Every reliable Gemini prompt follows the same three-part shape — Intent (what you want), Context (what Gemini can't infer), and Constraints (format, length, tone). That's the ICC framework, and these eighteen templates turn it into fill-in-the-blank skeletons. Swap the [brackets] with your own details, set a thinking level, and paste straight into Gemini 3.1 Pro or Flash. For ready-to-run examples instead of skeletons, see the best Gemini prompts roundup.
Email & communication templates
These skeletons draft, rewrite, and soften messages. Run them in Flash on the low thinking level, and @-mention the thread in the Gmail side panel so Gemini answers in context.
1. Reply-to-a-thread template
Intent: Draft a reply to the email thread [attach or @-mention the thread in Gmail].
Context: I am [your role]. The other person wants [what they asked for]. My position is [what you can/can't do].
Constraints: [Tone, e.g. warm but firm], under [N] words, [one clear ask or next step], sign off as [your name]. Give me two versions to choose from.How to use: swap the thread reference and your position — @-mention the actual Gmail thread so Gemini reads what was already said instead of guessing.
2. Cold outreach template
Intent: Write a cold outreach email to [recipient role] at [company type].
Context: We offer [product/service] that helps [audience] achieve [outcome]. I found them because [reason/trigger].
Constraints: Under [N] words, one specific hook tied to their [situation], a single low-friction CTA ([e.g. a 15-min call]), no jargon. Subject line plus body.How to use: replace the trigger with a real, specific reason you're reaching out — a recent hire, launch, or post — so the opener doesn't read as a template.
3. Difficult-message rewrite template
Intent: Rewrite my draft below so it lands as [desired tone, e.g. calm and professional].
Context: The situation is [what happened]. My relationship with the reader is [peer/manager/client]. My raw draft: "[paste your unfiltered draft]".
Constraints: Keep the core message, remove any blame or heat, propose [a next step], under [N] words. Flag any line that could still be read as passive-aggressive.How to use: paste your honest, unedited draft into the Context beat — Gemini rewrites the tone while keeping the point you actually need to make.
Docs & writing templates
Skeletons for drafting and reshaping long-form text. In Google Docs, @-mention the source file so Gemini works from your real document, and use the standard thinking level.
4. First-draft document template
Intent: Write a first draft of a [document type, e.g. project brief] about [topic].
Context: Audience is [who reads it]. It must cover [point 1], [point 2], and [point 3]. Background material: [@-mention any Drive files or paste notes].
Constraints: [Length], [tone], use headings and short paragraphs, end with [a summary / next steps]. Leave a [TODO] marker anywhere you need input from me.How to use: list the three must-cover points explicitly — that's the Context beat that keeps the draft from wandering off scope.
5. Rewrite & tighten template
Intent: Tighten the document [@-mention the Doc] without changing its meaning.
Context: It is currently [length] and too [wordy/dense/formal] for [audience].
Constraints: Cut it to roughly [target length], keep every fact and heading, prefer plain words, and show me the trimmed version plus a one-line note on what you cut.How to use: set a real target length so Gemini commits to cuts instead of lightly paraphrasing the whole thing.
6. Slides outline from a doc template
Intent: Turn the document [@-mention the Doc] into a slide outline for Google Slides.
Context: The talk is [duration] to [audience] about [goal of the talk].
Constraints: [N] slides, one idea per slide, a headline plus 3 bullet points each, and a speaker note under each slide. Start with a hook slide and end with a call to action.How to use: match the slide count to the talk length — roughly one slide per minute keeps the deck from overflowing.
Research & Deep Research templates
These need grounding with Google Search or Deep Research turned on, and reward the high thinking level. They return sourced, current answers rather than the model's memory. See how to prompt Gemini for Deep Research for the full workflow.
7. Grounded fact-check template
Intent: Fact-check the claims below using grounding with Google Search.
Context: The claims: "[paste claims or the passage]". This is for [use, e.g. a published article].
Constraints: Verify each claim as true, false, or unclear; cite a source link for each; note anything that changed after [date]; flag any claim you can't confirm. Use high thinking level.How to use: turn on grounding with Google Search first, then paste the exact claims — Gemini checks each one and links a source instead of asserting from memory.
8. Deep Research brief template
Intent: Run a Deep Research report on [topic/question].
Context: I need this to [decision or deliverable]. My current knowledge: [what you already know]. Scope: [geography, timeframe, or segment].
Constraints: Cover [sub-question 1], [sub-question 2], and [sub-question 3]; compare at least [N] sources; give a sourced summary, a table of key figures, and open questions. Cite every claim.How to use: switch on Deep Research (add Max for the deepest pass), then name the sub-questions so the agent plans its search around what you actually need to decide.
9. Compare-options template
Intent: Compare [option A], [option B], and [option C] for [my use case], grounded in current sources.
Context: My priorities are [priority 1] and [priority 2]. My constraints are [budget/time/team size].
Constraints: Return a comparison table across [the criteria that matter], a one-line verdict for my case, and source links. Call out anything that's changed recently. Use grounding with Google Search.How to use: spell out your priorities in the Context beat so the verdict is tailored to your case, not a generic "it depends."
Coding templates
Skeletons for reading, writing, and fixing code. Set the high thinking level for anything with real logic, and paste the surrounding code so Gemini matches your style. For a full pack, see Gemini prompts for coding.
10. Explain-this-code template
Intent: Explain what the code below does, step by step.
Context: Language is [language]. It fits into [where it runs / what calls it]. Code:
[paste the code].
Constraints: Walk through it in plain English, name any side effects or edge cases, and flag anything that looks buggy or risky. Keep it under [N] words. My level is [beginner/intermediate].How to use: state your experience level so the explanation is pitched right — beginners get the concepts, experts get straight to the gotchas.
11. Write-a-function template
Intent: Write a [language] function that [does the thing].
Context: It takes [inputs] and returns [output]. It runs in [environment/framework]. Match the style of this existing code: [paste a sample].
Constraints: Handle [edge case 1] and [edge case 2], add brief comments, no external libraries beyond [what's allowed], and include a couple of usage examples. Use high thinking level.How to use: paste one sample of your existing code so Gemini matches your naming and formatting instead of inventing its own conventions.
12. Debug-an-error template
Intent: Find and fix the bug causing the error below.
Context: Language/framework is [stack]. What I expected: [expected behavior]. What happens: [actual behavior]. Error message: "[paste error]". Relevant code: [paste code].
Constraints: Explain the root cause, give the corrected code, and list one way to prevent it next time. Don't rewrite unrelated parts. Use high thinking level.How to use: include the exact error text and both expected and actual behavior — that trio lets Gemini isolate the cause instead of guessing.
Data & Sheets templates
Skeletons for spreadsheet formulas and dataset analysis. @-mention the Sheet in the side panel so Gemini reads real columns, and use the high thinking level for multi-step analysis. See Gemini prompts for data analysis for more.
13. Sheets formula template
Intent: Write a Google Sheets formula that [does the calculation].
Context: My data is in [@-mention the Sheet or describe the ranges], with [column A = ..., column B = ...]. I want the result in [cell/column].
Constraints: Use a single formula if possible, explain each part in one line, and note any assumption about my data layout. Give a fallback if [an edge case, e.g. blanks or errors] appears.How to use: describe your columns exactly (or @-mention the Sheet) so the formula points at the right ranges the first time.
14. Analyze-a-dataset template
Intent: Analyze the dataset [@-mention the Sheet or attach the file] and tell me [the question you care about].
Context: Each row is [what a row represents]. The columns I care about are [columns]. Timeframe is [range].
Constraints: Give the top [N] findings with the numbers behind them, one surprising insight, and any data-quality caveats. Suggest the chart type that best shows the main finding. Use high thinking level.How to use: name the question you actually care about — "which channel drove the most signups" beats "analyze this," and the findings come back focused.
15. Pivot & chart plan template
Intent: Plan a pivot table and chart to show [what you want to see] from [@-mention the Sheet].
Context: The relevant columns are [columns]. My audience is [who sees it].
Constraints: Tell me which fields go in rows, columns, and values; which aggregation to use; and the single chart type that reads clearly for my audience. Give the step-by-step to build it in Sheets.How to use: state who the chart is for — an exec summary and an analyst deep-dive call for different pivots and chart types.
Study & learning templates
Skeletons that explain, condense, and quiz. These run well in Flash on the standard thinking level, and you can @-mention your own notes as the source. For students, see Gemini prompts for students.
16. Explain-a-concept template
Intent: Explain [concept] so I actually get it.
Context: I already understand [what you know]. I get stuck on [where you're stuck]. I'm studying this for [exam/project/curiosity].
Constraints: Use one everyday analogy, then the precise definition, then a worked example. Keep it under [N] words and end with the single most common misconception about it.How to use: tell Gemini where you get stuck — the explanation then targets your actual gap instead of starting from zero.
17. Study guide from notes template
Intent: Turn my notes into a study guide for [subject / exam].
Context: Source material: [@-mention your notes Doc or paste them]. The exam covers [topics] and is [format, e.g. multiple choice].
Constraints: Organize by topic, give a one-line summary plus key terms for each, add [N] flashcard-style Q&A per topic, and flag the [N] topics most likely to be tested. Keep definitions in my own note's wording where possible.How to use: @-mention your real notes so the guide reviews what you were taught, not a generic version of the subject.
18. Practice quiz template
Intent: Make a practice quiz on [topic] at [difficulty level].
Context: I'm preparing for [exam/goal]. Base it on [@-mention notes or describe the scope].
Constraints: [N] questions in [format, e.g. mix of multiple choice and short answer], no answers shown until I ask, then give the answer key with a one-line explanation for each. Focus on [the areas you're weakest on].How to use: ask it to withhold the answers first so you actually test yourself, then request the key to check your work.
Frequently Asked Questions
How do I fill in the [brackets]?
Replace each bracketed phrase with one specific, concrete detail. Instead of [audience], write "our enterprise customers on annual contracts." Gemini follows a clear Intent, real Context, and named Constraints far better than vague placeholders — that is the ICC framework these templates are built on.
Which thinking level should I set?
Use the low thinking level for quick email and rewrite templates, standard for docs and Sheets formulas, and high for the research, debugging, and data-analysis templates that need multi-step reasoning. Higher levels are slower but reason through more edge cases, so match the level to the difficulty of the task.
Do these templates work in Gemini 3.1 Flash?
Yes. Every template runs in Flash, which is faster and cheaper for straightforward email, doc, and study tasks. For the Deep Research, coding, and data-analysis templates, Gemini 3.1 Pro on a higher thinking level gives more reliable results, but Flash handles the rest well.
How do I reference my own Workspace files in a template?
Where a template says [attach or @-mention the file], type @ in the Gemini side panel inside Gmail, Docs, Sheets, Slides, or Drive and pick the exact file or thread. Gemini reads that file as context, so name the file and the specific tab, range, or section you want it to use.
Can I turn a template into a Gem?
Yes. Paste the filled-in Intent and Constraints beats into a new Gem's instructions, leave the Context beat for the user to supply each time, and save it. The Gem then runs your template on demand without re-pasting, which is ideal for repeated tasks like weekly reports or standard replies.
Do I need grounding with Google Search for every template?
No. Only the research and fact-check templates need grounding with Google Search, and they say so in the prompt. Email, doc, coding, and study templates work from the context you provide, so leave grounding off for those to keep answers focused on your material.
How do I adapt a template to a totally different task?
Keep the three-part ICC skeleton — Intent, Context, Constraints — and swap the domain nouns. A study-guide template becomes an onboarding-guide template by changing [subject] to [new-hire role] and the source from [notes] to [handbook], while the structure that makes it reliable stays intact.
What's the ideal length for a filled-in Gemini template?
Enough to be unambiguous but no more. State the intent in one line, give the context Gemini can't infer, and list three to five concrete constraints. Gemini 3.1 handles a 1M-token context, so pasting a long source document is fine — it is vague instructions, not length, that produce weak output.