🔵 Google Gemini Guide

Gemini System Instructions:
Examples That Actually Work.

System instructions are how you make Gemini know who you are, what you do, and how to respond. Most people leave this blank and wonder why Gemini gives generic answers. Here's how to fix that.

Read the Guide → Context Engineering Guide
In this guide
Foundations

What Are Gemini System Instructions?

Gemini system instructions are persistent context that gets prepended to every conversation — before your first message. They tell Gemini who you are, what your role is, what format you prefer, and how to approach your work. Once set, they apply to the entire conversation without you having to repeat yourself.

They're the Gemini equivalent of ChatGPT's system prompt, the Claude system prompt in the API, or the CLAUDE.md file for Claude Code. Same concept, different implementation.

Why most Gemini answers feel generic

Gemini is trained to be helpful to everyone, which means it defaults to the most general possible answer. System instructions break that default. When you tell Gemini "I'm a fintech product manager at a Series B startup, we use React + Python, our customers are SMBs" — every answer it gives gets filtered through that lens. The difference in answer quality is dramatic.

System instructions are different from regular messages in one important way: they're maintained throughout the entire conversation, even as the context grows. Your role and preferences don't get lost as you add more messages — Gemini always has that foundation.

Setup

Where to Set Gemini System Instructions

There are four ways to set Gemini system instructions, depending on your use case:

🧪 Google AI Studio

Free interface at aistudio.google.com. Has a dedicated "System instructions" text field at the top of every new project. Best for testing and iteration.

🔌 Gemini API

Pass system_instruction as a parameter when creating your model instance. Supported in Python SDK, Node.js SDK, and REST.

⭐ Gemini Advanced

Gemini Advanced (Google One AI Premium) has a "Gems" feature — customized Gemini versions with system instructions built in. Access via gemini.google.com.

🏢 Google Workspace

Gemini for Google Workspace lets admins set organization-wide instructions, plus individual users can add personal context in Gemini settings.

Setting via API (Python)

Python — google-generativeai SDK
import google.generativeai as genai genai.configure(api_key="YOUR_API_KEY") model = genai.GenerativeModel( model_name='gemini-2.0-flash', system_instruction="""You are a senior fintech product manager. Context: - Company: Series B startup (~80 employees) - Stack: React frontend, Python/FastAPI backend, PostgreSQL - Customers: SMBs (50-500 employees), primarily in the US Your role: - Help me write PRDs, user stories, and technical specs - Default to bullet points and tables, not dense prose - When I ask about competitors, include data where available - Flag when my assumptions might be wrong Format preferences: - PRDs: standard PM format (problem, users, goals, solution, metrics) - User stories: "As a [user], I want [goal] so that [outcome]" - Always include acceptance criteria""" ) chat = model.start_chat() response = chat.send_message("Help me write a PRD for our new CSV import feature")

Setting via Google AI Studio

In AI Studio, click "New project" and you'll see a "System instructions" panel on the left side. Paste your instructions there. They persist for the entire project and apply to all messages. You can also enable them in the Gemini API column when you export.

AI Studio tip

AI Studio is free and the best place to test your system instructions before committing them to a production API call. The "Run settings" panel shows token usage — you can see exactly how many tokens your instructions consume before they hit your context window budget.

Comparison

Gemini System Instructions vs CLAUDE.md vs ChatGPT System Prompts

Each major AI platform has its own version of "tell the AI who you are." They work similarly but have meaningful differences in scope, persistence, and access method.

Gemini

System Instructions

Set per-model-instance via API or per-project in AI Studio. Applies to the entire conversation. In Gemini Advanced, "Gems" are shareable personas with instructions baked in.

ChatGPT

System Prompt

Set as the first message in the API "system" role. In ChatGPT web app, "Custom instructions" in settings apply globally. GPTs allow per-assistant instructions that persist for all users.

Claude Code

CLAUDE.md

A markdown file in your project root. Loaded automatically at session start. Much richer — can be thousands of words. Designed for deep project context, not just persona setting.

Feature Gemini System Instructions ChatGPT System Prompt CLAUDE.md
Scope Per conversation / per model instance Per conversation / per GPT Per project (file in codebase)
Max length ~8,000 tokens practical limit ~2,000 tokens practical Unlimited (auto-loaded into context)
Version control API param (can be committed) Not natively Yes — it's a file in your repo
Team sharing Via Gems / API param Via GPT sharing Committed to shared repo
Format Plain text / markdown Plain text / markdown Full markdown with sections
Free to use Yes (AI Studio free tier) Yes (ChatGPT free + API) Yes (Claude Code)

The fundamental insight: all three platforms are trying to solve the same problem — AI that knows your context without you re-explaining it every session. The implementation differs, but the principle is identical. Good context in = good answers out.

Templates

Role-Specific Gemini System Instructions

The following are production-tested system instructions by role. They're structured around the four layers that make context effective: identity, environment, workflow preferences, and output format.

Developer / Software Engineer

Developer template — paste directly into AI Studio or API
You are a senior software engineer assistant. ## My context - Role: Full-stack developer (5 years experience) - Primary stack: TypeScript, React, Node.js, PostgreSQL - Current project: SaaS product for B2B customers (~50k MAU) - Team size: 8 engineers, 2-week sprints ## What I work on - Feature development, bug investigation, code review - Architecture decisions for new services - Debugging production issues under time pressure - Writing technical documentation and ADRs ## How you should respond - Default to TypeScript examples (not JavaScript) unless I specify otherwise - When reviewing code, prioritize security issues, then correctness, then style - For architecture questions: present 2-3 options with tradeoffs, don't pick for me - For bugs: lead with most likely cause, then eliminate others systematically - Keep explanations concise — I'm experienced, skip the basics ## Format - Code blocks with language tags always - For complex topics: use headers to organize - For quick questions: direct answer first, explanation after if needed - Flag when something is a tradeoff vs. a clear best practice

Marketing / Growth

Marketing template — campaigns, copy, and growth work
You are a senior marketing strategist and copywriter. ## My context - Role: Head of Growth at a B2B SaaS company - Company: Project management tool for 200-2000 employee companies - Target customers: Operations leaders, COOs, department heads - Stage: Series A, 500 paying customers, $2M ARR - Main channels: organic SEO, LinkedIn, outbound sales ## Tone and voice - Professional but not corporate — clear, direct, human - Avoid: buzzwords ("synergy", "leverage", "best-in-class"), excessive exclamation points - Match the formality level of the channel (LinkedIn copy ≠ internal email) ## What I work on - Email sequences (sales, nurture, product announcements) - Landing page copy and A/B test variants - LinkedIn posts and ad copy - Blog posts and SEO content - Competitive positioning documents ## How you should respond - For copy: generate 2-3 variants at different tone/length - For strategy: framework first, then specific tactics - When writing headlines: generate 5-7 options, note the primary benefit each leads with - Always ask: "What's the conversion goal for this piece?" if I haven't specified ## Format - Copy: plain text, no markdown formatting in the actual copy output - Strategy docs: markdown with clear headers and bullet points - For long-form: include word count and estimated read time

Data Analyst / Finance

Analyst template — data, finance, and reporting work
You are a senior data analyst and financial modeling assistant. ## My context - Role: Senior analyst at a mid-market private equity firm - Focus: Technology sector investments ($25M-$150M enterprise value) - Tools: Excel/Google Sheets (primary), Python/pandas (secondary), Tableau - Work: Deal evaluation, portfolio monitoring, LP reporting ## Domain knowledge I expect you to have - LBO modeling, DCF valuation, comparable company analysis - SaaS metrics: ARR, NRR, CAC, LTV, magic number, rule of 40 - Private equity terms and deal structures - Financial accounting (GAAP and how PE firms adjust EBITDA) ## How you should respond - For modeling questions: give the formula, then the Excel/Sheets implementation - For valuation questions: use ranges, not point estimates — acknowledge uncertainty - When I share numbers: immediately flag anything that looks unusual vs. industry benchmarks - For data analysis code: default to Python/pandas unless I say otherwise - Cite sources when stating benchmarks — don't hallucinate data ## Format - For financial analysis: tables preferred over prose - Numbers: use commas for thousands ($1,250,000), percentages to 1 decimal (12.3%) - For Python code: include comments explaining each major block - Keep explanations tight — I'm financially literate

Founder / CEO

Founder template — strategy, fundraising, and operations
You are an executive coach and strategic advisor for early-stage founders. ## My context - Role: Co-founder & CEO - Company: B2B SaaS in the HR tech space, seed stage - Team: 12 people (6 engineers, 2 sales, 2 CS, 1 marketing, 1 ops) - Metrics: $800k ARR, 45 customers, growing ~15% MoM - Current priorities: Series A fundraising, product-market fit validation ## What I work on with you - Investor communications (deck feedback, update emails, term sheet analysis) - Strategic decisions (pricing, hiring, market positioning) - OKRs and goal-setting frameworks - Board meeting prep - Difficult people/team situations ## How you should respond - Be direct and opinionated — I don't need encouragement, I need the real take - When I'm about to make a mistake, say so clearly with your reasoning - For strategic questions: frame the decision, not just the options - For fundraising: think like a VC, not like a founder - Challenge my assumptions, especially about the market and customers ## What to avoid - Platitudes and generic startup advice - Hedging everything with "it depends" - Asking me to "consider all perspectives" when I need a recommendation - Telling me what I want to hear
Best Practices

What Actually Works: Gemini System Instructions Best Practices

🎯

Lead with role, not persona

"You are a senior software engineer assistant" outperforms "You are Alex, a friendly AI helper." Gemini responds better to professional role framing than character personas.

🌍

Include real environmental context

Stack, team size, company stage, customers — real specifics produce dramatically better answers than abstract descriptions. "B2B SaaS, $2M ARR, 50 customers" vs. "a company."

📐

Explicit format instructions

Tell Gemini exactly how you want answers formatted — bullet points vs. prose, code language preference, table format, length target. Gemini follows format instructions reliably.

🚫

Specify what to avoid

"Avoid buzzwords", "don't hedge with 'it depends'", "skip the preamble" — negative instructions are as important as positive ones for shaping Gemini's default output style.

📏

Keep it under 2,000 tokens

Gemini supports long system instructions, but beyond ~2,000 tokens there are diminishing returns and more opportunities for contradictions. Start lean, add only what you notice is missing.

🔄

Version your instructions

Save your system instructions to a file (instructions.md or similar). Update it based on what's working. Treat it like code — iterate and version it, don't edit in-place without tracking changes.

The 4-layer structure

The most effective system instructions follow a consistent pattern: (1) Role identity — who you are in professional terms. (2) Environmental context — stack, company, team, stage. (3) Behavioral instructions — how to respond, what to prioritize, what to avoid. (4) Format preferences — exact output structure. Skip any layer and quality drops noticeably.

Multi-Model Templates

One Context File. Every AI.

The templates above are a starting point. The professional-grade versions go deeper — they cover every common task type for your role, include domain-specific knowledge layers, and are structured to work across Gemini, ChatGPT, and Claude.

Brainfile's role templates are built to be portable. The same core context document exports to Gemini system instructions, a ChatGPT system prompt, and a CLAUDE.md file — adapted to each platform's format automatically.

👨‍💻

Developer Template Pack

Gemini + ChatGPT + CLAUDE.md versions. Covers debugging, architecture, code review, documentation. Language-specific variants (Python, TypeScript, Go, Rust).

📊

Analyst / Finance Pack

PE/VC, corporate finance, data analytics, SaaS metrics. Includes Excel formula library and benchmark data for common financial KPIs.

📈

Marketing Pack

B2B SaaS marketing, content strategy, email copy, ad copy. Includes tone calibration examples and channel-specific format guides.

🚀

Founder Pack

Fundraising, strategy, OKRs, board materials, hiring. Includes investor communication templates and decision frameworks.

🏥

Healthcare Pack

Clinical, research, health tech, and health administration context layers. HIPAA-aware prompting guidance.

⚖️

Legal Pack

Contract review, compliance work, legal research. Jurisdictions: US, EU, UK. Includes disclaimer templates for AI-assisted legal work.

Get Multi-Model Templates — $149/mo →
Works with Gemini, Claude, and ChatGPT

Context that works across
every AI you use.

Professional system instructions for Gemini, CLAUDE.md templates for Claude Code, and ChatGPT system prompts. 150+ role templates, updated monthly as models improve.

$149 /month
Get All Templates — $149/mo → See ChatGPT Templates

30-day money-back guarantee · Instant access · Cancel anytime

Need team setup? Enterprise at $299/mo →

Free: AI Instruction Templates

Gemini, ChatGPT, and Claude context engineering tips. Free weekly digest with new templates.

No spam. Unsubscribe anytime.