The best prompts for Gemini

Gemini is Google's AI assistant and the family of models behind it. You may meet it as a standalone app, as a helper built into Google's productivity tools, or through the developer API. The surface changes what it can see, such as an open document or files you attach, but the way to ask for good work stays much the same.

Google's prompt design documentation treats prompting as an iterative process. You are not expected to get it right on the first attempt. Start with a clear instruction, look at what comes back, and adjust: rephrase, add an example, change the order of the parts. That framing is worth adopting, because it turns a disappointing answer into information about what the prompt was missing.

A Gemini prompt generally has up to four ingredients: the task, any constraints on how it should be done, the context or source material, and the format of the response. Leaving one out is the usual cause of generic output. If you do not say how long the answer should be or who it is for, the model has to guess.

Standing guidance has its own place. Developers set system instructions that apply to a whole session; in the app, saved instructions and reusable custom assistants play a similar role. Use these for a persona, a house style or rules that should hold across many requests, and keep one-off details in the prompt.

Gemini accepts more than text. Images, documents and other files can be part of a prompt, and the same principles apply: say what the file is and what you want done with it.

Version names, limits and pricing are omitted here because they change. Which features you have depends on your account and region.