Gboard could soon use your screenshots and chats to draft much better replies than you

At a glance:

• Google is testing custom prompt instructions for Gboard's Writing Tools to move beyond predefined styles.
• New features may allow the keyboard to draft messages using on-screen context or gallery screenshots.
• The AI functionality is expected to run on-device via Gemini Nano to maintain privacy and speed.

Expanding beyond predefined writing styles

Google is significantly evolving its Gboard Writing Tools, moving away from a rigid set of presets toward a more fluid, user-driven experience. While the current iteration of Writing Tools allows users to select from established styles—such as professional, friendly, or "emojify"—new code discovered in the Gboard beta suggests a much more powerful capability is on the horizon. Users will soon be able to input custom instructions via a dedicated text box to dictate exactly how their text should be modified.

This shift toward custom prompting allows for much more granular control over tone and intent. Instead of being limited to a "professional" button, a user might type specific instructions like "less robotic," "make a joke," or "use corporate jargon." This level of customization mimics the experience of using advanced LLM interfaces like ChatGPT or Gemini directly within the keyboard, effectively turning Gboard into a sophisticated, context-aware writing assistant.

Contextual awareness through screenshots and screen reading

Perhaps the most ambitious update involves how Gboard understands the world around the user. Recent findings indicate that Google is testing the ability for the keyboard to leverage on-screen context and even screenshots from the device's media gallery to draft text. This would allow the AI to "see" what you are looking at—be it a text conversation, a social media post, or an email—and provide highly relevant, contextually accurate replies without the user needing to copy and paste information.

To facilitate this, users would likely need to grant Gboard specific permissions to access their screenshot folders. While the exact implementation remains unclear, there are hints that the feature might focus on the most recent screenshot to maintain a balance between utility and privacy. A similar workflow already exists for Pixel users through the integration of Gboard and the Pixel Screenshot app, suggesting Google is working to unify these intelligence layers across the broader Android ecosystem.

On-device intelligence and the role of Gemini Nano

At the heart of these upcoming features is Google's on-device AI model, Gemini Nano. By processing these complex writing tasks locally on the device rather than in the cloud, Google can offer faster response times and enhanced privacy, as sensitive chat data and screenshots wouldn't necessarily need to leave the handset. This approach aligns with Google's recent push to integrate generative AI directly into flagship hardware, a trend that began with the launch of the Pixel 10 and its integrated Writing Tools.

However, the rollout of these advanced features may not be universal. Because running large language models on-device is computationally expensive, Google's internal guidelines suggest that certain features might be limited to devices with sufficient RAM. This could create a tiered experience where high-end flagship phones enjoy the full suite of contextual drafting and custom prompting, while mid-range or older devices are restricted to simpler text correction and basic rewriting tasks.