Google has quietly launched a new AI-powered dictation app for iPhone users that turns spoken words into refined text entirely offline, underscoring the company’s growing focus on on-device, privacy-conscious AI experiences. The low-key rollout marks a significant move in the race to redefine how people write, work and communicate using voice instead of keyboards.
Google’s new app: AI Edge Eloquent
The app, called Google AI Edge Eloquent, appeared on the iOS App Store without a formal announcement, blog post or launch event, signaling a deliberate “soft launch” rather than a headline-grabbing debut. It is free to download, carries no subscription fees and does not advertise strict usage caps, making it immediately attractive to students, professionals, journalists and creators who dictate frequently.
After installation, users are prompted to download AI models to their device, which power the app’s offline speech recognition capability. Only once these models are stored locally does full dictation unlock, allowing the system to listen, transcribe and clean up speech even when the phone is in airplane mode or away from a network.
In its App Store description, Google pitches the app as “an advanced dictation app engineered to bridge the gap between natural speech and professional, ready-to-use text.” It stresses that Eloquent focuses on what the user meant to say rather than echoing every stumble or false start, positioning it as a writing assistant rather than a bare-bones transcription tool.
How it works: from raw speech to clean copy
When a user starts dictating, Eloquent displays a live transcription line by line, with words appearing on screen in real time as they are spoken. Once the user stops or taps to pause, the app performs a second pass, automatically cleaning up the text in the background.
According to the app’s description, “Unlike standard dictation software that transcribes stumbles and filler words verbatim, Eloquent utilizes AI to capture your intended meaning. It automatically edits out ‘ums,’ ‘uhs,’ and mid-sentence self-corrections, outputting clean, accurate prose.” Everyday speech, often filled with pauses, restarts and half-finished phrases, is turned into more polished, structured writing.
Beyond simple cleanup, the app also gives users some control over the tone and length of the final text. Early descriptions highlight options to shorten content or make it more formal, so a loosely spoken brain dump can be transformed into something closer to a professional email or summary. For people who think out loud, real-time transcription plus automated editing could significantly reduce the time spent rewriting dictated drafts.
Offline-first, with optional cloud polishing
One of the app’s standout characteristics is its “offline-first” design. The core speech recognition runs directly on the device using compact AI models, enabling fast response times and improving privacy by keeping audio data off remote servers by default. This is especially useful for users who travel, work in the field or simply distrust always-online AI services.
At the same time, Eloquent offers an optional cloud-enhanced mode that can be switched on by the user. In that configuration, the initial transcription still happens locally, but the text can be sent to larger cloud models for extra refinement, stylistic tweaks or more advanced processing. Users who care most about privacy can stay strictly offline, while those who want maximum polish can opt in to cloud assistance.
This hybrid approach reflects a broader shift toward combining edge AI and cloud AI instead of treating them as separate worlds. Lightweight models on the phone handle the bulk of everyday tasks, while heavier cloud models come into play only when needed and only with explicit user consent.
Privacy, custom vocabulary and integration
Eloquent also addresses a long-standing pain point in dictation apps: proper nouns, unusual names and domain-specific vocabulary. The app can build a personalized vocabulary over time, improving its ability to recognize terms the user frequently speaks.
To jump-start that process, it can draw on a user’s recent communication patterns to surface commonly used names and phrases. This helps the app better recognize contacts, brands or project terms that would otherwise be misheard. At the same time, that kind of integration naturally raises questions about what data is scanned, how it is retained and how clearly users can control or disable it.
The app stores previous dictations and offers search across them, turning it into a repository of past notes, ideas and drafts. It also surfaces statistics such as words per minute and total words dictated, a useful touch for power users who rely heavily on voice for long-form writing.
With its offline-first architecture and optional cloud layer, Google is clearly framing Eloquent around privacy and control. Local processing by default, combined with opt-in cloud enhancement, is designed to reassure users who are wary of sending raw voice recordings to remote servers whenever they dictate.
iOS first, Android later
Perhaps surprisingly for a Google product, AI Edge Eloquent has launched first on Apple’s iOS rather than on Android. The app is currently listed on the App Store for iPhone users, while references in descriptions and external reports indicate that an Android version is planned but not yet available.
Documentation suggests that when it does arrive on Android, the app will integrate more deeply into the system. It is expected to be usable as a default dictation layer and may offer a floating button to trigger voice input from any text field, making it feel less like a standalone app and more like part of the operating system.
The decision to test the waters on iOS first could be strategic. By introducing the tool in Apple’s ecosystem, Google can observe how users adopt an offline, AI-powered dictation assistant without immediately overlapping with existing Android input features. It also serves as a showcase of on-device AI capabilities on a competing platform.
For now, Android users will have to wait for a formal rollout date, but early signals suggest Google sees this app as more than just another utility. If the experience proves successful on iOS, similar capabilities could be baked deeper into Android’s native keyboard and system-level voice input.
Taking on Wispr, SuperWhisper and others
Eloquent enters a busy and fast-evolving field of AI dictation tools. Recent months have seen a wave of apps focused on fast, accurate, AI-enhanced transcription, including Wispr Flow, SuperWhisper and Willow. Many of these products lean on large language models to polish text, summarize recordings and plug into notes, documents and task managers.
Google’s strategy is to bundle several key demands: offline operation, automatic removal of filler words, style adjustment and a free price point into a single package. By relying on efficient on-device models tuned for consumer hardware, the company can deliver an experience that feels responsive without connectivity, while still reserving more advanced cloud capabilities for those who want them.
Industry watchers view this quiet launch as part of a shift away from promoting only chatbots and toward embedding AI into everyday workflows such as typing, emailing and note-taking. If Eloquent gains momentum, it could pressure competitors that depend heavily on cloud processing and subscriptions to expand their offline capabilities and rethink pricing.
“Clean, accurate prose” on the go
The core promise of Eloquent is captured in its own description: the app “utilizes AI to capture your intended meaning outputting clean, accurate prose.” For people constantly on the move, that promise speaks to a real need getting from spoken thoughts to usable text with minimal friction.
Rather than treating dictation as a rough first draft that always needs heavy editing, Eloquent aims to deliver something much closer to a final version. A user can talk through an idea on a walk or between meetings and end up with text that is already structured and relatively polished, ready to paste into an email, report or document.
Built on years of work shrinking speech recognition models to run directly on phones, the app turns that underlying research into a focused productivity tool. Its quiet debut suggests Google is willing to let users discover it organically as an offline-first AI assistant that listens carefully, edits intelligently and, if it succeeds, may change how many people prefer to “write” in the first place.
Comments