01
Multimodal from the start
Many assistants gained image and audio understanding by attaching separate components to a text model. Google designed Gemini to take multiple input types natively.
In practice this shows in tasks that mix modes: reading a chart and reasoning about the numbers in it, working from a screenshot, or following something across video frames. It is less about any single benchmark and more about not losing information when converting between formats.
02
Grounding, and why you should demand sources
Gemini can retrieve from Google Search while generating an answer, which means a response can be anchored to real pages rather than to training data alone.
This is a genuine advantage over an ungrounded model, but it is not automatic and it is not a guarantee. Grounding narrows the gap between a fluent answer and a true one; it does not close it. Asking for the source and date of each factual claim is cheap and changes the quality of what you get back.
03
The Workspace connection is the real differentiator
Gemini can reach content in your Google account: mail, documents, files. That changes the shape of a good prompt. Instead of pasting a document into the chat, you point at it.
It also changes the privacy calculation, because you are granting a model access to your correspondence and files rather than choosing what to paste. Worth deciding deliberately rather than by default.
- Name the file rather than pasting its contents.
- Ask for source and date on anything factual.
- Say which register of language you want, or you will get a neutral corporate tone.
- Remember that what you say is stored in Gemini Apps Activity until you turn that off.
Frequently asked questions
Is Gemini the same as Google Assistant?
No. Assistant was a command-and-control system built around triggers and actions. Gemini is a generative model that can hold a conversation and produce new content. Google has been replacing Assistant with Gemini across its products, which is why the assistant on many Android phones changed without being asked.
Does Gemini search Google every time?
Not necessarily. It can ground answers in Search, but whether it does depends on the surface and the query. Because you cannot tell by looking at a fluent answer, explicitly asking for sources and dates is the reliable way to find out whether anything was actually retrieved.
Is Gemini free?
There is a free tier at gemini.google.com. Google AI Pro and Google AI Ultra are the paid tiers, which raise usage limits and unlock additional features. Pricing and what sits in each tier change, so check Google's own pricing page rather than trusting a figure in an article.
Can Gemini read my Gmail without permission?
Access depends on the Google account settings and smart features you have enabled. It is worth reviewing those deliberately rather than assuming the default is what you want, and the same settings control Gemini features across Gmail, Docs, Drive and Maps.