When is Gemini genuinely the best choice over ChatGPT or Claude?

Three situations: (1) the input is multimodal — images, video, mixed PDFs — and you want one model to reason over all of it; (2) the input is very large (hundreds of pages, multiple documents, a full transcript corpus) and you need a model with the context window to handle it natively; (3) the work lives inside Google Workspace or needs current Google Search grounding, where Gemini's native integration removes friction other models can't match.

Which Gemini model should I default to — 2.5 Pro, 2.5 Flash, or 1.5?

2.5 Flash is the right default for ~80% of work — daily writing, summarisation, structured extraction, fast multimodal. Reach for 2.5 Pro on hard reasoning, multi-document synthesis, or anything that has to be right (financial analysis, complex code, careful research). Use 1.5 Pro when you need 1–2M-token context windows and cost matters more than peak quality. Use 1.5 Flash for high-volume, latency-sensitive automations.

How do I stop Gemini from making things up on long documents?

Three moves, in order: (1) put the document at the top of the prompt and your instructions at the bottom; (2) add an explicit grounding instruction ("answer only from the attached document; do not use outside knowledge"); (3) require citations ("cite the page number for every claim; if the document doesn't cover it, say so explicitly"). Together they cut hallucination dramatically.

What's the difference between Gemini in the app, Google AI Studio, and the Gemini API?

The Gemini app is the consumer interface — best for quick tasks, Workspace integration and chat. Google AI Studio is the developer playground — best for prototyping prompts, testing structured output and previewing model behaviour before shipping. The Gemini API is the production endpoint — used for structured output, function calling and embedding Gemini in your own product. The prompts in this guide work in all three; the API benefits most from explicit JSON schemas and grounding flags.

Can Gemini really read entire YouTube videos?

Yes — on 2.5 Pro and 1.5 Pro you can pass YouTube URLs (or direct video files) and ask Gemini to summarise, extract timestamps, transcribe specific sections, or answer questions about what was said and shown. Ask for timestamp citations explicitly ("for every claim, cite the [mm:ss] timestamp") to keep the output verifiable.

How reliable is Gemini's structured JSON output?

Very, when you use it properly. Request a tight schema, set null rules for missing fields, forbid preamble and markdown fences, and prefer 2.5 Pro for complex schemas (2.5 Flash for simple ones). On the API, use response_mime_type: "application/json" with a response schema — adherence jumps to near-perfect compared to free-form prompting.

Does Gemini work well for coding compared to GPT-5 and Claude?

Yes — Gemini 2.5 Pro is strong on multi-file reasoning, large diffs, and tasks where understanding a screenshot or diagram matters. For raw boilerplate generation and quick fixes, GPT-5 is often faster; for the most careful PR-quality reviews, Claude is still the day-to-day pick for many teams. Gemini's edge in code is long-context plus multimodal — feed it a screenshot of a UI bug or a diagram of your architecture and the answers get noticeably sharper.

Can I port my ChatGPT or Claude prompts to Gemini?

Most will work, but you'll leave quality on the table. To port: put any artifact (document, image, code) at the top of the prompt, add an explicit grounding instruction if accuracy matters, move the most important constraints to the end, and — for API work — convert any structured output instructions into a real JSON schema. Five minutes of porting usually lifts Gemini output meaningfully.

Guide · Gemini

The Best Gemini Prompts

A curated, copy-ready library of the highest-leverage Google Gemini prompts — built around Gemini's strengths in multimodal reasoning, massive context and live Google data.

Updated June 26, 202614 min readBy Prompt InFlow Team

PromptInFlow AI Guides

General AI Models

Image & Video Models

AI Task Guides

AI Writing
Read guide
AI Marketing
Read guide
AI Coding
Coming soon
AI Business
Coming soon
AI Education
Coming soon

Gemini is the model most people underuse. It's a Google-native, multimodal-first reasoning system with a context window large enough to swallow entire books, slide decks, codebases and video files in a single turn — and it's tightly wired into Google Search, Workspace and the Android stack. Used well, it stops being "another chatbot" and starts being the model you reach for whenever the input is messy, visual, very long, or needs to be grounded in something verifiable. The prompts in this guide are written specifically for how Gemini behaves: they front-load the artifact, ask for explicit grounding, and lean on Gemini's multimodal and structured-output strengths instead of treating it as a text-only competitor to ChatGPT.

This is a working reference, not a roundup. Every prompt has been tested across the current Gemini lineup — Gemini 2.5 Pro for deep reasoning, 2.5 Flash for fast everyday work, and 1.5 Pro / Flash for cost-sensitive workloads — inside both the Gemini app and Google AI Studio. You'll find prompts for research and synthesis, document and PDF analysis, image and video understanding, long-form writing, data work in Sheets, coding, Workspace automation (Docs, Gmail, Slides) and structured-output extraction. Each one is built around the four moves that consistently lift Gemini quality: ground the model, give it the artifact, ask it to plan, and constrain the output.

Who is this guide for? Researchers and analysts who need to chew through long PDFs, transcripts, datasets or YouTube videos and get back synthesis they can cite. Marketers and writers who want on-brief drafts and Workspace-ready outputs. Developers shipping with Gemini's API, function calling and structured JSON. Operators inside Google Workspace who want Docs, Sheets, Gmail and Slides on autopilot. If you've used Gemini and thought "this should be more useful than it is," the gap is almost always the prompt — and this guide closes it.

To get the best from Gemini, treat each prompt as a brief paired with an artifact. Attach the PDF, paste the transcript, drop the image, link the URL — Gemini is at its best when it has something concrete to reason over. Ask Gemini to ground its claims in the artifact or in current Google Search results, and to flag anything it can't verify. For structured work, request explicit JSON schemas or markdown tables; for long contexts, put the artifact first and your instructions last. Bookmark this page — we refresh it whenever Google ships a new Gemini version or a meaningful capability change.

Editorial standards

How we selected these prompts

Why each prompt earns its place in the library.

Every prompt in this guide earns its place. We test on Gemini's current production lineup (2.5 Pro and Flash, with fallbacks on 1.5 Pro and Flash) inside the Gemini app, Google AI Studio and the Gemini API, and only ship prompts that produce a usable first output without a chain of follow-ups. Prompts come from our own daily workflows, patterns Google publishes in its Gemini and AI Studio documentation, and contributions from the Prompt InFlow community.

Built for Gemini's strengths

Every prompt leans into what Gemini does best — multimodal input, very long context, grounded answers and structured output — instead of treating Gemini as a text-only stand-in for ChatGPT.

Grounded by default

Where accuracy matters, prompts explicitly ask Gemini to cite the attached artifact or current search results, and to flag claims it can't verify. Hallucination shows up less when you ask.

Adaptable placeholders

Every template uses clear {placeholders} so the same prompt works for a solo operator, a Workspace team or a product team shipping with the Gemini API — swap five words, keep the structure.

Practical, not novel

We publish prompts you'll reuse weekly: research synthesis, document Q&A, on-brand drafts, Workspace automations, structured extraction. Demo-only prompts that look impressive once and never get used twice don't make it in.

Who this guide is for

This guide is built for anyone who wants Gemini to do real, grounded work across text, images, video, data and Google Workspace.

Researchers & analysts
Pulling synthesis from long PDFs, transcripts, datasets and YouTube videos — and needing answers they can cite.
Knowledge workers in Google Workspace
Drafting Docs, summarising Gmail threads, building Slides outlines and pushing analysis through Sheets.
Marketers & writers
Briefing on-brand drafts, repurposing across channels, and using Gemini's image understanding for creative review.
Developers
Shipping with the Gemini API: structured JSON output, function calling, code reasoning and multimodal endpoints.
Educators & students
Turning textbooks, lecture videos and slide decks into explainers, study plans and Socratic question chains.
Operators & founders
Competitive teardowns from public Google data, structured strategy memos and lightweight Workspace automations.

In this guide

What you'll learn

Skim the highlights, then jump to the section you need.

When Gemini is the right model

Where Gemini wins (multimodal input, very long context, Google grounding, Workspace integration) and where it doesn't — so you pick the right tool for the job.

40+ copy-ready Gemini prompts

Research, document analysis, multimodal, writing, coding and Workspace prompts you can paste into the Gemini app or AI Studio and run today.

Gemini-native techniques

Grounding instructions, artifact-first prompting, structured JSON output, function-calling-friendly briefs and the "plan, then answer" pattern.

How to use the full context window

Patterns for putting artifacts at the top, instructions at the bottom, and asking Gemini to cite the page, timestamp or row it relied on.

Framework

How great Gemini prompts work

The six-part anatomy behind every prompt that consistently ships usable output.

Gemini behaves like a research analyst with a search engine, a multimodal eye, and a very long memory — but only if you brief it like one. The prompts that consistently outperform on Gemini share the same underlying shape, which we map onto a six-part framework: R-A-C-G-O-V (Role, Artifact, Context, Grounding, Output, Verification). Use all six on serious analysis; use the first four on everyday work.

Role

Open with a specific role. "You are a senior equity research analyst" or "You are a Workspace-savvy executive assistant" sets Gemini's vocabulary, depth and assumed audience before it sees the task — and Gemini calibrates depth from role faster than most models.

Artifact

Give Gemini something concrete to reason over — a PDF, image, video, URL, transcript, spreadsheet or pasted block. Gemini is multimodal-first and at its strongest when there's an attached or pasted artifact, not just a question hanging in the air.

Context

State who the audience is, what's been tried, what success looks like, and what's off-limits. Two sentences of real context beats ten adjectives, and Gemini uses context to decide which parts of the artifact to weight.

Grounding

Tell Gemini where its answer should come from: "only from the attached PDF," "using current Google Search results," "only from the rows in the sheet." Grounding instructions cut hallucination noticeably — Gemini honours them more reliably when they're explicit.

Output format

Spell out the shape: a markdown table with these columns, JSON matching this schema, a memo with H2s, a Slides outline with speaker notes. For API work, request a JSON schema directly — Gemini's structured output mode is fast and reliable when you ask for it.

Verification

Ask Gemini to cite the page, timestamp, row or URL it relied on, and to flag any claim it can't verify. On long-context and search-grounded work this is the single biggest lever on trust — Gemini will hedge honestly when you invite it.

Three examples — and why they work

Example 1

Multi-PDF research synthesis

You are a senior research analyst. I'm attaching three PDFs on {topic}. Your job is to produce a synthesis a non-expert decision-maker can act on.

Instructions:
1. Read all three PDFs in full before answering.
2. For each major claim in your synthesis, cite the source document and page number in parentheses.
3. Where the documents disagree, surface the disagreement and which source is more credible and why.
4. If a question I'm likely to ask isn't covered by the documents, list it under "Open questions."

Output: a markdown memo with four H2 sections — Bottom line, Key findings (with citations), Disagreements, Open questions. Keep the memo under 600 words.

Why it works: Artifact-first (the PDFs do the heavy lifting), grounding is explicit (cite source and page), and the output format forces decision-ready structure. The "Open questions" section gives Gemini permission to admit gaps instead of inventing answers.

Example 2

Image-grounded creative review

You are a senior creative director reviewing concept art. I'm attaching one image.

Review the image against these criteria, in this priority order:
1. Brand fit for {brand description}
2. Compositional clarity (focal point, balance, negative space)
3. Technical execution (lighting, colour, anatomy / proportions if applicable)
4. Production readiness for {channel: e.g. Instagram square ad}

For every observation, reference what in the image you're pointing to ("the figure in the foreground left," "the warm highlight along the top edge"). End with a single "Ship / Iterate / Reject" verdict and the one change that would most improve the image.

Output: a bulleted list grouped by the four criteria, then the verdict line.

Why it works: Uses Gemini's native image understanding (artifact = image). Prioritised criteria stop Gemini from drifting into generic praise. The "reference what in the image" instruction forces grounded, specific feedback instead of vague aesthetic notes.

Example 3

Structured JSON extraction for the API

You are a structured-data extraction service. Read the email thread below and return a single JSON object matching this schema exactly — no preamble, no markdown fences, no commentary.

Schema:
{
  "deal_stage": "discovery | evaluation | negotiation | closed_won | closed_lost",
  "next_step": string,
  "owner_action_required": boolean,
  "open_questions": string[],
  "risks": string[],
  "confidence": "low | medium | high"
}

Rules:
- If a field is not determinable from the thread, return null for that field and add a short explanation to open_questions.
- Set confidence based on how directly the thread supports your inferences.

<thread>
{paste email thread}
</thread>

Why it works: Plays directly to Gemini's structured-output strength: a tight schema, explicit nullability rules, and a confidence field that pairs naturally with downstream automation. The "no preamble, no markdown fences" instruction makes the output API-safe on the first call.

Browse by use case

Featured prompt categories

Jump into the library that matches what you're working on — each category includes guidance on when to use it and what good output looks like.

Writing

When to use: Reach for Gemini when the writing job starts from real source material — a brief, a transcript, a spec, a slide deck, a research dump — and you want a draft that stays faithful to the inputs.
What you can achieve: On-brief drafts grounded in attached artifacts, Workspace-ready Docs you can drop straight into a shared folder, fast multi-format repurposing (Doc → email → Slide outline), and Gmail-native replies that match an existing thread's tone.
Best practices: Attach or paste the source artifact first; Gemini drifts less when it's writing from something concrete than from a topic.
Specify the exit channel (Doc, Gmail reply, Slides outline, LinkedIn post) — Gemini calibrates length and structure to the destination.
Ask Gemini to flag anything it added that isn't in the source — it's a fast way to catch helpful hallucinations before they ship.

Marketing

When to use: Use Gemini for marketing work that benefits from grounding in real data — competitive teardowns from public Google sources, SEO briefs informed by current SERPs, repurposing campaigns across channels, and creative review of attached imagery.
What you can achieve: Competitor breakdowns sourced from current Google results, SEO outlines that match real search intent, multi-channel repurposing of a single brief, and structured creative reviews of attached ads or concept art.
Best practices: Ask Gemini to ground claims in current Google Search results when freshness matters (pricing, positioning, launches).
Attach the brief or current asset as an artifact instead of describing it — Gemini's multimodal review is sharper than text-only critique.
Request 3 variants max per request; quality drops sharply past three, the same as on other models.

Coding

When to use: Use Gemini for code tasks that benefit from long context (multi-file reasoning, large diffs), structured output (JSON, schemas, function calling), or multimodal input (screenshots of UI bugs, architecture diagrams, error screens).
What you can achieve: Multi-file refactors that respect the full repo, function-calling-ready JSON specs, structured bug reports from screenshots, and architecture sketches that hold up across services.
Best practices: Paste the smallest reproducing snippet plus the surrounding files Gemini needs — large context is fine, but signal beats volume.
State runtime, language version and constraints up front ("Node 20, TypeScript strict, no new dependencies").
When you need machine-readable output, request a JSON schema directly — Gemini's structured-output mode is built for this.

Business

When to use: Use Gemini for strategy work where the value is in synthesis — pulling a single answer out of many sources — and for Workspace-bound deliverables: memos, board updates, hiring scorecards, SOPs that live in Docs and Sheets.
What you can achieve: Decision-ready memos sourced from multiple attached artifacts, competitor teardowns grounded in current Google data, sanity-checked financial commentary over pasted tables, and SOPs your team can actually follow.
Best practices: Give Gemini your actual numbers, transcripts and decks — synthesis from real artifacts is where Gemini outperforms.
Ask for the counter-argument: "steelman the opposing view, then recommend" surfaces blind spots faster than asking for risks.
Request decision-ready output (Recommendation → Rationale → Trade-offs → Open questions) instead of exhaustive analysis.

Writing

Articles, emails, scripts and stories that read like a human wrote them.

Marketing

Ads, landing pages, SEO and campaigns that actually convert.

Coding

Debugging, refactors, code review and architecture prompts.

Business

Strategy, ops, sales decks and financial analysis prompts.

Hand-picked

Curated prompts

The prompts we reach for most often — copy, paste, adapt.

Marketing

ChatGPT

High-Converting Cold Sales Email

Write a 4-line cold outreach email with a personalized hook, value prop, social proof and soft CTA.

You are a B2B sales copywriter. Write a 4-sentence cold email to {persona} at {company}. Open with a specific observation about {trigger_event}, then connect it to {value_prop}. Include one short social proof line and a low-friction CTA.

View prompt →

Marketing

ChatGPT

SEO Topic Cluster Builder

Generate a pillar page plus 12 supporting article ideas with target keywords and search intent.

Act as an SEO strategist. For the seed topic {topic}, produce one pillar page idea and 12 supporting articles. For each: working title, primary keyword, search intent, and one internal link target.

View prompt →

Coding

Claude

Senior Engineer Code Review

Get a staff-engineer-grade review of a diff with risk callouts, simpler alternatives and test gaps.

You are a staff engineer. Review this diff and return: (1) correctness risks, (2) simpler alternatives, (3) missing tests, (4) one nit. Be specific. Diff: {diff}

View prompt →

At a glance

Quick reference tables

Cheat-sheets for picking the right model and the right prompt style.

Gemini models at a glance

Pick the tier that matches the job — Gemini's tiers differ noticeably on cost and latency.

Model	Best for	Watch out for
Gemini 2.5 Pro	Deep reasoning, multi-document research, complex multimodal analysis	Higher latency and cost — overkill for short, routine tasks
Gemini 2.5 Flash	Daily writing, summarisation, structured extraction, fast multimodal	Slightly less rigorous on long-chain reasoning than 2.5 Pro
Gemini 1.5 Pro	Very long context (up to 1–2M tokens) when cost matters more than peak quality	Older; 2.5 Pro now beats it on most benchmarks
Gemini 1.5 Flash	High-volume, latency-sensitive tasks (classification, extraction, tagging)	Weaker at nuanced writing and multi-step reasoning

Gemini vs ChatGPT vs Claude — pick the right tool

All three are excellent. They're best at different jobs.

Task	Pick Gemini when	Pick ChatGPT when	Pick Claude when
Multi-document / long PDFs	You have 5+ PDFs or a 500-page document and want citations	Document is short and you want a fast, polished summary	Document is long and you want careful, hedged analysis
Multimodal (image / video)	Image AND video understanding, or live URL / YouTube input	Quick image Q&A or voice conversations	Image analysis with careful captioning, no video
Workspace integration	Output goes into Docs, Sheets, Gmail or Slides	Output is standalone copy or chat	Output is a long memo or analysis document
Structured JSON / function calling	You need reliable schema-conformant JSON via API	You need quick prototypes and conversational tool use	You need careful, schema-respecting output with reasoning
Live web / search grounding	Freshness matters (pricing, news, current SERPs)	You have built-in browsing turned on and want fast retrieval	Grounding is not native; pair with your own retrieval

Gemini-specific prompting techniques

Patterns that lift quality on Gemini specifically.

Technique	When to use	Example trigger
Artifact-first prompting	Any task with a document, image, video or URL	Attach the file, then write the brief beneath it
Explicit grounding	Anywhere hallucination is costly	"Answer only from the attached PDF; cite page numbers."
Structured output	API integrations, automations, data extraction	"Return a JSON object matching this schema, no preamble."
Plan, then answer	Reasoning, analysis, multi-step synthesis	"Plan your approach in 3–5 bullets, then write the final answer."
Search grounding	Freshness-sensitive prompts (pricing, news, launches)	"Use current Google Search results; cite the source URL."

Method

How to write better prompts

Four habits that lift the quality of every prompt you write.

Lead with the artifact

Whenever you have a PDF, image, video, URL or pasted block, put it at the top of the prompt — then write the brief beneath it. Gemini is multimodal-first; an artifact-first prompt consistently outperforms a topic-first one.

Tell Gemini where the answer should come from

Explicit grounding ("only from the attached PDF," "using current Google Search results," "only from rows in the sheet") cuts hallucination noticeably. Gemini honours grounding instructions more reliably when they're stated up front.

Ask for structured output when the consumer is a machine

If the output is going into a script, Sheet or downstream tool, ask for JSON matching an explicit schema — and tell Gemini to return no preamble and no markdown fences. Structured output mode is fast and reliable.

Plan before you answer on reasoning tasks

For anything analytical, instruct Gemini to plan its approach in a short bullet list before writing the final answer. It's a small move that meaningfully lifts quality on synthesis and code reasoning.

Avoid

Common mistakes

The patterns that quietly tank output quality.

Treating Gemini as text-only and skipping the artifact — its biggest edge is multimodal grounding.
Asking freshness-sensitive questions without telling Gemini to use current Google Search results.
Pasting a long document and putting your instructions at the top — Gemini weights end-of-prompt instructions higher.
Forgetting to ask for citations on long-context or grounded tasks; the model will hedge less than you'd like by default.
Requesting unstructured prose for downstream automation instead of asking for a strict JSON schema.
Defaulting to 2.5 Pro for everything — 2.5 Flash is faster and cheaper for ~80% of real tasks.

FAQ

Frequently asked questions

Quick answers about this guide and the prompts in it.

Keep reading

Related guides

Guide

The Best ChatGPT Prompts

A curated, copy-ready library of the highest-leverage ChatGPT prompts — organized by use case and tested across GPT-4o and GPT-5.

Guide

The Best Claude Prompts

A curated, copy-ready library of the highest-leverage Claude prompts — engineered for Claude's strengths in long-context reasoning, careful writing and structured analysis.

Guide

The Best AI Writing Prompts

Frameworks, model picks and copy-ready prompts for every writing task — from blog posts and emails to SEO content, product copy and storytelling.

Guide

The Best AI Marketing Prompts

A curated, model-aware library of marketing prompts that ship — built for the channels and tasks marketers actually run every week.

Explore

Related categories

Writing

Articles, emails, scripts and stories that read like a human wrote them.

Marketing

Ads, landing pages, SEO and campaigns that actually convert.

Coding

Debugging, refactors, code review and architecture prompts.

Business

Strategy, ops, sales decks and financial analysis prompts.

Compatible models

Works well with

Gemini

Multimodal reasoning

ChatGPT

General-purpose reasoning

Claude

Long-form writing & analysis

Fresh

Latest prompts

Recently added to the Prompt InFlow library.

Marketing

ChatGPT

High-Converting Cold Sales Email

Write a 4-line cold outreach email with a personalized hook, value prop, social proof and soft CTA.

View prompt →

Marketing

ChatGPT

SEO Topic Cluster Builder

Generate a pillar page plus 12 supporting article ideas with target keywords and search intent.

Act as an SEO strategist. For the seed topic {topic}, produce one pillar page idea and 12 supporting articles. For each: working title, primary keyword, search intent, and one internal link target.

View prompt →

Coding

Claude

Senior Engineer Code Review

Get a staff-engineer-grade review of a diff with risk callouts, simpler alternatives and test gaps.

You are a staff engineer. Review this diff and return: (1) correctness risks, (2) simpler alternatives, (3) missing tests, (4) one nit. Be specific. Diff: {diff}

View prompt →

Video AI

Veo

Veo Character Consistency Sheet

Keep the same character across multiple Veo shots with a reusable description block and shot list.

Character: 32-year-old woman, shoulder-length auburn hair, freckles, olive linen jacket. Reuse this exact description for every shot. Shot 1: medium close-up, golden hour. Shot 2: walking shot, overcast street. Shot 3: cafe interior, warm tungsten.

View prompt →

In summary

Conclusion

The ideas to carry forward — and where to go next.

If there's one idea to carry out of this guide, it's that Gemini is at its best when it has something concrete to reason over — a document, an image, a video, a URL, a dataset. The prompts that consistently produce shippable Gemini output give the model a clear role, an attached artifact, the context it needs, an explicit grounding instruction, a defined output format, and permission to flag what it can't verify. Skip the artifact or the grounding, and Gemini starts averaging the internet instead of analysing the thing in front of it.

Treat every prompt in this library as a starting point. Copy it, attach your real artifact, swap the {placeholders} for something specific, run it on 2.5 Flash first for speed, and only escalate to 2.5 Pro when the answer needs more rigour. Iterate with short follow-ups ("tighten section two," "add timestamp citations," "return as JSON") instead of rewriting from scratch — Gemini is unusually responsive to small corrections, especially when the artifact stays in context.

Prompting is a skill, and it compounds across models. The more prompts you ship on Gemini, the faster you'll feel when to reach for it — long context, multimodal input, Workspace deliverables, grounded research — and when ChatGPT or Claude is the better tool. When you're ready for more, explore our other model guides, browse the prompt library by use case, or open a generator to build a Gemini-ready prompt in seconds.

Continue Learning

Prompt Engineering Learning Path

A guided reading order through every Prompt InFlow flagship guide — start with the model you use most, then layer in topical skills.

Build prompts faster with Prompt InFlow

Browse 500+ tested prompts across every major AI model — or use our generators to build your own in seconds.

Browse all prompts Open ChatGPT generator