A well-written Gemini prompt gets you a better answer. The bigger time savings come from using AI regularly, and a set of saved prompts makes regular use easier.
Research from the Federal Reserve Bank of St. Louis shows how much frequency matters. Workers who used generative AI for their job said it saved them 5.4% of their work hours on average, or about 2.2 hours in a 40-hour week.
Among workers who used AI every day that week, 33.5% said it saved them 4 hours or more.
This guide gives you 30 Gemini prompts, sorted by role, that you can save and reuse. It also covers one part of Google’s prompting advice you can skip: telling Gemini to act like an expert.
Short answer: Google recommends four ingredients for a prompt: persona, task, context, and format, but you don’t need all four every time. The persona is the debatable part. This article covers 30 Gemini prompts organized by role, plus the limits you’ll run into and a better way to save prompts you’ll use more than once.
30 Gemini Prompts for Work (And the Google Tip to Skip)
- What Makes a Gemini Prompt Work?
- 30 Gemini Prompts for Work, Sorted by Role
- How Should You Prompt Gemini for Image Generation?
- How to Save a Gemini Prompt with Gems
- Where Gemini Prompts Break in Real Use
- Five Gemini Prompt Mistakes, and What to Do Instead
- How to Keep Gemini Prompts Next to the Work in ClickUp
- Keep the Prompts That Earn Their Place
- Frequently Asked Questions About Gemini Prompts
Summarize this article for me pleaseWhat Makes a Gemini Prompt Work?
A good Gemini prompt can be short. It just has to give the model enough context to work with. According to Google’s own data, successful prompts in Workspace Labs averaged about 21 words.
Google breaks its prompts into four parts: Who should Gemini respond as? What should it do? What information should it use? What should the finished answer look like?
| Element | What it looks like | What it changes |
|---|---|---|
| Task | “Summarize,” “draft,” “compare,” “list every” | Gives Gemini a clear action to perform |
| Context | A file, pasted text, audience details, constraints | Supplies information Gemini would otherwise have to guess |
| Format | “Put this in a table with X, Y, and Z columns.” “Keep it under 120 words.” | Defines the shape of the answer and cuts cleanup |
| Persona | “You’re a program manager at a FinTech company.” | Shifts tone, framing, and emphasis; has little proven effect on factual accuracy |
Task, context, and format are easy to defend. Persona is where things get interesting.
Google lists persona as optional, but recent research suggests you should expect little from the familiar “You are an expert…” opening. A Wharton study tested expert personas across six language models, including two Gemini models, and found no consistent improvement in factual accuracy over prompts without a persona.
The one exception was Gemini 2.0 Flash, where expert personas produced a modest gain on one benchmark. Gemini 2.5 Flash showed none, so the researchers treat it as a model quirk. Personas are still useful when you want a particular point of view, vocabulary, or level of explanation. They just won’t make the answer more accurate.
Context has a bigger effect on the answer, especially when Gemini can read the source material itself. In supported Gemini and Workspace surfaces, you can type @ to point Gemini to a Drive file directly, so you don’t have to describe the file or paste parts of it into the prompt.
Take this prompt:
Summarize our Q3 performance.
With context and format added, it becomes:
Summarize @Q3 Revenue Review for the leadership team. Give me five bullets covering revenue, pipeline, churn, risks, and next steps.
The second prompt leaves Gemini less room to improvise. It knows what to read, who the answer is for, what to extract, and how to present it.
For a broader look at the same mechanics across other AI tools, our guide on how to write AI prompts covers prompting for ChatGPT, Claude, and more.
Prefer to see the framework in action? This walkthrough covers the same ideas, plus the common mistakes that weaken a prompt.
Summarize this article for me please30 Gemini Prompts for Work, Sorted by Role
Here we cover 30 copy-and-paste Gemini prompts across six job functions, five per role: marketing, project management, sales and client services, product and engineering, HR and people operations, and data and spreadsheets. Each one follows the task-context-format structure from the previous section, with notes on where Gemini needs a guardrail.
Pro Tip: Add one line to any prompt that reads a document: “If the answer isn’t in the file, say so.” It costs nine words and cuts down on confident guesses you’d otherwise have to catch in review.
Marketing prompts
Marketing prompts work best when Gemini starts from something real: a brief, an interview transcript, or a competitor’s page. Without that material, Gemini falls back on generic copy that could describe any product. Each prompt below either points Gemini to a specific source or sets a hard limit, like a character count or a cap on channels.
- Turn @[campaign brief] into a channel plan. Use one table with columns for channel, audience, core message, required asset, and owner. Limit to five channels.
- Rewrite the three paragraphs below for a [audience] reader who has 20 seconds. Keep every number. Cut the hedging. Return the rewrite only.
- Read @[customer interview transcript] and pull every sentence where the customer describes a problem in their own words. Group by theme. Quote verbatim.
- Draft six subject lines announcing [feature]. Three should lead with the outcome and three with the pain point. Keep each under 45 characters.
- Compare the messaging at [competitor URL] against @[our positioning doc]. Use a two-column table showing where the positioning overlaps and where it differs.
Once a prompt works, save it so you don’t have to rewrite the instructions next time. Our roundup of AI prompt workflow templates covers how to turn recurring prompts like these into a repeatable workflow.
Project management prompts
Project problems start out as gaps: a task with no owner, a date that ignores a dependency, or an assumption that turns out to be wrong. These prompts ask Gemini to find those gaps in your meeting notes, plans, and timelines. Two of them also produce something you can send or assign right away: a status update and a subtask list.
- From @[meeting notes], list every commitment someone made. Use a table with task, owner, and due date. Flag any item with no owner.
- Read @[project plan] and list the five assumptions that would cause the most damage if they turned out to be wrong. Give one early warning sign for each.
- Write a status update for [stakeholder] from @[project doc]. Use three bullets: what shipped, what slipped and why, and what you need from them.
- Break [deliverable] into subtasks that each take under a day. Use a numbered list with one acceptance criterion per item.
- Compare the dates in @[timeline sheet] against the dependencies in @[scope doc]. List every place where the two disagree.
Also Read: AI Agents for Project Management
Sales and client services prompts
Sales prompts produce text a buyer or client will read, so accuracy and tone count as much as speed. Two of the prompts below add a check: one keeps what the buyer said separate from what Gemini inferred, and the other states what’s out of scope.
- Summarize @[email thread] into a client-facing recap. Start with one paragraph of context, then bullets for decisions, open questions, and next steps with owners.
- Draft a reply to the objection below. Acknowledge it in one sentence, give two specific counterpoints, and end with a question. Keep it under 120 words.
- Read @[discovery call transcript] and list what the buyer said about budget, timeline, and approval. Mark anything you inferred separately from what the buyer stated outright.
- Turn @[statement of work] into a one-page scope summary for the client. Use plain language and call out anything explicitly out of scope.
- Draft a check-in email for an account that has been quiet for [X] weeks. Give them one specific reason to reply.
Pro Tip: For the objection reply, paste a real objection from a lost deal along with what your best rep said back. Gemini matches the register of what you show it far better than the register you describe.
Product and engineering prompts
Product and engineering teams spend a lot of time turning one document into another. A bug report becomes steps for the engineer who will fix it, or a PRD becomes stories for the team building it. These prompts make the process less painful.
- Read @[bug report thread] and write reproduction steps a new engineer could follow. Use a numbered list. Flag every step where the report is ambiguous.
- Turn @[PRD] into user stories using “As a [role], I want [action] so that [outcome].” Group them by epic.
- Review the spec below and list the edge cases it doesn’t address. Sort them by likelihood of reaching production.
- Summarize @[incident doc] into a postmortem draft covering timeline, impact in numbers, root cause, and three follow-up actions with owners.
- Explain what the function below does in plain English for a non-engineer. Then list what breaks if the input is empty.
Note: Gemini can run Python in a sandbox, but it can’t execute your codebase, install your dependencies, or see the systems the function touches. Treat the “what breaks” answer as a list of hypotheses to test, and ask it to state any assumptions about input types or dependencies it can’t see.
HR and people operations prompts
HR prompts deal in personal information, such as survey answers and performance feedback, and that has to remain private. The prompts below keep Gemini close to the source by asking for verbatim quotes, word limits, and observable behavior only. Read the data note after the list before you paste in survey exports or feedback drafts.
- Rewrite this job description for [role] to cut jargon. Keep the responsibilities specific. Preserve the structure but shorten the sentences.
- From @[survey results], list the three themes that appear most often in free-text answers. Quote two responses per theme verbatim.
- Draft a 30-day onboarding checklist for a [role] joining a [size] team. Group it by week. Give every item an owner.
- Turn @[policy doc] into an FAQ with eight questions an employee would reasonably ask. Keep each answer under 60 words.
- Review this feedback draft. Flag every sentence that describes a personality trait. Observable behavior only.
Note: Check where the data goes before you paste it. Qualifying Google Workspace accounts have stronger data protections than the consumer Gemini app, where chats may be reviewed and used to improve Google’s AI when Keep Activity is on. Strip names and other identifying details from survey exports and feedback drafts by default.
Data and spreadsheet prompts
In spreadsheets and reports, a wrong number is harder to spot than a wrong sentence. Each prompt below asks for output you can check: a percentage for each column, a formula explained part by part, a page number beside every figure, or a set threshold for flagging differences.
- Look at @[sheet] and tell me which columns contain missing values and what percentage of rows are affected in each.
- Write a Google Sheets formula that [describe the calculation]. Explain what each part does in one line.
- From @[sheet], build a table of the top 10 rows by [metric], with a column showing each row’s share of the total.
- Read @[report PDF] and pull every figure that has a date attached. Put the source page number beside each.
- Compare @[sheet A] against @[sheet B] and list every row where the numbers disagree by more than [X]%.
Summarize this article for me pleaseHow Should You Prompt Gemini for Image Generation?
The first thing to decide is what kind of image task you’re giving Gemini.
In its prompting guide for Nano Banana 2 and Nano Banana Pro (Gemini’s image models), Google separates image prompting into three jobs: creating a scene from scratch, editing an existing image, and working from reference images. Each one needs a different kind of instruction.
For a fresh generation, give Gemini enough detail to build the shot you have in mind. Start with what’s visible: the subject, what they’re doing, where they are, how the scene is framed, and the visual style.
Compare this prompt:
Make a professional headshot.
with this one:
A woman in a navy blazer seated beside a window, medium close-up, soft light from camera left, shallow depth of field, neutral office background.
The second prompt hands over far less control. You’ve already chosen the framing, lighting, setting, and distance from the subject. Gemini still has room to interpret the scene, but the big visual decisions are yours.
Use these rules to get the same control:
- Describe what should be in the frame. “Professional,” “cinematic,” and “premium” are weak on their own because they can mean almost anything. Objects, poses, backgrounds, light sources, and camera position give Gemini something concrete to render
- Give the camera a job. If the result still feels generic, add terms like medium close-up, low angle, overhead shot, shallow depth of field, or wide-angle. Camera language often fixes a composition faster than another pile of style adjectives
- Stop restarting once the composition works. If the face, pose, and framing are good, keep them and edit the weak part: “make the light warmer,” “remove the plant,” or “change the jacket to charcoal.” A fresh generation can solve one problem and bring three new ones with it
- Protect the parts you already like. “Change the background to a hotel lobby” leaves too much open. Add “keep the subject, pose, clothing, framing, and lighting unchanged,” and the edit has much tighter boundaries
- Give reference images narrow jobs. If you upload three references, say what each one contributes: pose from Image A, lighting from Image B, background from Image C. A vague “combine these” invites the model to blend everything in ways you never asked for
- Describe the scene you want. “An empty street at dawn” gives Gemini a picture to build. “No cars, no pedestrians, no signs” gives it a blacklist and leaves the scene itself vague
Three prompt shapes cover most of the work:
- Starting from scratch: “[subject] + [action] + [setting] + [composition] + [style]”
- Editing an existing image: “Change [specific element] to [new state]. Keep [protected elements] unchanged.”
- Using reference images: “Use [image] for [specific trait]. Apply it to [base image or subject]. Keep [protected elements] unchanged.”
For text work, Gemini usually has the source material in front of it and needs clear instructions. Image generation shifts more of that burden onto you. The less you leave undefined about the scene, the less time you spend fixing choices Gemini made on your behalf.
Summarize this article for me pleaseHow to Save a Gemini Prompt with Gems
If you rebuild the same prompt every week, save it as a Gem.
A custom Gem stores your instructions and can include reference files. Open it again later, and the setup will still be there. It works well for recurring jobs like reviewing copy against a style guide, turning meeting notes into a fixed status format, or drafting from the same positioning docs each time.
Here’s how to set one up:
- Open Gemini and select Gems in Settings
- Choose New Gem, give it a name, and write the instructions the way you’d write any prompt, covering the task, the context, and the output format
- Add reference files under Knowledge, either from your device or from Google Drive. Test it in the preview pane, then click Save
A few setup details are easy to miss:
- Previewing doesn’t save the Gem. Google’s current Gems documentation treats preview as a testing step. You still need to click Save when you’re done
- Drive files need the right account settings. Keep Activity must be enabled, and Google Workspace must be connected to Gemini Apps before you can add Drive files. On a work account, your admin may also need to enable app access
- Drive-linked files stay current. If your style guide or product brief changes, the Gem picks up the latest version without another upload. Files uploaded from your device are static copies
There’s one hard limit to plan around: a custom Gem can hold up to 10 files, each up to 100 MB.
That’s enough for a focused working set, such as a style guide, positioning doc, approved examples, and a template or two. It gets awkward once you try to turn one Gem into a full company knowledge base.
At that point, split the job. A “Blog Editor” Gem can hold editorial guidance and strong examples, while a separate “Product Messaging” Gem carries positioning and product docs. Smaller knowledge sets also make it easier to tell which source Gemini used (especially when the output goes wrong).
Remember: Even with a clean split, a Gem has a fixed view of the job. It knows only the files you gave it. You still have to add every new brief, transcript, or spreadsheet by hand before Gemini can use it.
Summarize this article for me pleaseWhere Gemini Prompts Break in Real Use
A clear prompt can still produce a bad result when the problem sits somewhere else: the wrong context, a product limit, or a task that asks Gemini to do too much at once.
These are the failure points that come up most often:
| What breaks | What happens in practice | What to do instead |
|---|---|---|
| Context window | Accounts without a Google AI plan work with a 32,000-token window, so a long PDF or transcript can get cut off with no warning | Split long files, or ask Gemini to confirm the last section it read |
| File uploads | 10 files per prompt, and every file inside an uploaded folder counts toward the cap | Merge small files before uploading, or work in batches |
| Usage allowance | Gemini Apps now use compute-based limits that refresh every five hours, up to a weekly cap. Deep Research, image generation, and Pro models burn through it faster, and free accounts can lose Deep Research during high demand | Save Deep Research for questions that need many sources, and watch for the in-app warning that you’re close to a limit |
| Audio Overviews in Drive | Capped at 20 a day, and closing or refreshing the page mid-generation cancels it | Start them early and leave the tab open |
| Personas | The Wharton study from above found no consistent accuracy gain from expert role prompts across six models | Keep the role to one line tied to the job: “review this as a compliance lead” |
| Overloaded Gems | 10 files may fit one workflow, but unrelated material makes it harder to tell which source shaped the answer | One Gem per job |
Most of these limits shift by plan, and Google notes they can change without notice during periods of high demand.
Before you rewrite a failed prompt, check three things: whether Gemini has the right source material, whether you’ve hit a usage or file limit, and whether the task needs to be split into smaller steps. If one of those is broken, better wording won’t fix it.
Also Read: Claude Skills for Writing
Summarize this article for me pleaseFive Gemini Prompt Mistakes, and What to Do Instead
Each mistake below has a symptom you can spot in the output. Once you recognize the symptom, you can usually fix the next prompt without rewriting everything from scratch.
1. Stacking several jobs into one prompt
Symptom: You asked for four things. The first comes back detailed, the fourth gets two sentences.
Fix: Split work that requires different kinds of reasoning. Ask Gemini to analyze the material first, then draft, compare, or recommend in the next turn. You keep the same conversation context without asking one response to carry the whole workflow.
2. Letting Gemini fill in missing facts
Symptom: The answer sounds complete, but some of the specifics came from the model, and your source material never mentioned them.
Fix: Tell Gemini how to handle gaps. “Use only @[project plan]. If the document doesn’t say, mark it unknown” is much safer than leaving it to complete the picture.
This is especially useful for meeting notes, research, policy documents, and client work, where a plausible invention can be harder to catch than an obvious error.
3. Correcting the answer without diagnosing it
Symptom: “Try again” gives you a different version of the same problem.
Fix: Name what failed. “Keep the structure, but remove claims that aren’t supported by the source” gives Gemini a correction it can act on. The same goes for tone, length, missing fields, weak examples, or too much detail.
4. Starting a new chat when the old one already has the context
Symptom: You’re pasting the same brief, audience, and background material several times in one session.
Fix: Continue in the existing thread when the next request depends on the same material. Gemini can use the conversation context, within the limits of its context window, which makes small revisions much cheaper than rebuilding the setup each time.
Start a new chat when the thread gets cluttered or the job changes.
5. Leaving good output in the chat window
Symptom: You got the answer you needed, but two days later, it’s buried inside an old conversation.
Fix: Move finished work to where it belongs: the project doc, task, spreadsheet, comment, or reusable Gem. A strong answer that stays in the chat has to be found, copied, and rebuilt the next time the job comes around.
Summarize this article for me pleaseHow to Keep Gemini Prompts Next to the Work in ClickUp
A Gem holds up to 10 files, and you have to add every new brief, transcript, or spreadsheet yourself. Mistake #5 happens for a similar reason: good output stays in a chat window, away from the task it was for. Saved prompts are easiest to reuse when they sit next to the work they’re for.
ClickUp keeps your prompts in the same place as your tasks, Docs, and projects. A reusable prompt can live in a Doc or on a recurring task. ClickUp Brain reads the task or Doc you point it to, so the current project supplies the context, and you don’t have to re-upload files each week.
Here’s a simple setup for managing the prompts using ClickUp:
- Keep reusable prompts in a ClickUp Doc. Group them by job, such as campaign briefs, meeting follow-ups, research, or editing. ClickUp Docs can sit alongside the projects that use them, which beats maintaining a prompt graveyard in another tab
- Put task-specific prompts on the task itself. A recurring task can carry the prompt in its description or comments, right next to the brief, files, owner, and due date
- Pull the work into the prompt with a
@mention. Inside Brain,@mention points to tasks, Docs, people, Teams, or Channels. A reusable prompt can stay short while the current project supplies the changing context - Save prompts you run often. Type
/in Brain to add a new prompt and choose who can see it: just you, everyone, admins, or specific people. Saved prompts show up in the same menu across every project, so you don’t have to search through old chats - Turn a prompt into a Skill when it should run itself. An AI Skill is a saved set of instructions that ClickUp Brain applies automatically whenever the request fits
- Move the answer straight into the workflow. Brain responses can become Docs, tasks, subtasks, comments, and other ClickUp items. That closes the gap from mistake #5, where a good AI answer ends up stranded in chat history
Say you keep this prompt for weekly project updates: “Read @[project task] and @[meeting notes]. Draft three bullets covering what shipped, what slipped and why, and what needs a decision this week.”
The wording barely changes from week to week. The task and meeting notes do.
And if you want Gemini for the generation itself, there’s no need to leave ClickUp. Brain’s model selector includes Gemini models, and the Workspace context stays attached when you switch mid-conversation.
Best for: Teams that reuse prompts around recurring work and want the prompt, source material, and final output in the same workspace.
Where it taps out: If your workflow already lives entirely inside Gemini and Google Workspace, moving prompts into ClickUp can add an extra layer you don’t need. It’s strongest when ClickUp is already where you plan and ship work.
Summarize this article for me pleaseKeep the Prompts That Earn Their Place
You don’t need a library of 200 prompts. Start with the prompts in this guide that match the work you do every week.
Use the examples here as working drafts. Edit them for your team, and save the ones you keep using as a Gem or in ClickUp. If a prompt stops working, check the source material and limits before you rewrite it.
The rest of the improvements come through trial and error.
And when your prompts need to become a part of how the work gets done, try ClickUp.
Summarize this article for me pleaseFrequently Asked Questions About Gemini Prompts
Why does the same Gemini prompt give a different answer each time?
Two things are usually at play. Language models add some randomness to every response, so identical prompts rarely produce identical text. On top of that, Gemini in Workspace apps searches Drive, Gmail, Chat, and Calendar, so the material it pulls in can change between runs. Add the files you want as sources, and the answers get far more consistent.
Does Gemini remember my earlier chats when I write a new prompt?
For eligible accounts, yes. Personal context is on by default, so Gemini can draw on past conversations to shape a new answer. You can turn it off in Settings, or use a Temporary Chat for a one-off request that stays out of your history and doesn’t affect later responses.
How do I stop Gemini from pulling in files I never asked about?
Manage your sources. In Docs, Drive, and Gmail, the Ask Gemini panel has a source settings menu where you can switch off Drive, Gmail, Chat, or Calendar search. Add the specific files you want with @ or Add from Drive, and Gemini will prioritize them. Below every answer, click Sources to see which files it used.
Can I share a Gem with my team?
Yes. Sharing works like sharing a Google Doc: you can make someone a Viewer or an Editor, and if the Gem includes uploaded files, you’ll be asked to share those too. Shared Gems are stored in Drive, so your organization’s Drive sharing rules apply. Admins can also turn Gem sharing off for the whole organization.
Is Gemini free to use?
Yes, Gemini has a free tier that covers chat, file uploads, and custom Gems. Paid plans raise usage limits and add access to more capable models. Gemini features inside Google Workspace apps depend on your organization’s plan and admin settings.
Everything you need to stay organized and get work done.
Recent Articles
30 Gemini Prompts for Work (And the Google Tip to Skip) 
30 Gemini Prompts for Work (And the Google Tip to Skip)
Claude Skills for Writing: How to Build One That Actually Fires 
Claude Skills for Writing: How to Build One That Actually Fires
