Tired of typing the same instructions into the chat every time - like "keep it brief, no fluff, write for decision-makers"? It’s a waste of time and gets annoying fast. Custom instructions solve this. You define how ChatGPT should respond once, and the model sticks to it for every chat from then on. Tone, format, level of detail, and target audience are set before you even ask your first question. Technically, these are persistent instructions - guidelines that an LLM keeps in mind permanently.
This feature is available from all major providers, though the names vary. ChatGPT calls them "Custom Instructions," Claude uses "Instructions for Claude," and others refer to them as "Saved Info" or personalization. Here, you'll learn what these instructions are good for, where to set them up, and how to structure them. At the end of this article, you'll find seven templates you can copy and adapt right away.
What are custom instructions in ChatGPT?
Custom instructions are a text field in your ChatGPT settings where you define once how you want ChatGPT to respond. These guidelines then apply automatically to your chats, so you don't have to repeat them every time. Typical content includes tone, response format, target audience, and your professional role.
What are custom instructions used for?
These instructions turn a general-purpose chatbot into a reliable work tool. You define how the model should respond once, and the output is ready to use every time. Most often, this involves rules for format and quality, style and tone, addressing a specific target audience, and focusing on roles and context. Many people also use this feature as a safety net: asking for clarification when things are unclear, not guessing, and avoiding sensitive data. The common goal is always the same: you need less repetition and fewer follow-up prompts, and you get more consistency because your preferences don't have to be renegotiated with every request. In ChatGPT, these instructions are stored centrally and applied to all conversations.
Typical use cases
To show you how custom instructions help in everyday work with LLMs, here are a few concrete examples.
Work mode for actionable answers
Always the same output format: key takeaway, three options, risks, next steps. This leads to fewer follow-up questions and avoids walls of text.
Setting a permanent tone and writing style
For example, "short, factual, no buzzwords, in German" or "friendly, empathetic, informal." This saves you from having to make constant style corrections.
Target audience translator
"Explain everything for ambitious laypeople or decision-makers and define technical terms the first time they appear." This makes content ready to share immediately.
Role and focus persona for recurring tasks
For example, "Act as a project manager and structure tasks, risks, and responsibilities." Or "Act as an editor and provide headline variations, a clear thesis, and no filler sentences."
Quality assurance and safety rails
"If information is missing, ask 1 or 2 follow-up questions. If you are unsure, say so. Do not speculate." This reduces hallucinations and false confidence.
Language and format standards for daily use
For example, "always use Markdown with headings," "always use bullet points," "always use a table for comparisons," or "always include a short summary at the end."
The best templates for custom instructions in ChatGPT
Everyone who uses LLMs has their own unique style. To get you started, here are seven templates you can copy, adapt to your job, and paste into your personalization settings. Just replace the info in brackets with your own details. The next section explains where to paste the template and how many characters are allowed. If you want to dive even deeper into the topic, check out our guide to prompt engineering.
Template 1: The All-Rounder
A short, structured standard for anyone who primarily wants actionable answers.
By default, provide short, structured, and actionable answers.
FORMAT
1) Key takeaway in 1 sentence
2) 3 options (1 to 2 sentences each)
3) Risks and trade-offs (max. 3 bullet points)
4) Next steps (max. 5 bullet points, concrete)
RULES
- No fluff, no repeating my question.
- If information is missing, ask a maximum of 2 follow-up questions and do not guess.
- If you are unsure, say so clearly.
Template 2: Executive Management
For decisions where time is tight and numbers matter.
I run a company with [number] employees in the [industry] sector.
Write responses as a decision brief: recommendation first, followed by the 3 most important reasons, then risks and costs.
State assumptions explicitly and highlight what you don't know.
Compare options in a table showing effort, benefit, and risk.
Keep it under 200 words unless I ask for more depth.
If key data (budget, timeframe, goal) is missing, ask questions first.
Template 3: Marketing
For copy, campaign ideas, and editorial work.
I work in marketing at [company] (industry: [industry]). Target audience: [target audience].
Write in a [tone, e.g., clear, friendly] style with short sentences and concrete examples.
Avoid buzzwords, superlatives, and filler sentences.
For writing tasks, always provide 3 headline variations and a clear core message.
For campaign ideas, always include the goal, channel, and how to measure success.
Check claims for evidence and highlight any unsupported figures.
Template 4: Sales
For prospecting, proposals, and meeting preparation.
I work in sales for [product or service] and talk to [target customers, e.g., mid-sized business executives].
Keep emails and messages short (max. 120 words), personal, and include exactly one clear next step at the end.
Prepare for meetings by outlining: the goal of the meeting, 3 questions for needs analysis, and 2 likely objections with appropriate responses.
Use placeholders like [Customer A] and do not invent references or figures.
Template 5: Customer Service
For friendly and solution-oriented responses to customer inquiries.
I answer customer inquiries for [company].
Write in a friendly, calm, and solution-oriented manner using a professional tone.
Start by briefly acknowledging the issue, then provide the solution in clear steps, and finish with the next step.
Do not promise anything I haven't explicitly authorized (refunds, deadlines, special terms).
If information is missing, ask the customer for clarification.
If a case needs to be escalated, say so and suggest forwarding it.
Template 6: Research
For those who need reliable information instead of quick claims.
I am researching technical topics and need reliable information.
In every response, clearly distinguish between verified facts, assessments, and open questions.
For every important fact, cite the source or explicitly state that you do not have one.
If sources contradict each other, present both positions.
State clearly if you do not know something or if information might be outdated.
At the end, summarize what is considered verified in 3 sentences.
Template 7: Excel and spreadsheets
For formulas, data preparation, and analysis.
I work in Excel ([Version, e.g., Microsoft 365]) daily with [Type of data, e.g., sales figures].
Always provide formulas in a way that I can copy them directly, and explain them afterward in 1 to 2 sentences.
Prefer standard functions like XLOOKUP, SUMIFS, and PIVOT before suggesting macros.
Ask for column names and example rows if the data structure is unclear.
Point out typical sources of error, such as spaces, text formats, and duplicates.
Where do you save custom instructions in ChatGPT?
Custom instructions are found in the settings, not the chat window. In ChatGPT, you'll find them under "Personalization" or "Customize ChatGPT," and in Claude under "Instructions for Claude." These settings are tied to your account and automatically apply to your chats. If you're building your own tools, you can alternatively set them in your code via the API.
In ChatGPT, personalization now consists of three areas:
- Personality and tone: Here, you can choose a preset like "Professional" or "Efficient" and use sliders to adjust how warm ChatGPT sounds and how often it uses emojis and lists.
- Custom instructions: The free-text field for your own rules. This article focuses on this section.
- Memory: ChatGPT remembers details from your conversations and refers back to them later. You can view and delete these entries in your settings.
The character limit is important: free accounts and the Go plan have a 1,500-character limit, while paid plans have had up to 5,000 characters since July 2026. All the templates below will fit within the smaller limit if you shorten them slightly.
The same principle works technically in developer setups. There, the instruction is embedded as a system message (system prompt) into every request, usually server-side or within a workflow tool. This ensures all users and sessions follow the same rules.
How many instructions can I use and how do I switch between them?
Most LLM apps have a global area for persistent instructions. You can store multiple rules there as bullet points, essentially creating one instruction with many sub-points. You no longer need to copy and paste for different modes, such as work mode and creative mode. ChatGPT and Claude offer projects with their own instructions. These apply to all chats within that project and supplement the global settings. For example, you can create a "Customer Emails" project and a "Blog Articles" project, each with its own appropriate rules. If you don't use projects, you can save 2 to 3 ready-made instruction blocks as text snippets in Notion, OneNote, or Word and swap them as needed. This is especially worthwhile in everyday professional life, and you can find more on this in our article on ChatGPT for companies.
UI vs. API
The difference between UI (User Interface) and API (Application Programming Interface) is important because it explains where you actually set persistent instructions. When you use an LLM via the website or app, you are working in the UI. You enter your instructions once in the settings, and they automatically apply to new chats. The API is the programming interface. It comes into play when you build or automate something yourself, such as an internal tool, a workflow, or a script.
Tips for storing system prompts
1. Drafting
Write instructions so the model knows exactly what you want. This works best with clear, positive rules like "Answer in 5 bullet points" or "Ask follow-up questions if information is missing." Vague prohibitions like "don't be too long" or "no bad answers" give the model little guidance. Keep instructions short and prioritize: 5 strong rules work better than 20 that are only moderately clear. After saving, test immediately with 2 or 3 typical questions to see if the format, tone, and level of detail are right.
Looking for proven templates and concrete prompting examples? Check out our collection of the best ChatGPT commands, codes, and prompts.
2. Data Privacy
Treat persistent instructions like regular work data. Do not enter passwords, health data, or personal information about customers or employees. If you work in a team, only include information that would be appropriate for an internal handbook. Use placeholders like "Customer A" or "Project X" and keep sensitive details in the specific chat or, better yet, in secure, authorized systems.
3. Testing and Operations
Changes to the system prompt should not be made live. Test new versions first in a small sandbox, such as a separate test chat or test profile, using the same standard cases every time. Define 1 or 2 example answers as a reference and version your prompts (v1, v2, v3) so you can quickly roll back if there are issues. After rolling out, monitor briefly to see if the rules are being followed consistently before refining further.
The next step: Agents as an extension
Persistent instructions ensure consistent answers. If you want to automate recurring tasks, you reach the next level of maturity: agents. An agent is more than just good prompting. It is a system that performs work. It has a goal, follows a workflow logic, can access tools depending on the platform (such as files, calendars, email, the web, or apps), and delivers a result that goes beyond mere text. Our guide on creating AI agents shows you how to build your own.
"Custom instructions improve responses in chat. But if you want real scalability, build agents. They use your instructions to autonomously access tools and deliver finished work instead of just text." Learn how to build an agent without developer skills here: Building an AI Agent for Businesses – No Developers Required.
In practice, this is the transition from getting answers to implementing automation solutions. Recurring processes are modeled as workflows and executed semi-autonomously by the agent. A typical pattern is: check input, retrieve information, create a proposed decision, ask follow-up questions, and write the output to a target system. Depending on the level of maturity, this ranges from simple templates where the agent guides you step-by-step to true process automation using tools like Power Automate, n8n, Zapier, or internal APIs. The added value lies in fewer context switches, fewer manual handovers, and reproducible quality.
Conclusion
Persistent instructions are the fastest way to take an LLM from "somewhat helpful" to "reliably useful." They save time because you don't have to re-explain tone, format, and quality rules for every request, and they increase consistency because the model has a stable framework. The most important success factor is clarity: a few prioritized rules that you test and refine as needed. If you implement this properly, you'll get fewer follow-up prompts, less junk text, and significantly more output that you can use right away.
The best way to start: Take one of the templates above, adapt it to your job, and test it in your daily routine for a week. If you want to turn these into repeatable processes, agents are the logical next step. They combine these rules with goals, tools, and workflow logic to handle tasks semi-automatically.
Tired of settling for chat answers and want to run processes completely autonomously? Click here for our major comparison test: The 7 best automation tools of 2026 compared.
Get a free strategy call and discover how AI can help you automate workflows, create better content, and scale your marketing.
Book a call
FAQ
Where can I find custom instructions in ChatGPT?
In the settings under "Personalization" or "Customize ChatGPT." There are three areas there: personality and tone, custom instructions, and memory.
How many characters are allowed for custom instructions?
For free accounts and the Go plan, it's 1,500 characters. As of July 2026, paid plans allow up to 5,000 characters.
Can I use multiple persistent instructions at the same time?
There is usually one global instruction area per account where you can store multiple rules as bullet points. For different modes, use the "Projects" feature offered by ChatGPT and Claude, which allows for their own specific instructions.
What is the difference between a persistent instruction and a regular prompt?
A regular prompt only applies to the current request. Persistent instructions set a lasting framework for tone, format, and behavior across many requests.
What should I never include in persistent instructions?
No passwords, no health data, and no personal information like customer names or private contact details. Use placeholders and only add sensitive details where it is truly necessary and authorized.
When are instructions no longer enough, and when do I need agents or automation?
When you don't just want answers, but need to reliably handle recurring processes—like checking, summarizing, drafting, and transferring data to a tool. That’s when agents and workflows make sense, as they combine goals, steps, and tool access.
Posted by:
Jens is the co-founder of bakedwith, a boutique consultancy firm specialising in smart automation and AI. He helps medium-sized companies and corporations to optimise processes, reduce manual work and achieve growth through efficient workflows. Previously, he was Growth Lead at OMR and worked as a freelancer, helping numerous companies optimise their CRM and automation systems. Jens is passionate about combining growth and efficiency to enable teams to focus on what matters most.
.webp)
.webp)
.webp)
