ChatGPT haircut prompt: what to paste, what to check, and where AI try-on goes wrong

Yes, as a brainstorm. ChatGPT is good at turning a description into a shortlist and at explaining haircut vocabulary. What it doesn't do is measure anything. Give it a selfie and it guesses your curl pattern, density and hairline from pixels; describe your hair loosely and it suggests cuts your hair can't hold.

The answer is only as good as the facts you feed it, so feed it facts. The prompt below asks for them hair first and routine second, and it tells the model not to rate your face, which keeps the answer about haircuts.

Copy it, replace every bracket, delete the options that don't apply, and paste it into a new chat. You can paste the same text into other chat assistants, such as Gemini or Claude.

Act as an experienced barber and hairstylist. Suggest five haircuts for me and explain each one.

My hair:
- Curl pattern: [straight / wavy / curly / coily, plus a 1A–4C code if I know it]
- Density: [low / medium / high] (how much scalp shows when it's dry)
- Strand feel: [fine / medium / coarse]
- Hairline: [no recession / higher at the temples / receding / thinning crown / widening part]
- Current length: [buzz / short / medium / long]; how much I'll cut: [keep length / a little / a big change]

My routine:
- Styling time I'll really spend: [time] a day
- Visits I'll keep: every [interval]
- Heat tools: [yes / no]; product: [none / a little / any]
- Context: [conservative job / helmet or hats / gym or swimming / humid climate / none]

Avoid: [things you don't want, e.g. a fringe, a skin fade, a blow-dry every morning]

For each cut, give: why it suits my hair (not my face), the main trade-off, how often it needs a trim, and the exact words to tell my barber or stylist (sides, top length, fringe, neckline). Don't rate my face or looks. If my hair facts rule a cut out, say so instead of suggesting it.

Two lines do most of the work. “Why it suits my hair, not my face” stops the answer sliding into face-shape rules, and “If my hair facts rule a cut out, say so” gives the model permission to cross things off instead of agreeing with everything.

Pick the closest option and say when you're unsure; “wavy, maybe 2B” is more useful than a confident wrong code. The hair type guide walks through checking curl pattern, density and strand thickness at home.

Get the curl pattern right first. In Fitcut's catalog it changes how many cuts are possible at all: compare 2A, with 38 cuts built for it, and 3C, with 10. A chat model that doesn't know your pattern will happily fill the gap with cuts from the wrong row of this table.

For the hairline line, plain words work: “higher at the temples” or “thinning at the crown”. If you know your Norwood stage, add it; the receding hairline guide explains the stages.

Run each suggestion through four questions before you book anything:

1. Does it hold my curl pattern and density? Open the cut's page, for example the textured crop, and read who it suits and who should skip it.
2. Does it work with my hairline? Long, swept-back cuts need a full front.
3. Can I keep up the trims and the styling? Ask the model for the trim interval, then compare it with your calendar.
4. Is the reason about my hair? If every reason is “it balances your face shape”, ask again.

Check the names too. A chat model can blend two cuts into one or use a trend name a barber won't recognise. Ask for each cut “in barber terms: sides, top length, fringe and neckline” and you'll get something you can read out in the chair.

• “Which of these needs the least daily styling for my hair, and what does it trade away?”
• “Write the exact words I'd say to a barber for your second suggestion, with clipper guard numbers.”
• “Which of these would stop working if my hairline moved back?”
• “What would you change for a humid climate, or for wearing a helmet?”
• “Drop anything that needs a blow-dryer and replace it with something that air-dries.”

Ask one follow-up at a time, and paste your hair facts again if the chat gets long, so the model doesn't drift back to generic picks.

Generative image models usually don't paste a haircut onto your photo; they redraw the picture, and every redraw is a chance for your face to shift. In a test write-up updated in June 2026 by AIChangeHair, which sells its own hairstyle tool, Google's Nano Banana image model held likeness well on the first one or two edits, but after about four back-and-forth tweaks the testers saw faces soften and identity drift ^[2].

The same write-up warns that wording matters: get it wrong and the model changes your face, not just your hair. Its fix is simple: re-upload your original selfie and redo the hair change in a single step rather than stacking edits ^[2].

Whatever tool you use, compare the hairline, the forehead and the shape of the head with your original photo. A haircut preview has to redraw those areas, so check them before you trust it. Fitcut's previews come from an image model (Google's Gemini image model, with FLUX as a fallback) that is told to change only the hair. Each preview starts again from your original photo, so edits never stack. Before you see a render, it is checked against your photo for drift around the eyes, nose and mouth, for a shift in skin tone and, on cuts without a fringe, for a moved hairline. A render that fails is redone or rejected.

Further than arm's length. A 2018 study in JAMA Facial Plastic Surgery modelled selfie distortion and found that at 12 in from the face, selfies make the nose look larger relative to the width of the face, by 30% in men and 29% in women compared with a distortion-free projection, while a photo from 5 ft, a standard portrait distance, shows essentially no difference ^[1].

That matters for any tool that reads proportions from a photo, and for a chat model guessing from one. For a fair photo:

• Have someone else take it, or prop the phone up across the room and use the timer.
• Keep the camera at eye level, your face straight on and your expression neutral.
• Pull your hair back for proportions, then take a second photo with your hair as you wear it, for texture.
• Use even light from the front, with no hat or sunglasses.

These are three different kinds of tool, and they're good at different things. Fitcut is a measurement-based recommender, so it sits in the last column, and it's our own product. For named apps, prices and published ratings, see the AI haircut apps comparison; we don't repeat that list here.

• You don't know your curl pattern or density. The model accepts whatever you type, so a wrong guess flows straight into the picks.
• Your hairline is changing. The right cut depends on the stage, and a chat model can't see it reliably. Start with the receding hairline guide.
• You have tight curls or coils. If the suggestions read like straight-hair cuts, they probably are; the coily hair guide lists cuts built for 4A to 4C.
• You want to see the cut on your own face across several versions. That's exactly where drift adds up.
• You want your proportions measured, not described. The face shape detector measures ratios from a photo instead of guessing a label.

In those cases, Fitcut's free analysis reads your hair from a photo and asks you to confirm it, measures your proportions on your device and shows your top match with its main reason; members see the full ranked list. You can check its picks the same way you'd check a chat answer.

Read the tool's privacy notice before you upload. Look for where the photo is processed, how long it's kept, whether it's used to train models and whether you can delete it yourself. For a chat assistant, that means the provider's policy and your account's data settings.

For Fitcut, the privacy notice puts it in plain words: “Your face is measured on your device.” and “Your selfie is never stored at rest. At most, it's held encrypted for 24 hours while your previews render.” Its list of what Fitcut never does includes “Rate your face or looks.” and “Train models on your photos, unless you give separate, opt-in consent.”