How to add your AI profile to ChatGPT

The AI profile is the Markdown file that comes with your analysis. It is plain text, 29 KB for the sample analysis, and it opens with six rules for the assistant that reads it. Then come your season with your best white, power color and metal, a verdict and a reason for all fifty colors, your body shape and line, the cuts, materials and patterns that work and the ones to avoid, and the body answers you gave us, the height and weight you entered included. ChatGPT takes it as an ordinary text file, and OpenAI caps a text file at 2 million tokens, so size is never the problem. The job is to put the file somewhere ChatGPT keeps seeing it, so you set this up once instead of attaching a file to every conversation.

Three ways to give ChatGPT the file

OpenAI's own rule of thumb is that a project is the better fit when you expect to come back to the work, and a regular chat is enough for something quick. For this file you will come back, which is why this guide builds a project.

The screenshots come from the ChatGPT phone app and were taken in September 2026. The menu names, file limits and settings below were checked against OpenAI's help pages in October 2026.

Before you start

The file is on your analysis page under Connect to AI. Press Download .md and the card above the preview shows the file's name and size. The name is wearly- plus your season. For the sample analysis, a Deep Autumn, it reads wearly-deep-autumn.md, 29 KB. Our screenshots show a copy called wearly-labs-profile.md, so yours will look different. If you are on a phone and the download is awkward, use Copy instead and skip to the shortcut at the bottom.

Two things to check when it arrives. The name should end in .md; if your phone or browser saved it as .md.txt, rename it back before you upload. And the size should be tens of kilobytes, not zero.

Step 1: find Projects

In the ChatGPT phone app, open the menu from the top left. The list holds Images, Library, Projects, Plugins, Scheduled and Remote. Pick Projects. On the web, Projects is in the sidebar, with New project to start one (OpenAI Help Center).

A project keeps related chats, files and instructions together, so ChatGPT can use the same context in every chat inside it. It also follows your account: OpenAI lists starting on one device and continuing on another among its uses, so a project you build on a laptop is there on your phone.

Step 2: make the project

Create a new project and give it a name you will recognize in six months. "Wearly Labs" is what we use in the screenshots. "My style" or "Wardrobe" works as well.

If it offers a memory setting, pick Project-only memory. In that mode your saved memories are not referenced in the project's chats, and the chats cannot see conversations from outside the project, so nothing you said elsewhere competes with your profile. Default memory lets ChatGPT draw on both. You can switch later under the more options menu, then Project settings, then Memory, and a change can take a few hours to apply (OpenAI Help Center).

Step 3: add the file as a Source

Open the project and switch from Chats to Sources. OpenAI's docs describe a project the same way: a Chats section and a Sources section for uploaded files.

An empty project says "Give ChatGPT more context. Upload Sources, link drives, or connect apps like Slack to give ChatGPT deeper context about your project." Press Add, then pick your .md file from your files.

It comes back listed under its file name with Markdown under it. That listing is the confirmation: the file is attached to the project, not to a conversation, so it survives every new chat you start in here.

Step 4: tell it what the file is for

The file already tells an assistant how to read it. Its six rules include never suggesting a color from the Not your colors list, always naming where a conditional color goes (below the waist only, or as an accent only), judging a garment that is not in the file against the cut rules, asking when it does not know, and not softening a refusal. What a file cannot carry is how you want ChatGPT to answer, and an order to open it for every question. That is the standing instruction, and the difference between a useful setup and a disappointing one is mostly in this step.

Open the project's more options menu (•••), pick Project settings, and paste the text below as the instructions. Project instructions apply only inside that project and override your global custom instructions (OpenAI Help Center), so they do not collide with the ones you set for everything else. If your app has no such field, paste the text as the first message of your first chat in the project.

You have my personal style profile as a source in this project. Act as my shopping and styling assistant. Whenever I ask what to buy, share a product link, or send a photo, check it against my color palette, my body shape and line, and my recommended cuts before answering. Tell me straight whether it works and why, in one or two sentences. If it does not work, name a specific alternative that does. Never recommend anything from my Not your colors list or my Cuts to avoid. Do not hedge.

Adjust the last line to taste. "Do not hedge" is there because the default register of an assistant is to give you three options and let you decide, which is exactly what you paid not to have to do. The two lists are named as they are in the file. In the file, Wear is "Wear anywhere", Depends is "Yours, but only where the line says" (below the waist only, or as an accent only), and Skip is "Not your colors".

Step 5: test it with something you already know the answer to

Do not start with a real shopping question. Start with checks you can mark yourself.

Ask: What is my season, my best white and my power color?

All three are near the top of the file, under Color season. For the sample analysis the answer is Deep Autumn, ivory and rust. If you get a paragraph about whites in general, the source is not being read. Go back to step 3 and confirm the file is listed under Sources.

Then ask about something you know is a no: a color from your Not your colors list, or a cut from Cuts to avoid. The answer should carry the verdict and the reason your own page gives. These four come from the sample analysis, the real result of co-founder Filip Poláček, a men's Trapezoid with a medium line:

If ChatGPT approves something your page puts under Not your colors, it is answering from general knowledge, not from your file. Check Sources, then the instruction in Project settings.

What to ask once it works

The useful prompts are the specific ones.

• The link check. Paste a product URL. "Does this work for me?"
• The rack. A photo from inside a shop. "Anything here worth trying on?" A rack photo is taken in shop light, so treat a yes as a shortlist. The sample analysis's advice for a Deep Autumn is to hold the piece against your skin in daylight: if it still looks rich and warm rather than washed out, it is yours to take home.
• The pack. "Five pieces for four days in Copenhagen in October, all from my palette."
• The audit. A photo of a shelf in your wardrobe. "What earns its place here?"
• The size question. Expect ChatGPT to ask for a chest, waist or inseam measurement. The file does not carry them and tells the assistant to ask before it recommends a size.

Keeping it current

The profile is a text file, so you can edit it. Open it in any text editor, add a line under a heading that already exists, and re-upload it. Useful additions: sizes, a brand you buy a lot, things you will not wear whatever the analysis says.

Replace the file, do not add a second one. Open the project's Sources, open the old file, delete it, then upload the new one. If you upload a file with the same name, ChatGPT asks whether to upload it anyway or skip it, and uploading anyway adds it next to the old one (OpenAI Help Center). Two versions of your profile in one project is how you get contradictory answers.

If you do not want a project

Attach the file at the start of a conversation (open the tools menu, then Add photos or files) and paste the instruction from step 4 as your first message. It works exactly as well, for exactly that conversation. You will do it again tomorrow.

The other shortcut, and the right one on a phone: press Copy on the Connect to AI screen and paste the whole profile as your first message, with the instruction under it. No file handling at all.

Troubleshooting

ChatGPT says it cannot see the file. Open the project's Sources and confirm the file is listed there and not attached to a single chat, then start a new chat inside the project and ask again.

The answers are generic. Ask it which part of your profile its last answer came from. If it can name the section, the file is being read and the instruction needs to be firmer. If it cannot, it is answering from general knowledge: check Sources, then the instruction in Project settings. If a project keeps ignoring the file, paste the profile as the first message instead, so the text sits in the conversation itself.

The answers contradict the profile. Check the project's memory setting. With default memory, ChatGPT also draws on your saved memories. Switch to Project-only memory under Project settings, then Memory, and allow a few hours for it to apply.

It keeps recommending one color. The profile names a power color and the model latches onto it. Tell it: "I already own three rust things, rotate."

ChatGPT says the upload limit is reached. Free accounts are limited to 3 file uploads a day, and a failed attempt can count toward the cap (OpenAI Help Center). Paste the profile instead, or try again tomorrow.

The download became a .txt file. Rename it back to .md. Nothing inside changed.

Related reading

• Make any AI your personal stylist (the .md trick)