How Google Now, Siri & Cortana Predict What You Want
Google, Apple and Microsoft all have agents that want to be your personal assistant. But how well Google Now, Apple's Siri and Microsoft's Cortana can predict your needs depends on how much you want to share, how wedded to particular platforms …
Context & Ripple Effects
This comparison lands in the opening phase of the assistant wars: months earlier, Apple announced its Proactive Assistant as a direct Google Now competitor, putting all three platform owners — Google, Apple, Microsoft — in the same predictive-assistant race simultaneously.
The piece frames the core trade-off that defined the next decade: an assistant's ability to predict needs scales with how much personal data you hand it and how locked into one vendor's platforms you are. Within a year Google answered the prediction problem with structure, replacing passive cards with the conversational Google Assistant unveiled in May 2016, and Sundar Pichai laid out the company's broader vision for AI across messaging and beyond.
First-order effects
- Users of all three agents face an immediate choice with no free option: get better predictions by sharing more data and staying inside one company's ecosystem, or keep privacy and accept a dumber assistant.
Second-order effects
- Platform ownership becomes the competitive weapon rather than algorithmic cleverness — Google's response was to push Assistant everywhere it could ship software, per its later plan to make the assistant ubiquitous against Siri's and Alexa's incumbent device advantages, while Microsoft leaned Cortana into Windows and adjacent surfaces.
- Automakers spotted the same anticipation logic and began embedding mood-aware, Alexa- and Nuance-powered assistants into cars, extending the assistants' battleground beyond phones and desktops.
Third-order effects
- A decade of escalation produced strikingly flat results: 2025 YouGov polling shows US adults using assistants mostly for weather, music, web answers and timers — essentially the same task profile as 2018 — suggesting prediction quality never broke through to new behaviors and that whoever controls the distribution surface, not the smartest model, keeps the user relationship.
The trend: Personal assistants have evolved from passive prediction engines like Google Now into conversation-first agents pushed onto every surface, yet adoption remains anchored to a handful of simple utility tasks.