Best Gemini Prompts

Start with the source, then the job

What sets Gemini apart is what it can reach: your Gmail, Docs and Drive, public YouTube videos, Google Search, and whatever you upload. A prompt that points at none of that gets the answer any chatbot would give. That's why every prompt on this page opens by naming the material, and then spends the rest of its length on what to do with it. Five things carry most of the weight.

1. The source, named the way Gemini finds it. Type @, pick the service, then describe the file with words that are in it: the sender, the subject line, the title. Google's own help page says not to paste the URL of a document or email. «@Gmail the thread with Northwind about the September renewal» is what works.
2. The decision it feeds. «Compare these proposals» gets a summary. «Compare them so I can choose on two-year cost and how easy it is to leave» gets a table you can act on.
3. The shape of the answer. Rows and columns for anything headed into Sheets, numbered sections for anything you'll forward.
4. What to write when the file is silent. «Not mentioned», or [confirm figure]. Without that line, a gap in your Doc gets filled from general knowledge, and it reads exactly like something the file said.
5. Where each point came from. A page, a section, an email date. Google warns that Gemini can answer from an older email when a newer one exists, and the dates are how you'll spot it.

Here's the pattern short enough to type from memory:

@[Gmail / Drive / Docs] Find [the file, in words it contains].
I need it to decide [what you're deciding].
Answer as [a table / a numbered list] with [the parts].
If the file doesn't say, write "not mentioned".
Put the date or section next to every point.

The prompts below are that pattern worked out for one job each. The Gmail Thread Summarizer treats the newest message as the current state, so a reversed decision doesn't come back to life. The Document Comparison Table lists the files it opened before comparing anything, which tells you straight away if it grabbed the wrong one.

Which prompt for which job

The third column matters more here than on most hubs. Have that material ready before you paste.

Editing your own photos in Gemini is a different kind of prompt, and it has its own collection: Gemini photo prompts. For prompts that don't need your Google account at all, the ChatGPT and Claude pages run fine in Gemini, and our Gemini prompt examples sort eighteen more blocks by what you hand over.

One request, before and after

We wrote this pair to show what naming the source does. The first is how a busy Monday request tends to look:

Summarize the budget emails.

Gemini has to guess which emails, from whom, and over what period. You'll get a summary of something, and you won't know what. The second version names all of it:

@Gmail Find the emails from Dana Ortiz about the Q4 event budget,
sent in September and October. I need to know what's approved
before Thursday's planning call. Give me the approved total, the
lines still open, and anything Dana is waiting on from me. Where a
figure was revised, use only the latest one and show the email date.

Now the search has a sender, a topic and a window. The latest-figure rule deals with the budget that changed twice, and the date beside each number lets you open the right email in one click if something looks off. For a single thread, the Gmail Thread Summarizer asks for all of this already.

Follow-up lines that fix an answer

Stay in the same chat. Gemini keeps track of the source you named, so the repair is one line.

Where Gemini answers go wrong

These come from what Google's help pages say Gemini can't do, and they explain most odd results with your own files.

• It can't see everything in a file. Comments and images inside Docs and Gmail are off limits, and so are photos and videos stored in Drive. A folder of receipt photos in Drive won't work through @Drive, so upload the pictures straight into the chat.
• It can't count your stuff. «How many invoices did I get this year?» asks Gemini to count items in Gmail or Drive, which Google lists as something it can't do. Ask it to list them for a short window and count the list yourself.
• Work and school accounts may be blocked. An administrator has to allow connected apps first. If @Gmail does nothing on a work account, check that first; pasting the text gets you going meanwhile.
• Long uploads get skimmed. When a file is too big for the plan's reading capacity, Google says the answer can miss connections scattered through it. Split the file, or ask about one section at a time.
• A new chat forgets the service. You have to mention Gmail, Docs or Drive in the first prompt of every conversation. Follow-ups in the same chat don't need it.

Gemini settings that change the result

These are as of October 2026, from Google's Gemini help center. Google changes them often.

• Models and thinking. Gemini 3 Flash-Lite, Flash and Pro are available on every plan, including no plan at all. Standard thinking is the default; Extended thinking reasons longer, and Deep Think needs Google AI Ultra. Usage is measured by compute, so Pro and higher thinking levels use your allowance faster. It refreshes every five hours until you reach a weekly limit. For the Spreadsheet Data Analyst or the Document Comparison Table, Extended is worth it.
• How much it can read at once. The context window is 32K tokens without a plan, 128K on Google AI Plus, and 1 million on Pro and Ultra. A long PDF that works on Pro can lose detail on a free account.
• Uploads. Up to 10 files per prompt, 100 MB each, or 2 GB for a video. Uploaded video is capped at five minutes in total without Pro or Ultra, and at one hour with them.
• Deep Research. It's on every plan for users 18 and over, with higher daily limits on Pro and Ultra. Google Search is a source by default; untick it if the report should come only from your Gmail, Drive, uploads or NotebookLM notebooks. Gemini shows a research plan first, and Edit plan is where the Deep Research Brief's scope goes. A report usually takes five to ten minutes and can be exported to Docs.
• Gems are turning into skills. Starting in November 2026, Gems go away on personal Google accounts and are recreated automatically as skills, which you call with a slash in any chat. Work accounts follow in 2027. Skills don't yet work with Canvas or Deep Research. What the Gem Builder writes still fits, because a skill takes the same name, description and instructions.

Our Gemini review keeps the plans and prices, and Gemini vs ChatGPT covers which one to open for what.