Custom Instructions
Standing rules you set once so an AI tool applies them to every request, rather than restating the same context each time. They are the stable half of personalisation; memory is the remembered half.
What Custom Instructions are
Custom instructions are standing rules you set once so an AI tool applies them to every request. Without them, each conversation starts from nothing and you restate who you are, how you want the reply written and what you are working on. With them, that context is already in place, and a prompt only has to carry what is specific to the task in front of you.
The aim is plain. The AI Agents and Personalisation session states its outcome as participants who can set a chatbot up around their own role, so it stops asking for the same context every time. Custom instructions are the first tool for that.
The Stable half of Personalisation
Personalisation on this site has two halves, and custom instructions are the stable one. They are rules you write yourself and they stay as written until you change them. Memory is the other half. It is what the tool keeps from one conversation to the next, and it changes as you work.
The two are taught together for that reason. The personalisation session sets them out as stable rules against remembered context, and gives the HR course a session of its own on custom instructions versus memory and when to use which. Choosing between them is a judgement about how long a fact stays true. A rule that holds for every request goes in the standing set. Something that belongs to this month’s project is context, and it may sit in memory or in the prompt.
What goes in them
The personalisation outline gives a good picture of the ground they cover. It works through traits and style, personality and tone, occupation and background, and naming and identity. In plain terms, that is who you are, what your work is, how you want replies to sound and what the tool should be called.
Two other parts of the catalogue add to that. The prompting workshop teaches role and backstory as part of the context you give a tool, and personas as a way to make an answer come from a specific kind of expert or in a house style. Custom instructions are where the parts of that which never change can live. Everything that changes with the task stays in the prompt.
How they fit with the four parts of a Prompt
The prompting workshop teaches four essential parts of a prompt: goal, context, source and expectations. Read against that method, custom instructions carry the parts that never change. Your role and background are context, and the shape you want replies to take is an expectation. The goal and the source belong to the task, so they stay in the prompt each time.
The editors and authors session lays the sequence out in order. It teaches custom instructions, then traits, then memory, then when to use which, and only after that moves on to advanced persona design and meta prompting. Standing rules come first in that order, ahead of the more advanced techniques.
A Prompt that writes them for you
The personalisation session includes a prompt that writes both your custom instructions and your memory for you, so you do not have to start from a blank box.
The one-to-one programme takes the same idea further. It drafts instructions so the AI writes the way this person writes, and lists custom instructions among its first sessions, so that everyday requests need no set-up at all. Those early sessions clear the ground before anything gets built.
A worked example
The personalisation session gives a use case that shows the point: standardised ticket replies. Replies of one kind should come out the same way each time, and standing rules are how a tool learns that once instead of being told for every ticket.
The one-to-one programme adds a learning loop to the same idea, in which a person’s corrections become dated rules the tool keeps. A correction made once is not made again, and a dated rule can be reviewed later.
Across the Tools
The idea is not tied to one product. The ChatGPT courses list personalisation and settings as a part of the syllabus, and the Mistral workshop covers custom instructions, memory, stable preferences, changing context and a personalised chatbot set-up. The editors and authors session and the advanced generative AI course both include it, and the advanced course adds a topic on personalising your chat.
The site does not describe any product’s settings screen, so a team should check the tool it uses. What carries across is the skill: deciding which of your preferences are stable enough to write down.
Custom Instructions, Projects, Custom GPTs and Agents
Custom instructions are the lightest way to make a tool behave as you want, and the catalogue treats the heavier ones as choices. The Mistral workshop closes on a decision framework that asks when to use chat, a project, a library, memory, custom instructions, an agent, a connector, Deep Research or Work Mode. The point is to choose the smallest option that does the job.
A custom GPT is the next step up in the ChatGPT courses. It is an assistant that holds a role and a set of rules for one job, so whoever uses it does not write the brief each time. The advanced course asks when one is worth building and where a saved prompt is enough. An AI agent goes further again. Its instructions are a prompt it holds permanently, and it adds a source and tools. Standing rules for you, in general, come first.
When a rule works against you
A standing rule applies to every request, and that is both its value and its risk. The Mistral workshop says it directly: the useful judgement is which of your preferences are stable enough to set once and which change with the task, because a rule written for one job can quietly spoil the next.
When a tool starts giving results that do not suit the task, the standing set is one of the first places to look. The fix the workshop points to is the same judgement: keep in it only what is stable, and leave what changes with the task in the prompt.
What not to put in
Everything you type into an AI tool is data, and custom instructions are typed text like any other. The site’s article on why banning AI at work backfires argues for a rule drawn by kind of data and not by tool. Client information, contracts and HR records should not go in unless a rule says they may.
The same rule covers the standing set. Keep it to how you work and what your role is, and leave out anything a rule says may not be entered. Vendors state that business content typed into their tools is not used to train their models by default. That is a policy each vendor publishes, and a business should read it for itself, for the account it will actually use.
Mix-ups worth avoiding
Four confusions come up often enough to name. The first is treating custom instructions and memory as one thing, when one is written by you and stays fixed and the other is kept by the tool and changes. The second is treating custom instructions as a custom GPT, which holds a role and rules for one job and not for every request.
The third is filling the standing set with rules that only suit one kind of work. The fourth is assuming a well-briefed tool is a correct one. Standing rules change how a tool writes, not whether what it says is true.
Where it goes wrong
Custom instructions do not check facts. A tool can follow every rule you wrote and still produce a reply that reads well and is wrong. That is hallucination, and it is a property of how generative AI works, not a fault in your instructions.
The working habit does not change. Ask for the source, open the source and treat the first reply as a draft. The person who sends the result on is responsible for what it says. A tool that writes in your voice makes this easier to forget, because the reply sounds like you, and that is the moment to read it most carefully.
Where it is taught and where it appears in an Audit
The AI Agents and Personalisation session is where custom instructions and memory are set up together, and the one-to-one programme sets them up in its first sessions. The Mistral Le Chat workshop covers them for its own tool, and the HR course teaches when to use them against memory. Six of the courses are the core AI courses, and the rest sit in the wider training catalogue.
An audit meets custom instructions as a sign of repetition. Where a team restates the same context in every request, the work repeats, and the prompting workshop’s decision framework sorts it by how often it repeats and how much judgement it needs. A task that recurs and follows rules suits a template or a scheduled prompt. One that needs several steps, a source and a tool suits an agent.
FAQ
Questions about Custom Instructions
What are Custom Instructions?
Custom instructions are standing rules you set once so an AI tool applies them to every request, instead of you restating the same context each time. They are the stable half of personalisation. Memory is the remembered half.
What is the difference between Custom Instructions and Memory?
Custom instructions are stable rules, written by you, that apply to every request. Memory is what the tool keeps from one conversation to the next, and it changes as you work. The personalisation session teaches the split as stable rules against remembered context, and the HR course teaches when to use which.
What should go into Custom Instructions?
The things that stay true whatever the task. The personalisation session works through traits and style, personality and tone, occupation and background, and naming and identity. Your role, the way you want replies written and the format you prefer are the usual candidates. Anything that changes with the job belongs in the prompt.
Can Custom Instructions make an AI Tool write in my voice?
They are the start of it. The one-to-one programme drafts instructions so the AI writes the way this person writes, sets them up in the first sessions, and adds a learning loop in which corrections become dated rules the tool keeps. The instructions carry the standing style. You still read and correct what comes back.
Are Custom Instructions the same as a Custom GPT?
No. Custom instructions are standing rules applied to your requests in general. A custom GPT is an assistant that holds a role and a set of rules for one job, so whoever uses it does not write the brief each time. The advanced course also asks where a saved prompt is enough, and does not treat a custom GPT as the answer to every repeated task.
Can a rule I set for one job spoil another?
Yes. The Mistral workshop makes the point directly: a rule written for one job can quietly spoil the next. The judgement it teaches is which of your preferences are stable enough to set once and which change with the task. If a rule only suits one kind of work, keep it out of the standing set.
Mistral Le Chat Workshop: From Chatbot to Personalised AI Agents
Is it Safe to put company Information into Custom Instructions?
Treat the text as data, because it is. Everything you type into an AI tool is data, and the site's rule is drawn by kind of data and not by tool. Client information, contracts and HR records should not go in unless a rule says they may. Vendors state that business content is not used to train their models by default, which is a policy each vendor publishes and a business should read for itself.
Do Custom Instructions stop a Tool getting things wrong?
No. They set how the tool writes and what it assumes about you. They do not check a fact. A tool can follow every standing rule and still produce a reply that reads well and is wrong, which is hallucination. The answer is a draft, and the person who sends it on is responsible for it.
What it means for a Business
They are taught as one of the six ways to personalise a tool, so that everyday requests need no set-up at all, and the same sessions include a prompt that writes both your custom instructions and your memory for you. On the one-to-one programme they are part of the first sessions, which clear the ground before anything gets built.
Ready to Put This to Work?
Tell us where your team is with AI and we will tell you honestly what would make the biggest difference.