How to Get ChatGPT to Pick Your Outfit (Actually Wearable Version)
Most people ask ChatGPT to pick an outfit exactly once. They type something like "what should I wear today", get a paragraph describing a white shirt and tailored trousers they do not own, and conclude the whole idea is nonsense.
The idea is not nonsense. The question was. "What should I wear today" contains no information: not what you own, not where you are going, not what the weather is, not what you are willing to put on. ChatGPT answered the only question it was actually asked, which was "describe a generically acceptable outfit", and it answered it correctly.
This is the method that produces an outfit you will genuinely wear. It takes about four minutes the first time and thirty seconds every time after.
The short version
A wearable answer needs six inputs: the clothes you own, where you are going, the weather, the formality level, your hard limits, and what you wore recently. Give ChatGPT all six and forbid it from naming anything outside your list. Give it fewer than four and you will get a magazine outfit. The single highest leverage instruction is the one that says unlisted items do not exist.
Why "what should I wear today" fails
Three things go wrong at once, and it is worth separating them because they have different fixes.
No inventory. ChatGPT has never seen your clothes. Without a list, it is describing an outfit in the abstract.
No constraints. With nothing to work against, it defaults to the statistical centre of everything it has read about getting dressed, which is a white shirt, dark trousers and a neutral shoe. That answer is not wrong. It is just not about you.
No stakes. "Today" could be a funeral or a hardware store. Formality is the variable that changes the answer most and it is the one people omit most often.
Fix all three and the same model produces something completely different.
The six variables that decide the answer
Every good outfit prompt is these six things in some order. You do not need elaborate wording. You need the information present.
1. The clothes. A list of what is available. Not your whole wardrobe necessarily. For a single outfit, the twenty five to forty items you would actually consider today is plenty and works better than two hundred, because a shorter list is a list the model can hold entirely in view.
If you would rather solve this once for a whole season than every morning, The ChatGPT Capsule Wardrobe Prompt does the same work in bulk.
2. The occasion. Where you are going and who will see you. "Office" is weak. "Office, but I have a client meeting at two and drinks after" is strong, because it tells the model the outfit has to work across three registers.
3. The weather. Temperature range, not season. "12 to 19 degrees, windy, no rain" beats "winter". Include whether you will be outdoors and for how long, because that changes the layer, not the outfit.
4. The formality target. Give it a number. One to five, where one is you are not leaving the house and five is a wedding. Numbers work far better than adjectives, because "smart casual" means eleven different things and three is three.
5. Your hard limits. The things you will not wear regardless of how well they work. Nothing cropped. No heels above four centimetres. Never a tuck. Arms covered. This list is short and it is the difference between an outfit that is correct and an outfit you will actually put on.
6. What you wore recently. Two or three days is enough. Without it, you will be handed your most reliable combination repeatedly, because it is genuinely the best answer and nothing is stopping the model from giving you the best answer twice.
The prompt
Paste your item list first, then this underneath in the same message.
Pick what I wear today from the list above.
RULE: You may only use items on that list. If something is not on
the list it does not exist. Do not name it, do not suggest it, do
not build an outfit that needs it. If the list cannot produce a
workable outfit, say so plainly and tell me which slot is empty.
TODAY
- Going: [where, who is there, how long, any change of setting]
- Weather: [X to Y degrees, wind, rain, time outdoors]
- Formality target: [1 to 5]
- Hard limits: [things you will not wear]
- Worn in the last 3 days: [items]
GIVE ME
1. One outfit. List each piece using the exact wording from my
list. Nothing else in this section.
2. Two lines on why it works, referring to the specific pieces.
3. One swap: which single item to change if it turns out warmer
or more formal than expected, and what to change it to.
4. A second complete outfit that is meaningfully different from
the first, not a variation of it.
Do not describe my body. Do not use the words elevated, effortless,
timeless or versatile. Plain language.The banned words at the end are not a joke. Those four words are the fingerprint of generic styling content, and forbidding them measurably pushes the output towards concrete description. "The wool trousers because they hold a line without needing a heel" is useful. "An effortless, elevated look" is not.
Working the answer
The first answer is a draft. The value is in the next two messages, and these are the ones people skip.
If it is close but wrong in one place: Keep everything except the shoes. Give me three alternatives from my list and say what each one changes about the outfit.
If it is too much: Take one thing away. Which piece is doing the least work?
If it feels like someone else: This is not me. It reads as [too corporate / too styled / too casual]. Same occasion, same list, different register.
If you want to know whether it is actually right: What is the weakest part of this outfit and what would someone notice first?
That last one is the most useful prompt on this page. ChatGPT will defend an outfit indefinitely if you let it, and asking it to attack its own suggestion produces genuinely better information than asking it to explain why the suggestion was good.
The checks that catch a bad answer
Three quick tests, worth running the first few times until you know how your setup behaves.
The invention check. Read the outfit and confirm every single item appeared on your list, word for word. Invented items are usually the most ordinary ones, which is exactly why they slip past. A plain white tee. A denim jacket. A simple gold hoop.
The omission check. Ask: List every item from my list you did not consider. If that list is long, your inventory was too big for the model to hold, and the outfit was chosen from a fraction of your wardrobe. Shorten the list and ask again.
The repeat check. After a week, ask: Across everything you have suggested in this conversation, which items have you used more than twice, and which have you never used? The never used group is where the interesting outfits are hiding, and it is the starting point for The ChatGPT Wardrobe Audit.
Making it work more than once
Everything above works in a single conversation. The problem starts on day two, when you open a new chat and none of it exists any more.
There are three levels of fix, in increasing order of effort and effectiveness.
Level one: keep one long conversation. Do all your outfit questions in the same chat. Your list stays visible for a while and the model builds up a sense of what you rejected. This works for about a week before the early messages start falling out of view and the model begins forgetting the trousers.
Level two: put your rules into memory or a custom GPT. Your hard limits, your formality vocabulary, your banned words and your general context are stable facts and they store well. Remember that I never wear anything cropped, that heels above four centimetres are out, and that I dress for a cold open plan office. This removes about half of the typing. It does not store your wardrobe, and trying to make it do so is the mistake covered in Why ChatGPT Forgets Your Wardrobe.
Level three: give it a wardrobe it can look up. Instead of the model holding your list in its head, the list lives somewhere structured and ChatGPT queries it. This is what a connected wardrobe does, and it changes the shape of the interaction completely: the six variables reduce to one line, because five of the six are already known.
With Modebase connected, the four minute setup above becomes: Cold and windy today, client meeting at two, drinks after. What am I wearing? The wardrobe, the wear history and the hard limits are already there. What comes back is built from the pieces you actually own, and it knows you wore the charcoal trousers on Monday.
A note on photos
You can photograph an outfit you are already wearing and ask ChatGPT to assess it, and this works well. It is genuinely good at spotting a proportion problem or a formality mismatch in an image.
What does not work is photographing your entire wardrobe and expecting the model to plan from it. Images consume far more of the available context than text does, only a limited number can be considered at once, and none of them persist into the next conversation. Photos are for judging one outfit, not for describing a closet. The full comparison of methods is in Four Ways to Tell ChatGPT What Is In Your Closet.
Frequently asked questions
How do I get ChatGPT to only use clothes I actually own?
Put a hard rule at the top of the message stating that items not on your list do not exist, and instruct it to report an empty slot rather than fill it. Restate that rule at the top of any follow up message that matters, because instructions given early in a long conversation lose force as the conversation grows.
Why does ChatGPT keep suggesting the same outfit?
Because it is the best answer and nothing is telling it not to repeat. Include what you wore in the last three days in every prompt, or ask it directly which items it has never used and to build an outfit from those.
Can ChatGPT pick an outfit from a photo of my closet?
It can identify a reasonable number of items from a photo, but a full closet shot is unreliable because garments overlap and colours read wrongly under wardrobe lighting. Photographing individual pieces works better and still does not persist between chats. For a single outfit, a typed list of the items you would consider today is faster and more accurate.
What is the best ChatGPT prompt for what to wear?
The one that includes all six variables: your available clothes, the occasion, the weather as a temperature range, a formality number from one to five, your hard limits, and what you wore recently. Missing any of the first four is what produces the generic white shirt answer.
How many clothes should I list for a single outfit question?
Twenty five to forty items. That is enough for real variety and short enough that the model can hold the whole list in view. Pasting two hundred items to pick one outfit makes the answer worse, not better, because parts of the list stop being visible.