Gemini
by Google DeepMind
Getting-started guide
Google's assistant, free and genuinely capable — with a privacy model built on human review rather than a training switch.
Current prices: Free, from $4.99/month · See all plans
Last reviewed 2026-10-14 · for Gemini
You open Gemini in a browser or its app, signed in with your Google account, and ask in plain language. It answers, and you keep going — attach a file, ask a follow-up, ask it to try again — in one running conversation. The same assistant also appears inside Google's other products, working on whatever you have open there.
A general-purpose assistant with an unusually capable free tier — chat, image generation and editing, deep research, and live voice conversation at no cost.
Tied into the Google account you already have, with paid tiers bundling cloud storage and other Google subscriptions alongside higher usage.
Strong on very long inputs, with paid tiers accepting document sets far larger than most assistants handle.
Governed by Google's activity and human-review model rather than by a single train-or-don't switch.
The free tier needs nothing but a Google account — no card, no trial clock. It includes chat, image generation and editing, deep research, live voice, and the notebook research tool.
Paid tiers are now named Google AI Plus (2× Free usage), Google AI Pro (4×), and Google AI Ultra, the top tier; current prices are on its Launchpad page. Do not use the retired “Gemini Advanced” name.
Limits are compute-based rather than message-counted, refresh on a five-hour cycle against a weekly ceiling, and can be extended by buying credits.
Decide what you are willing to have a human read before you start typing. See the security section — this is the part that differs most from its competitors.
Gemini's privacy model is not a training toggle. Google's stated position is that a portion of conversations are read and annotated by human reviewers, including reviewers at service providers, to improve its models and services. Google's own guidance is unusually direct about the implication: do not enter anything confidential that you would not want a reviewer to see.
Set against a tool where one switch turns training off, this is a different shape of risk rather than a larger or smaller one. There is no configuration that gets you to zero retention, and the meaningful control is not a setting — it is what you choose to type. Use temporary conversations for anything sensitive, and treat the sensitive material itself as the thing to withhold.
I'm attaching several documents. Read all of them, then tell me where they disagree with each other. I care more about the contradictions than the summary.
Why this one: Its capacity for large inputs is the clearest advantage it has, and almost nobody uses it.
Tweak: Ask which document is the outlier and why — that surfaces the assumption the set is built on.
Searches across many sources and returns a structured report rather than a chat answer.
Why this one: Available on the free tier, and it is a different tool from ordinary chat — worth learning as its own thing.
Tweak: Give it the decision you are trying to make, not the topic. "Should I do X or Y, and what would change my mind" beats "tell me about X".
[Start a temporary conversation, then ask your question as normal.]
Why this one: The one concrete control that meaningfully reduces exposure, and it is off the beaten path.
Tweak: Make it a habit tied to content, not to mood — anything with a name, a number, or an employer in it.
Answer the question, then list the specific sources you used and which part of the answer each one supports.
Why this one: Builds the verification reflex early, while the stakes are low.
Tweak: Pick the claim that most surprised you and check that one. Surprise is a good relevance filter.
Want the full picture first? See the Gemini overview for pricing, the free tier and related tools.
