Stop Re-explaining Yourself to AI

You open a new chat.

Before you can ask the thing you actually came to ask, you find yourself typing the same introduction again. I'm a freelance consultant. I work with small businesses. I write in British English. Please don't make it sound like LinkedIn. Keep it short.

Then you get a decent answer, close the tab, and next week you do it all over again.

If that sounds familiar, the problem is not the AI you chose. It is that you are using a very capable assistant as if it were a stranger you meet for the first time every morning.

In the last piece on The Working Frame, we looked at ChatGPT, Claude and Gemini and which one belongs in your work. This is the next step. You have picked your tool. Now make it remember who you are.

All three now have the settings to do this. Most people never touch them.

Why it keeps forgetting you

An AI assistant does not know you by default. It knows an enormous amount about the world, but nothing about your clients, your prices, the way you write, what you have already tried or what you are trying to build.

So every time you start from nothing, it fills the gaps with the most average version of whoever might be asking. That is why so much AI writing sounds the same. It is not that the model cannot write well. It is writing for nobody in particular.

The fix is not a cleverer prompt. It is context: giving it the background once, in the right place, so it carries into everything after.

There are three places that context can live. They do different jobs, and the difference matters.

01 / The brief

Think: the thing you would tell a new assistant on their first day.

Every one of the big three has somewhere to tell it about yourself permanently.

In ChatGPT it is Custom Instructions, under Settings then Personalisation. In Claude it is your personal preferences in Settings. In Gemini it sits in the personal context settings, where you can save instructions you want it to follow.

Whatever you write there is read before every conversation. You never have to say it again.

This is the single most useful ten minutes you can spend on AI, and it is the part most people either skip or fill with one line: "be concise".

A good brief is not a CV. It does not need your whole career. It needs the things that would change the answer.

Compare these two.

"I'm an interior designer."

Versus:

"I run a small interior design studio in Manchester. Most of my clients are homeowners renovating period properties, and I also take on a few restaurant fit-outs each year. I write in British English. My tone is friendly and plain-spoken, never salesy. When I ask for writing, give me one strong version, not five options. If something I've said doesn't add up, tell me."

The second one takes a minute longer to write. It will save you hours.

Notice what is in it. Not just what you do, but how you want to be spoken to, what format you want back and permission to disagree with you. That last line does more than people expect. Left to its own devices, AI tends to agree with whatever you bring it. Telling it once that you would rather hear the problem changes the whole relationship.

The three things worth telling it first

If you only write three things in your brief, make them these.

1. What you do and who it is for. Not your job title. Your work. Who your clients or audience are, what you make and what you sell. This is what stops it suggesting Instagram ideas for a business you do not run, or pricing advice for a market you are not in.

2. How you sound. British or American English. Formal or relaxed. Phrases you hate. If there are words that make you wince when you see them in AI writing, list them. It will stop using them. This is the difference between drafts you have to rewrite and drafts you can actually use.

3. How you want it to work with you. Short answers or thorough ones. One option or several. Whether it should ask questions before starting or just make a reasonable assumption and get on with it. Whether you want it to challenge you.

That third one is the most overlooked. Most frustration with AI is not about the quality of the answer. It is about the way it answers: too long, too many bullet points, too many clarifying questions, too much enthusiasm. All of that is adjustable. You just have to say so.

Do not type it. Say it.

The easiest way to write a brief is not to write it at all.

Open the app on your phone, use voice mode or dictation, and talk for three minutes. Tell it what you do, what a normal week looks like, what you are working on, what drives you mad about the drafts you get back. Ramble. Go off on tangents.

Then ask it: "Turn that into a short set of instructions I can save so you remember this about me."

You will get a cleaner brief than you would have written, in a fraction of the time, and it will include things you would never have thought to type. People are much better at explaining themselves out loud than on a blank page.

The same applies to anything you already have. Your About page. A brand guide. A description of your services. The email you send to new clients. Upload it and ask the AI to pull out what it should know about you. You have probably already written most of your brief somewhere. It is just scattered.

02 / Projects

Think: a separate desk for each piece of work.

The brief is about you. Projects are about the work.

A project is a dedicated space with its own instructions and its own files. ChatGPT and Claude both have them, and in ChatGPT they are available on every plan, including the free one. Gemini's closest equivalent is Gems: custom versions of Gemini with their own instructions and files for a specific job.

This is where you put the context that only matters for one thing.

A project for your newsletter, with past issues uploaded so it knows the format. A project for a client, with their brief, their brand guidelines and the notes from your last call. A project for your business admin, with your prices, your terms and the email templates you use.

Every chat you start inside that project already knows all of it.

The instructions you give a project also take priority over your general brief while you are inside it. So your brief can say "keep it short" while your long-form writing project says "go into detail and challenge the structure". Both are true. They just apply in different places.

This is where AI stops being a chatbot and starts behaving like someone who has been working with you for a while. You stop explaining the background and start with the actual question.

A useful rule: if you have explained the same context in three separate chats, it belongs in a project.

03 / Memory

Think: what it picks up as you go.

Memory is the newest of the three, and the most misunderstood.

ChatGPT, Claude and Gemini can all now remember things from your conversations and bring them into new ones. Claude made memory available to everyone, including free users, in March 2026. ChatGPT can keep saved memories and refer back to your past chats. Gemini can use previous conversations to understand ongoing projects and preferences, although it depends on your account and settings, and it does not currently work inside Gems.

It sounds like this should make the brief unnecessary. It does not.

Memory is helpful, but it is passive. It picks up what happens to come up, not necessarily what matters. It might remember that you once asked about a trip to Lisbon and forget that you only ever write in British English. You have much less control over what it decides is important.

So treat the two differently.

Your brief is the thing you decide. Memory is the thing it notices.

The best setup uses both. Put the essentials in your brief, where they are guaranteed to be read every time, and let memory fill in the smaller details as you work.

When you want it to remember something specific, you can simply tell it: "Remember that I don't work on Fridays" or "Remember that my main client this quarter is a skincare brand."

Check what it thinks it knows

This is the part almost nobody does, and it is the most revealing.

Ask it: "What do you know about me?"

You may be surprised. Sometimes it is accurate. Sometimes it has hung on to something from six months ago that is no longer true, or drawn a conclusion from one throwaway question.

You can view and edit what it has stored. In ChatGPT it is under Settings, Personalisation, then Memory. Claude and Gemini have their own memory controls in settings. Correct anything that is wrong and delete anything you would rather it did not use.

Do this every couple of months. Your work changes. Your assistant should keep up.

And if you are switching

If you have spent a year teaching one AI about yourself and want to move to another, you do not have to start again. Ask your current one to write a summary of everything it knows about you, your work and your preferences. Then paste that into the new one's brief.

Claude has made this a built-in feature, with a prompt in its settings designed to bring memories across from other chatbots. But the manual version works with any of them.

What not to tell it

Context makes AI more useful. That does not mean it needs everything.

You do not need to give it passwords, bank details or anything you would not be comfortable seeing in a settings page. Be thoughtful about other people's information too, particularly clients'. Your client's brand guidelines are useful context. Their personal phone number is not.

All three have a temporary or incognito chat option for conversations you do not want remembered. Use it for anything sensitive or one-off. And if you are using a work account, check what your organisation's settings allow before assuming memory works the same way it does on your personal one.

A good test: if you would not put it in a briefing document for a new assistant, it probably does not belong in your AI's memory either.

What changes when it knows you

The difference is not one impressive answer. It is that every answer starts closer to what you need.

You ask "what should I post this week?" and it already knows what you sell, what you posted last month and what your audience responds to.

You ask it to draft a reply to a difficult client and it already knows your tone, your terms and how you usually handle it.

You paste in a messy page of notes and say "turn this into a plan" and it knows the plan is for you, with your hours, your priorities and your way of working.

That is the shift. You stop briefing a tool and start working with an assistant.

It also changes how you use it. When you no longer have to set the scene, you ask more often, for smaller things. The quick email. The second opinion. The "does this make sense?". Those small moments are where most of the value of AI actually lives, and they are exactly the moments you skip when every conversation needs a five-paragraph introduction first.

Where to start today

You do not need to build the whole thing in one go.

Today: write your brief. Three minutes of talking, one request to tidy it up, saved in your settings.

This week: create one project for the piece of work you repeat most. Upload the two or three documents you always end up explaining.

This month: ask it what it knows about you, and correct it.

That is enough to change how it feels to use.

Going further

Once your AI knows who you are, the next question is what it can actually run for you.

That is what the AI Powered Chief of Staff is built around. It takes the brief in this article much further: one document, your Chief of Staff Brief, built by the AI interviewing you while you do most of it by talking. Then eleven working functions run on it: your comms in your voice, your diary and travel, your meetings, what you owe and what you are owed, your projects, research and decisions, and a weekly review that holds it all together. It works in ChatGPT or Claude, and it takes about thirty minutes until it is useful.

If you would rather do it yourself, everything above is a very good start.

Either way, stop introducing yourself.

It should already know.

This reflects the products as they exist in October 2026. Settings move and names change, but the principle does not: give it the context once, in the right place, and keep it up to date.