compares with ChatGPT or Meta AI, and whether creative experimentation converts into durable paid subscriptions.
What is clear is that the next phase of AI competition in Southeast Asia will not be won only on model benchmarks. It will be fought on language, mobile behaviour, distribution, trust, and whether assistants can do useful work in the messy, multilingual reality of everyday life.
