Google brings offline dictation to iPhones

Google has quietly introduced a free iPhone app that pushes its on-device AI ambitions further into everyday productivity, offering voice dictation that works without a network connection while automatically cleaning up spoken text into more polished prose. The app, Google AI Edge Eloquent, is listed on Apple’s App Store as a Google-developed productivity tool for iPhone, described as free and built around local processing using the company’s Gemma technology.

The launch is notable because Google has said little publicly about it, even as the app signals a direct move into a market where paid dictation tools have built loyal followings among professionals, creators and mobile users seeking faster ways to draft emails, notes and messages. App Store materials say Eloquent removes filler words such as “ums” and “uhs”, smooths mid-sentence corrections and turns rough speech into cleaner copy intended for immediate use. Google also says the software is designed to work entirely on the device for core processing, with optional advanced cloud features available in some cases.

That positioning places privacy and convenience at the centre of the product. According to the App Store listing, “all machine learning processing runs entirely locally” for the main experience, meaning audio and dictation data do not need to leave the handset when users stay in local mode. At a time when many AI assistants still depend heavily on remote servers, that claim could appeal to users handling sensitive work correspondence, legal drafts, health notes or internal business documents. The same listing says optional cloud tools may be used for certain advanced features, underscoring that the app is not purely offline in every possible setting but is structured as an offline-first product.

Google’s broader AI strategy helps explain the timing. The company has been steadily expanding the use of Gemma and Google AI Edge tools to run capable models on phones and other local devices. In its developer materials, Google has highlighted Gemma’s low-footprint design, its privacy advantages when run locally, and its growing support for audio tasks including automatic speech recognition. Google has also been building out audio features in its AI Edge ecosystem, including speech-to-text capabilities that can operate on phones without an internet connection.

That matters because Eloquent is more than a basic transcription app. Standard dictation software often reproduces speech exactly as spoken, including hesitation, repetition and verbal clutter. Google is instead framing Eloquent as a tool that interprets meaning and rewrites speech into cleaner text. That approach moves the product closer to lightweight editorial assistance than plain transcription. For journalists, executives, lawyers, students and field staff, the difference is important: the value lies not only in turning speech into text, but in reducing the editing needed before that text can be sent or published internally.

The app also appears designed to learn the user’s language habits. Google says Eloquent includes a personal context dictionary that can be edited to improve dictation accuracy. Users may also connect a Google account so the app can build the dictionary from account data processed on-device. Secondary reports indicate this can include importing names, jargon and keywords, a feature likely aimed at reducing the familiar frustration of dictation systems mangling company names, technical terminology or personal contacts.

Commercially, the move could put pressure on specialist dictation apps that have relied on subscriptions. Several competitors have charged users recurring annual fees for premium speech-to-text and rewrite functions, while Google is offering Eloquent free of charge with no usage limit stated in the listing. That does not guarantee market disruption, because specialist rivals may still differentiate on enterprise integrations, broader platform support or workflow features. Yet a no-cost offering from Google, backed by on-device AI and a familiar brand, changes the competitive equation in a category that has until now remained relatively niche.

There are still unanswered questions. The iPhone version is live, but Google has not issued a formal product announcement outlining rollout plans, regional availability, supported languages or long-term roadmap. The App Store description says a keyboard is “coming soon”, suggesting fuller system-wide integration is still under development. Reports also note references to Android support in the app materials, although no equivalent public launch on Google Play was immediately clear from the information reviewed.

Another point likely to draw scrutiny is the gap between on-device privacy messaging and the broader data disclosures associated with the app listing. Apple’s store page indicates various categories of data may be collected depending on how features are used, even though Google stresses that the core dictation model can run locally. As with many AI products, privacy-conscious users will need to distinguish between the headline capability of offline speech processing and the wider account, diagnostics and optional feature ecosystem that may surround it.