Headline & intro
Typing on glass has always been a compromise. With its quietly released Google AI Edge Eloquent app for iOS, Google is signalling that compromise might finally be negotiable. This isn’t just another transcription utility; it’s Google’s first serious attempt to turn on‑device AI into a mainstream way of writing.
In this piece, we’ll look at why an offline‑first dictation app matters far beyond note‑taking, what it reveals about Google’s edge‑AI strategy, how it could reshape competition with Apple and a growing wave of startups – and why European users should pay close attention to the privacy and regulatory angles.
The news in brief
According to TechCrunch, Google has quietly launched an experimental dictation app for iOS called Google AI Edge Eloquent. The app is free and built around Gemma‑based automatic speech recognition models that run locally on the device. Once downloaded, users can dictate and see live transcription, which is then automatically cleaned up when they pause.
The app removes filler words, smooths self‑corrections and can transform text into different formats such as key points or more formal wording. Users can optionally enable cloud processing that calls Gemini models for additional text refinement. Eloquent can import custom vocabulary from a user’s Gmail account and lets them add their own terms. There is a searchable history of sessions plus basic stats such as words per minute.
For now, Eloquent is only on iOS, but the App Store description mentions an upcoming Android version that can act as a system‑wide keyboard and offer a floating transcription button, similar to rival Wispr Flow.
Why this matters
The obvious story is “Google ships a dictation app.” The more interesting one is this: Google is testing what happens when text input becomes an AI product, not a keyboard feature.
First, this is a clear on‑device AI showcase. Running reasonably advanced speech recognition locally – with optional cloud enhancement – is exactly the kind of hybrid model the industry has been promising. Offline mode means lower latency, better reliability on the move and a powerful privacy narrative: your raw voice data doesn’t have to leave the phone.
Second, this puts pressure on a fast‑growing niche of AI transcription startups. Tools like Wispr Flow, SuperWhisper and Willow have been charging or experimenting with subscriptions for very similar capabilities. Google is now offering a highly polished version for free, on Apple’s platform no less. That could compress pricing power across the segment and force smaller players to specialise – for example in team workflows, meeting intelligence or industry‑specific jargon – rather than core recognition quality.
Third, this is a subtle jab at Apple. Apple has invested heavily in on‑device models, but its built‑in dictation still feels like a feature, not a destination app. By shipping Eloquent on iOS first, and promising deep Android integration later, Google is effectively saying: “If Apple won’t make voice the primary input, we will – even on their phones.”
The immediate winners are mobile professionals, students, journalists and anyone who writes a lot on the go. The losers may be indie dictation apps and, if Google plays this right, traditional keyboards.
The bigger picture
Eloquent sits at the intersection of three larger trends.
1. The on‑device AI land grab.
Google has been talking up “AI on the edge” for years – from Tensor chips in Pixels to Gemini Nano on Android. Apple is widely expected to make on‑device AI a centrepiece of upcoming iOS releases. Qualcomm, MediaTek and Samsung are all shipping NPUs and bragging about TOPS figures. But users don’t care about TOPS; they care about experiences. Eloquent is one of the first consumer‑visible, always‑useful examples of what those chips are for.
2. The renaissance of voice as an interface.
We’ve been here before: Dragon NaturallySpeaking in the 2000s, then Google Voice Typing and Siri. Accuracy was good enough for commands, not for serious writing. With modern models, error rates have dropped and language understanding has improved. Startup products like Wispr Flow or SuperWhisper already demonstrated that you can dictate long‑form text comfortably. Google entering the ring validates that voice‑first writing is no longer a niche for accessibility or power users.
3. Input is becoming an AI service layer.
Look at Eloquent’s features: not just transcription, but automatic editing, summarisation, tone shifting, custom vocabulary and statistics. This looks less like a microphone and more like an AI writing assistant that happens to listen instead of type. The moment Eloquent becomes a keyboard replacement on Android, Google gains a new layer between users and every app they use – a layer that can observe, assist and, inevitably, monetise.
Historically, whoever controls the input method – from QWERTY to T9 to smartphone keyboards – has wielded enormous power. Eloquent is Google’s attempt to ensure that in the AI era, it still owns the gateway between human intent and digital text.
The European / regional angle
For European users and companies, the most important part of Eloquent is not its clever phrasing; it’s the offline‑first design.
Under GDPR, data minimisation and purpose limitation are core principles. An app that can credibly say, “We process your voice on‑device; the raw audio never leaves your phone unless you opt in,” has a much easier time gaining trust in privacy‑sensitive markets like Germany, the Netherlands or the Nordics. For regulated sectors – healthcare, legal, public administration – offline dictation can be the difference between pilot projects and real deployment.
The flip side is the Gmail vocabulary import. Even if technically compliant, the idea that Google is pulling names and jargon out of personal correspondence will raise eyebrows in Europe. Expect Data Protection Officers to ask detailed questions: Where is this processing done? Is any of that data used to retrain models? Can it be fully disabled at organisation level for Workspace customers?
There’s also a language equity question. English will clearly be first‑class; major EU languages like German, French and Spanish will follow. But what about smaller languages – Slovenian, Slovak, Croatian, Baltic languages? European startups have an opportunity here: narrow but deep coverage for specific languages and domains, with clear GDPR‑first architectures.
Finally, the looming EU AI Act will categorise and constrain certain uses of AI systems. While a generic dictation app likely won’t fall into the highest‑risk categories, the line can blur once Eloquent is used inside medical, legal or HR workflows. Vendors integrating it into business stacks will need to think carefully about compliance, documentation and transparency.
Looking ahead
Assuming Eloquent gains traction on iOS, the logical next step is deep integration into Android and Google’s productivity stack.
On Android, making Eloquent the default or recommended keyboard could trigger DMA scrutiny in the EU: is Google unfairly privileging its own AI service over rivals? Even without defaults, tight integration with Gboard, Google Docs and Gmail would make Eloquent hard to ignore for Workspace users.
From a product standpoint, expect:
- Richer editing controls: style presets for legal, academic, marketing text; multilingual dictation; live formatting.
- Team features: shared custom vocabularies for companies, domain‑specific models for medicine or law.
- Deeper telemetry (opt‑in, hopefully): analytics about how and where people dictate, feeding back into product decisions and ad strategy.
Unanswered questions remain. Will Google keep Eloquent completely free, or introduce a paid tier tied to Workspace or Gemini subscriptions? Will Apple respond by upgrading its own on‑device dictation and locking down deeper system hooks for third‑party keyboards? And crucially for European regulators: how transparent will Google be about what runs where – on device vs in the cloud – and how data is used?
The opportunity is clear: if Eloquent works well in many languages and respects privacy, it could become the default way millions of people “type”. The risk is that we end up with another opaque, unaccountable AI layer sitting between our thoughts and our documents.
The bottom line
Google’s offline‑first dictation app is less a quirky experiment and more a strategic bet on voice‑driven, on‑device AI as the next default input method. It challenges startups’ business models, pressures Apple to respond and opens fresh regulatory questions in Europe around data use and platform power. The real test will be language breadth, transparency and restraint.
The question for users and policymakers alike: who do you want sitting between your voice and your words – and on whose terms?
