Gemini App

The Gemini app is a program by Google for phone, tablet, and browser. You ask a question in it using plain language, and it answers in plain language. Behind it is a computer program that has learned how language works from enormous amounts of text. Such programs are called language models, and Google’s family of them is named Gemini. The app is therefore just the interface: the window through which you talk to this technology. Besides text, you can also upload images, photos, or documents and ask questions about them.

Google’s answer to ChatGPT

Google has earned its money from search for decades. Anyone wanting to know something types it in, clicks on links, and sees ads along the way. If users instead ask their questions to an AI that answers directly, this business comes under pressure. That is exactly what happened when ChatGPT, from the company OpenAI, appeared at the end of 2022 and gained millions of users within a few months.

The Gemini app is Google’s direct response to this. It was first called Bard and was renamed Gemini in 2024, matching the name of the models. For investors, the app is therefore a metric: How many people use it regularly? How fast is that number growing compared to ChatGPT? Google now mentions such user numbers in its quarterly reports, because the stock market watches this closely.

It is important to distinguish between the app and the model. The model is the learning part, the app is the packaging. Google also builds the same technology into Gmail, into Android, and into Search. The app is nonetheless special, because it’s where you get to see the newest features first.

From typing to answering

When you type something in, the actual computational work does not happen on your phone. Your input is sent to Google’s data centers. There, the model runs on specialized chips and computes the answer word by word. Each new word is chosen from whatever statistically fits best. The result comes back and appears on your screen, often while it is still being generated.

The app usually offers several model tiers. A fast, cheap variant answers simple questions. A more powerful variant takes more computing time for difficult tasks, such as mathematics or longer analyses. These more powerful tiers are often tied to a subscription, because every answer incurs computing costs.

A common misconception is that the app looks things up in a database. It normally does not. It generates language that sounds plausible, and in doing so can make things up. Such confidently delivered false statements are called hallucinations. That’s why, for current-events questions, Gemini can additionally run an actual web search and link to sources.

Gemini on the phone and in the headlines

On newer Android phones, Gemini replaces the old Google Assistant. You hold down a button or say a wake phrase and ask your question. Many people use the app to summarize texts, practice vocabulary, get homework explained, or write job applications. If you photograph a math problem, the app tries to explain the steps to solve it.

The same topics keep coming up in news coverage. First, user numbers and market share versus ChatGPT. Second, mishaps, such as incorrect historical claims or inappropriately generated images. Third, privacy, because conversations can be analyzed to improve the models. Anyone entering sensitive things should know the settings for activity history.

For context, a comparison helps: ChatGPT belongs to OpenAI, Copilot to Microsoft, Claude to Anthropic. All are interfaces for similar technology. The Gemini app is distinguished above all by its close connection to Google’s own services such as Search, Maps, YouTube, and Gmail.