These 14 Gemini prompts are built around what Google's own documentation says works: precise instructions, tags or headings to separate the parts, context first and the question last, few-shot examples, and features only Gemini has, like Deep Research over your Gmail and Drive and skills you can upload as a SKILL.md. Each one is a fill-in template you can paste as is.
General jobs that work the same in any chatbot, like editing a draft or planning a week, are in the best ChatGPT prompts and the Claude prompts. Nothing here repeats them.
How to use these prompts
Copy a prompt, replace each blank in double braces, or keep the default after the colon. {{audience:a product team}} means "a product team" unless you change it.
The prompts follow Google's prompting strategies guide for the Gemini API. It is written for developers, but its rules carry over to chatting in the app. Three of them shape almost every prompt below:
- Structure it. The guide says XML-style tags such as
<context>or<task>, or Markdown headings, are effective delimiters, and that you should pick one format and use it consistently in a prompt. - Context first, question last. For large inputs, the guide says to supply all the context first and put your instructions or question at the very end, then bridge with a phrase such as "Based on the information above...".
- Ask for the length you want. In the guide's words: "By default, Gemini 3 models provide direct and efficient answers." If you want detail, say so and name the sections.
Google's Workspace prompting tips add the plain-language side: "Talk to Gemini just like you would a friend or colleague," give it context, and break complex tasks into separate prompts.
How do you prompt Gemini Deep Research?
Deep Research runs many searches and writes a report. According to Google's Deep Research help page, Google Search is a source by default, and you can add Gmail, Drive, uploaded files or your NotebookLM notebooks. Gemini shows a research plan you can change with Edit plan before you click Start research. A report usually takes about 5 to 10 minutes.
1. A Deep Research brief that sets the scope
Use it to start any Deep Research run. A tight brief gives you a better plan to edit, which is the cheapest place to fix a report.
<task>
Research {{topic}} so I can decide {{decision:whether to enter this market next year}}.
</task>
<scope>
Time window: {{time_window:the last 24 months}}
Region: {{region:United States and the EU}}
Favor: {{favor_sources:official statistics, regulators, company filings}}
Avoid: {{avoid_sources:vendor marketing pages and undated blog posts}}
</scope>
<output>
Sections: summary in five bullets, current state with numbers,
key players, open risks, what would change the conclusion.
Date every statistic and name its source in the text.
Flag any claim supported by only one source.
</output>
2. Research that includes your own Gmail and Drive
Turn on Gmail and Drive under Sources first. This is the Deep Research job a web-only tool can't do: combining what the web says with what your team already wrote.
<task>
Build a briefing on {{project_or_client}} for {{meeting:my call on Thursday}}.
</task>
<sources>
Use my Gmail and Drive for what we have discussed and agreed.
Use Google Search only for public facts about {{project_or_client}}.
Keep the two clearly separated in the report.
</sources>
<output>
1. What we agreed, with the date and the email or doc it came from.
2. Open questions and promises still outstanding, with owners.
3. Public news from {{time_window:the last 90 days}} that affects us.
4. Three questions I should ask on the call.
</output>
3. Turn a finished report into a decision memo
Paste this as a follow-up in the same chat once the report is done. It forces the report into something you can act on.
Based on the report above, write a one-page decision memo for
{{reader:my manager}}.
Recommend one option. Give the three strongest reasons, each tied to a
finding in the report. Then give the strongest argument against it and
what evidence would make you change the recommendation.
Mark every number that came from a single source. Keep it under
{{length:400 words}}.
Which Gemini prompts work best with long documents?
When you paste a lot of material, placement matters. These prompts put the material first, the instructions last, and use the guide's bridging phrase.
4. Answer questions across several long documents
For contracts, research papers or a long spec plus its appendices. Tag each document so Gemini can cite which one it used.
<document name="{{doc_1_name:Contract}}">
{{doc_1}}
</document>
<document name="{{doc_2_name:Amendment}}">
{{doc_2}}
</document>
Based on the information above, answer the questions below. For each
answer, name the document and quote the sentence that supports it. If
the documents conflict, show both passages and say which one appears
to take precedence and why. If no document answers a question, say
so and name what kind of document would.
<questions>
{{questions}}
</questions>
5. Define the terms before the analysis
The guide tells you to "explicitly explain any ambiguous terms or parameters." Use this whenever a word like "active user" or "churn" could be read two ways.
<data>
{{data}}
</data>
<definitions>
{{term_1:Active user}}: {{definition_1:signed in at least once in the last 28 days}}
{{term_2:Churned}}: {{definition_2:no sign-in for 60 days after being active}}
</definitions>
Based on the data above and using only the definitions given, answer:
{{question:how did churn change month over month?}}
Show the calculation for each number. If the data can't support an
answer under these definitions, say what is missing. Give a detailed
answer with a short method note, not just the final figures.
How do you use few-shot examples with Gemini?
Google's guide is unusually direct here: "We recommend to always include few-shot examples in your prompts." It also warns that too many examples can make the model overfit, and that every example should use the same structure.
6. Classify items with consistent examples
For sorting support tickets, feedback or survey answers into your own categories. Three to five examples in an identical format work better than a long rule list.
Classify each item into one of: {{labels:Bug, Feature request, Billing, Other}}.
Answer with the label only, one per line, in the same order.
Item: The export button does nothing in Safari.
Label: Bug
Item: Could you add a dark mode?
Label: Feature request
Item: I was charged twice this month.
Label: Billing
Item: {{item_1}}
Label:
Item: {{item_2}}
Label:
7. Rewrite in a house style from examples
When you need copy that matches existing pages. The examples carry the style better than adjectives do.
<examples>
<example>
<before>{{before_1}}</before>
<after>{{after_1}}</after>
</example>
<example>
<before>{{before_2}}</before>
<after>{{after_2}}</after>
</example>
</examples>
<task>
Rewrite the text below the same way the examples do. Match the length,
sentence style and level of formality of the "after" versions. Keep
every fact. Return only the rewritten text.
</task>
<text>
{{text}}
</text>
How do you prompt Gemini with images and files?
The guide says to treat text, images, audio and video as equal-class inputs and to "clearly reference each modality as needed." In practice that means telling Gemini what each attachment is and what to do with it.
8. Check a screenshot against a spec
Attach the screenshot, paste the spec. Naming each input stops Gemini from describing the image instead of comparing.
The attached image is a screenshot of {{screen:the checkout page}}.
The text below is the spec it should match.
<spec>
{{spec}}
</spec>
Compare the screenshot to the spec. List every mismatch as a table:
element, what the spec says, what the screenshot shows, severity
(high, medium, low). Then list anything in the screenshot the spec
doesn't mention. Don't comment on visual taste, only on the spec.
9. Read a chart and test a claim
Attach a chart image from a report or slide. Useful before you repeat a claim someone made about it.
The attached image is a chart from {{source:a quarterly report}}.
Someone claims: "{{claim}}"
1. Describe what the chart shows: axes, units, time range, series.
2. Read off the values that matter for the claim, and say how
precisely they can be read from the image.
3. Say whether the chart supports the claim, partly supports it, or
contradicts it, and explain in two sentences.
4. Note anything misleading about the chart itself, such as a
truncated axis.
10. Current facts with dates and sources
The guide says grounding with search should be on "whenever the model may need to know obscure or recent facts." This prompt makes the answer show where each fact came from.
What is the current state of {{topic}} as of today?
Search the web. For every fact, give the date it was published and
name the source in the sentence. Prefer {{preferred_sources:official sites and primary sources}}.
If sources disagree or a fact may have changed recently, say so.
Finish with what I should check myself before relying on this.
How do you turn a prompt into a Gemini skill?
A skill is "a set of reusable instructions and additional context," per Google's skills help page. You can create one from a chat or the Skills page, or upload a SKILL.md file, or a folder or .zip with SKILL.md in its main folder. The name must be lowercase with hyphens, uploads must stay under 100 MB in total, and you need to be 18 or over with Keep Activity on. To use one in a chat, type / (or @ with the updated shortcut) and pick it.
Gems are on the way out for personal accounts of people 18 and over. Google's transition page says Gems become read-only on November 17, 2026 and are removed on December 17, 2026, converting to skills automatically. Workspace business accounts follow in March 2027 and education accounts in June 2027.
11. Draft skill instructions with persona, task, context and format
Google's tips for custom Gems name four areas: persona, task, context and format. The same four make good skill instructions.
Write instructions for a reusable Gemini skill.
Persona: {{persona:a senior editor for a B2B software blog}}
Task: {{task:edit drafts for clarity and cut length by a third}}
Context: {{context:readers are engineering managers; we never use hype words}}
Format: {{format:the edited draft, then a bullet list of the five biggest changes}}
Write the instructions in the second person, under 300 words. Add a
short line saying when the skill should be used, and a list of things
to ask the user for if they are missing.
12. Convert a Gem into a SKILL.md before it goes read-only
Paste your Gem's instructions. You get a file you can upload on the Skills page or keep in your own repo.
Below are the instructions from one of my Gemini Gems.
<gem_instructions>
{{gem_instructions}}
</gem_instructions>
Convert them into a SKILL.md file. Start with YAML frontmatter between
--- lines containing:
- name: lowercase letters, digits and hyphens, at most 64 characters
- description: one or two sentences, under 1,024 characters, saying
what the skill does and when to use it
Then write the instructions as clear Markdown sections. Keep every rule
from the Gem. List any file the Gem relied on that I need to add to
the skill folder.
How do you plan a multi-step task in Gemini?
The guide recommends breaking work into components: "Instead of having many instructions in one prompt, create one prompt per instruction," then feed each output into the next prompt.
13. Split a big request into a prompt chain
Use it when one giant prompt keeps giving you half of what you asked for.
I want to {{goal:turn 40 customer interviews into a prioritized roadmap}}.
Don't do the task yet. Break it into a chain of 3 to 6 prompts I can
run one after another, where each step's output is the next step's
input. For each step give:
- the exact prompt text, with a placeholder for the previous output
- what the output should look like
- how I can check it is right before moving on
14. Ask for a hard answer and the thinking budget to match
The guide notes that for heavy reasoning, a simple request like "Think very hard before answering" can help, at the cost of extra thinking tokens. Save this for problems that deserve it.
Think very hard before answering.
<problem>
{{problem}}
</problem>
<constraints>
{{constraints:must run on our current stack, no new vendors}}
</constraints>
Give the final answer first in two sentences. Then the reasoning in
numbered steps, the assumptions you made, and the one assumption that
would most change the answer if it were wrong.
Keep your Gemini prompts and skills in one place
Prompt 12 and the skills feature make this easy to keep tidy: once a prompt is a SKILL.md, it works in Gemini and in other tools that read the same format. The free SKILL.md generator turns any prompt above into a valid file in your browser and checks the name and description limits. If you also use Gemini in the terminal, the Gemini CLI guide and the Gemini CLI skills guide cover where skills live there.
To keep the prompts themselves one shortcut away, save them in Kipt. Your prompts live as Markdown in a private GitHub repo you own, blanks like {{topic}} stay fillable, and Cmd+K finds any prompt by name or tag. Kipt is not affiliated with Google.