How to use Gemini well

What Gemini is shaped for

Work that lives in Google Workspace, and prompts that include images, audio, or video rather than only text.

• Native handling of images, audio, and video as input, not as an add-on
• Direct, permissioned access to your own Drive, Gmail, and Docs
• Works inside the document you're already editing
• A deeper research mode that assembles longer, structured reports

Where it gets its information

Training data, Google Search, and — with your permission — your own Drive, Gmail, and Docs. The most important question before any prompt is which of those three it is actually drawing on this time.

How to prompt it well

• Name the source you mean. "Using the doc in my Drive called Q3 Plan…" removes the main ambiguity, since it can pull from your files, the web, or training.
• Give it non-text input. A screenshot of a broken chart, a photo of a whiteboard, or a recording is a first-class prompt here, not a workaround.
• Use it inside the document you're editing rather than copying text into a separate chat, so it has the surrounding context automatically.
• Ask it to show which sources it used when the answer mixes web results with your own files.

The fundamentals — goal, context, constraints, format — work the same everywhere. Build one in the prompt lab →

Giving it context

• Workspace access is permission-scoped — it cannot see a file you have not shared with yourself.
• Reference documents by their exact title.
• Long threads mixing personal files and web research get muddled; keep them separate.

Getting the output format you want

• Ask for output as a Doc or Sheet when the destination is Workspace.
• For multimodal input, say what to attend to: "read only the axis labels."
• Explicit bullet-count instructions are followed well.

Files, documents, and research

Strong on your own Drive contents and on mixed-media material. It has a deeper research mode that will go and assemble a longer report from many sources — useful, but it produces length, so state the scope you want. Verify anything it attributes to your own documents by opening them.

Example prompts

Using the doc "Q3 Plan" in my Drive, draft an email to the team explaining the two changes in section 4. Match the plain tone of the doc.

Research this topic and give me a structured brief: what's established, what's contested, what's unknown. Link every claim.

Here's a screenshot of the error and the file. Tell me what's wrong and give the fix.

Here's a photo of my lecture notes. Turn them into a study sheet, and flag anything my notes got wrong.

Compare the two spreadsheets in my Drive and list every row where the totals disagree.

Here's a recording of our 20-minute discussion. Pull out every idea anyone proposed, including the ones we talked past.

Find the thread in my mail about the venue booking and tell me what I still owe them.

Common mistakes

• Assuming it can see a file you never shared or named explicitly.
• Not noticing whether an answer came from your Drive or from the open web.
• Using the standalone chat when the in-document version has far more context available.
• Letting the deep research mode produce far more than was wanted, with no length given.

What it will not do well

• Cannot read files outside the permissions you have granted
• Does not keep a clean line between your private files and public web results unless asked to
• Stays lengthy in its research modes without an explicit limit

Privacy and what you send it

Tied to your Google account, with a meaningful split between personal accounts and managed Workspace accounts — the data handling and admin controls differ substantially. On a work account, your organisation's settings govern. Check which account you are signed into before sharing anything sensitive.

How to check its work

• Open the Drive file it cited and confirm the passage yourself.
• Ask it to separate "from your files" from "from the web" in its answer.
• For multimodal answers, check it read the image or audio correctly before trusting the conclusion.