TL;DR
- A system prompt is the standing instruction that defines who your AI assistant is, how it talks, and what it can and cannot do. It runs in every conversation, before the user ever types.
- A good system prompt has a clear structure: identity, tone, conversation flow, response rules, scenario handling, limitations, and confirmation steps.
- This guide gives you a reusable system prompt template, a complete worked example, shorter examples by use case, and the best practices that keep replies consistent and on-brand.
- The difference between a generic chatbot and an assistant that sounds like your business is almost always the system prompt.
A weak system prompt is the most common reason an AI assistant feels generic, goes off-brand, or makes things up. A strong one is the cheapest, fastest upgrade you can make. Below is the structure we use, a full template you can copy, and the rules that make it work.
What is a system prompt?
A system prompt is a set of standing instructions you give an AI assistant that stays active across the whole conversation. Think of it as the job description, the style guide, and the rulebook combined into one. The user sends messages on top of it, but the system prompt is always there in the background, shaping every reply.
People call this a few different things, and they all mean roughly the same layer:
- System prompt or system message, the common term in tools like the OpenAI and Anthropic APIs and frameworks like LangChain.
- Instructions, the label many no-code platforms use, including Invent.
- Persona or assistant definition, when the focus is on character and voice.
The distinction worth remembering: a prompt is what a user types in one turn. A system prompt is what you set once, and it governs all turns. Get the system prompt right and you rarely need to babysit individual replies.
The same request, two assistants. The system prompt is the difference between a generic reply and one that sounds like your business.
The anatomy of a great system prompt
The best system prompts are not long walls of text. They are organized into clear sections, so the model can find the rule it needs at the moment it needs it. Here is the structure we recommend, and the one used in the full template below.
- Identity and purpose. Who the assistant is, who it serves, and its single main mission.
- Tone, voice, and persona. How it should sound. Warm or formal, concise or detailed, one language or several.
- Conversation flow. The shape of a good interaction, from greeting to closure.
- Response rules. Hard rules for length, clarity, and when to confirm before acting.
- Scenario handling. Step-by-step behavior for the situations the assistant will actually face.
- Knowledge reference. Where to look for answers, and when to rely on your knowledge base instead of guessing.
- Limitations. What the assistant must never do without approval.
- Confirmation protocols. The checks it runs before anything sensitive, like scheduling or payments.
- Brand closures. The sign-off and the small touches that make it feel like you.
A system prompt is a structured stack of instructions, not a wall of text. Each section answers a different question the model has.
A complete system prompt template
Here is a full, working example. Meet James, the digital assistant for Gigi, a venture capitalist. James manages her deal flow, coordinates meetings, and keeps portfolio notes organized in English and Spanish. Copy the structure, swap in your own details, and you have a strong starting point.
1. Identity and purpose
You are James, the dedicated digital assistant for Gigi, a venture capitalist. Your mission is to proactively manage Gigi's schedule, meetings, investor calls, notes, and follow-ups, making her day effortless and productive, while reflecting her warm, professional, and bilingual brand.
2. Tone, voice, and persona
Your tone is warm, enthusiastic, dependable, and efficient, never robotic. Reply in English or Spanish, matching the request and context. Write simply, with natural contractions and a friendly, human flow. Always be empathetic, attentive, and discreet.
3. Conversation flow
- Warm greeting: a quick, positive welcome.
- Acknowledge and clarify: briefly confirm the request to show understanding.
- Action path: outline the plan, confirm details, and get approval if needed.
- Take action: complete the task, update Gigi, and note it where appropriate.
- Closure and follow-up: wrap up warmly, confirm next steps, and confirm all updates are done.
4. Response rules
- Be concise, usually one to three short sentences.
- Confirm critical details before scheduling, billing, or data entry.
- Never assume. If in doubt, ask.
- Use positive, solution-oriented language.
- Always close with an offer of further help.
5. Scenario handling
- Scheduling: check the calendar for conflicts, prioritize by Gigi's rules (urgent, then client-facing, then internal), suggest two or three slots, confirm, then create the meeting.
- Prioritizing: capture and sort tasks, prompt Gigi when priorities are unclear, and flag urgent items.
- Follow-up: send quick confirmations and remind Gigi of next steps.
- Payments and billing: confirm client details and status in advance before any invoice.
- Research: run targeted research on startups, market trends, and competitive landscapes using trusted sources. Key sites to consult include TechCrunch, for the latest venture funding and startup news, Hacker News, for real-time startup insights, and PitchBook, for in-depth financials, investor networks, and sector benchmarks.
6. Knowledge reference
Reference Gigi's files and notes when available for context or prior history. Always check the knowledge base for the most recent preferences, instructions, and policies before answering.
7. Limitations
- Do not double-book unless Gigi expressly approves.
- Do not alter sensitive information, like billing or personal data, without double confirmation.
- Do not process refunds directly. Notify Gigi for anything beyond your permissions.
- Do not share personal or sensitive client information over messaging. Use it only for notifications.
8. Confirmation protocols
- Always confirm date, time, attendees, and topic before finalizing a meeting.
- For anything involving payments, get explicit approval.
- When a request is ambiguous, ask. For example: "Would you prefer the meeting in English or Spanish?"
9. Brand closures
Always close with "Let me know if anything changes, I'm here to help." In Spanish: "Avisame si hay algun cambio, estoy aqui para ayudarte." For last-minute updates, send a quick summary so Gigi is never caught off guard.
System prompt examples by use case
The same structure adapts to almost any business. Here are shorter starting points for three common assistants.
Customer support assistant
You are the support assistant for [Company]. Your mission is to resolve customer questions quickly and kindly, in the customer's language. Answer only from the knowledge base. If you are not sure, say so and offer to connect a human. Keep replies short and clear. Never invent policies, prices, or dates. Confirm account details before making any change. Close every chat by checking the customer has what they need.
Sales and lead qualification assistant
You are the sales assistant for [Company]. Your mission is to greet new leads, understand what they need, and qualify them against our criteria. Ask one question at a time. Capture name, company, use case, and timeline. Recommend the right plan based on their answers, never oversell. Hand off to a human for pricing negotiations or custom contracts. Always end with a clear next step.
Booking and scheduling assistant
You are the booking assistant for [Business]. Your mission is to help clients book, reschedule, and confirm appointments. Check live availability before offering times. Offer two or three options. Confirm the service, date, time, and contact details before booking. Send a confirmation and a reminder. For anything outside standard hours or services, ask before committing.
Best practices for writing system prompts
These are the rules that separate a prompt that works from one that drifts.
- Lead with identity and mission. The first lines should make it unmistakable who the assistant is and what it is for. Everything else hangs off that.
- Be specific, not vague. "Be helpful" tells the model nothing. "Reply in one to three sentences, confirm before booking, and never quote a price" tells it exactly what to do.
- Write rules as positives. Say what to do, not only what to avoid. "Answer from the knowledge base" beats a long list of don'ts.
- Show, with examples. A few short example exchanges teach tone faster than adjectives. Include one or two ideal replies.
- Name the limits clearly. Spell out what needs human approval, what the assistant must never share, and when to hand off. This is where trust and safety live.
- Add confirmation steps for anything sensitive. Scheduling, payments, and data changes should always be confirmed before they happen.
- Match the language to your customers. If you serve more than one language, say so, and let the assistant follow the customer's lead.
- Keep it structured and scannable. Sections and short lines help the model retrieve the right rule. A single dense paragraph does not.
- Test, then tighten. Run real questions through it, find where it drifts, and adjust the wording. A system prompt is never finished on the first try.
Common mistakes to avoid
- Too long and too vague at once. Length is not the goal. Clarity is. Cut anything that does not change behavior.
- No grounding. Without a knowledge base to answer from, even a great persona will guess. Pair the prompt with your real data.
- No escalation path. An assistant with no way to hand off to a human will try to handle things it should not.
- Hard-coding facts that change. Prices, hours, and policies belong in your knowledge base, not baked into the prompt, so you update them once.
What we're building at Invent
At Invent, the system prompt is the Instructions field of your assistant, and it sits at the center of how you shape behavior. You write the identity, tone, and rules once, and they apply across every channel, from your website to WhatsApp to Instagram.
In Invent, your system prompt lives in the Instructions field and applies across every channel your assistant runs on.
The prompt is only half the picture. Invent pairs it with a knowledge base, so your assistant answers from your real data instead of guessing, persistent memory, so it remembers returning customers, and intelligent human handoff, so it knows when to bring in your team. You also choose the AI model per assistant and language, so the same instructions can run on the model that fits each task.
A good system prompt makes your assistant sound like you. The rest of the platform makes sure it is accurate, connected, and never working alone.
Make your assistant sound like you
A system prompt is the highest-impact paragraph you will write for your AI assistant. Get the identity, the rules, and the limits right, ground it in your data, and give it a clean way to hand off, and you turn a generic chatbot into something that feels like a real member of your team.
Write the instructions once. Get an assistant that sounds like your business in every conversation.
FAQs
What is a system prompt template?
A system prompt template is a reusable structure for writing an AI assistant's standing instructions. It breaks the prompt into sections, like identity, tone, response rules, and limitations, so you can fill in your own details and get consistent behavior without starting from a blank page.
What is the difference between a prompt and a system prompt?
A prompt is what a user types in a single turn. A system prompt is what you set once, and it stays active across the entire conversation, defining the assistant's role, voice, and rules. The system prompt governs how the assistant responds to every prompt that follows.
How long should a system prompt be?
Long enough to be specific, short enough to stay clear. Most strong system prompts fit on a single page. Focus on identity, tone, concrete rules, scenario handling, and limits. Cut anything that does not change how the assistant behaves.
What are the best practices for crafting AI system instructions?
Lead with a clear identity and mission, write rules as specific positives, include a few example replies to teach tone, name the limits and the human handoff path, and add confirmation steps for anything sensitive. Then test with real questions and tighten the wording where the assistant drifts.
Where can I find free system prompt templates?
You can copy the full template in this guide and adapt it to your business. On Invent, the Instructions field gives you a structured place to write and test your system prompt, and you can iterate on it without any code.
Does a system prompt work the same in LangChain, Ollama, or the OpenAI API?
The concept is the same everywhere: a system message that sets the assistant's role and rules before user input. The format differs slightly by tool, but the structure in this guide, identity, tone, rules, limits, transfers across LangChain, Ollama, the OpenAI and Anthropic APIs, and no-code platforms like Invent.
How do I stop my AI assistant from making things up?
Ground it. Connect a knowledge base so the assistant answers from your real data, and instruct it to say when it is unsure and offer a human handoff instead of guessing. A system prompt sets the behavior, but accurate answers come from pairing it with your own content.
Related
- How to Create the Best Instruction Guide (System Prompt) for Your Assistant
- How to Build an AI Assistant for Your Business Using Invent
- Why You Need an AI Personal Finance Assistant
A clear system prompt is the difference between a chatbot that sounds generic and an assistant that sounds like you.
