I Tested ChatGPT, Claude, Gemini, and Perplexity as Android’s Default Assistant
Pixel 9 Pro XL Test Puts Four Major AI Assistants Through Ten Chores
In a detailed assessment published on makeuseof.com, reviewer Bryan M. Wolfe tested how four major artificial intelligence assistants handle everyday mobile tasks when assigned as the default handler on a Google Pixel 9 Pro XL running Android 17. The evaluation compared ChatGPT, Claude, Gemini, and Perplexity by running each assistant through an identical script of ten routine phone commands.
The Tech TL;DR:
- Google’s preinstalled Gemini was the only assistant out of the four tested to successfully complete all 10 requested chores without requiring manual workarounds.
- Perplexity finished 8 out of 10 tasks but suffered from severe latency, taking nearly 45 seconds to respond to basic weather inquiries compared to Gemini’s 2-second response time.
- ChatGPT and Claude tied by completing 5 out of 10 chores, with both failing on core device controls like flashlight activation and screen summaries.
Configuring the Android Default Assistant Menu
The testing process relied on the native Android default assistant architecture found under Settings, Apps, Default apps, and Digital assistant app. Swapping the primary handler required approximately ten seconds per application. Wolfe maintained a strict testing protocol on the Pixel 9 Pro XL using default models for each platform: GPT-5.6 Sol for ChatGPT, Claude Opus 5.5 for Claude, Gemini AI Pro with version 3.8 Flash, and Perplexity’s automatic selection option. Every service received uniform permissions for notifications, location, calendar, and contacts, while Gemini required explicit screen access to process on-screen summaries.
Performance Breakdown Across Ten Everyday Chores
The evaluation script blended passive utility tasks with active information retrieval queries. Half the commands covered basic automation: setting a 12-minute timer, programming a 6:45 a.m. alarm, dispatching a text message stating running 10 minutes late, scheduling a Friday lunch event, and initiating music playback via Spotify. The remaining five tasks tested contextual awareness, requiring each assistant to summarize an active news article on the screen, locate a nearby gas station, check weekend weather for Miami, and provide currency tipping guidance for Cozumel.
Gemini emerged as the only application to complete the entire roster of ten chores. It activated the flashlight, sent text messages following user confirmation, and initiated Spotify playback without manual taps. Gemini checked the user calendar for scheduling conflicts before adding the Friday lunch entry and successfully read the active news article on the display. In contrast, ChatGPT and Claude could not parse on-screen text, asking instead for manual screenshots or pasted text blocks. Perplexity matched several contextual queries and successfully opened Google Maps for route navigation, but stumbled on contact retrieval by demanding manual phone number entry despite the contact existing in the local address book.
Gemini Leads in Response Time and Lock Screen Access
Response times varied significantly across the test group. Gemini initiated weather responses in approximately 2 seconds, followed by Claude at 3 seconds, ChatGPT at 10 seconds, and Perplexity lagging significantly at 45 seconds per query. Lock screen execution also exposed operational friction. ChatGPT executed commands from the lock screen without requiring device unlock, whereas Gemini and Perplexity demanded manual icon taps or device authentication before processing spoken instructions. Claude failed entirely on the lock screen, prompting the user to unlock the hardware first.
# Example intent resolution check for Android digital assistant services
adb shell cmd role query-holders android.app.role.ASSISTANT
Perplexity’s subscription model presented another operational quirk during testing, as its internal model selector excluded its proprietary Sonar architecture in favor of routing requests through third-party models maintained by OpenAI, Google, Anthropic, and xAI. Meanwhile, ChatGPT ran GPT-5.6 Sol on the Pixel 9 Pro XL while concurrent iOS environments accessed GPT-6.1 Sol, illustrating asynchronous feature rollouts across mobile operating systems.