Upload a selfie, ask ChatGPT to rate your face out of ten, get back a 7.2 with a tidy breakdown of your symmetry, jawline, and skin. That's the trend. Searches for it have grown roughly 3,700% and it works exactly as advertised — you will get a number, and the number will look authoritative.
Here is the part worth knowing before you run it: ChatGPT is not measuring anything. It's producing the number that reads as plausible next to your photo and your wording. That isn't a flaw in the prompt. It's what the model is.
This guide covers the viral prompts, a demonstration of why the score is unstable, and six prompts that give you feedback you can actually act on.
The prompts people are actually running
For completeness, since this is what you searched for. The most-shared version:
Rate my face out of 10 based on symmetry, proportions, bone structure, skin quality, and jawline. Be brutally honest. Give me a score, my three strongest features, my three weakest features, and what I could improve.
The "harsh critic" variant:
You are a brutally honest beauty analyst with no interest in sparing my feelings. Analyse this photo and give me an unfiltered rating out of 10, plus a breakdown of every flaw you can identify.
And the structured-report variant:
Act as a facial aesthetics analyst. Analyse this photo and return a report with: overall score /10, facial thirds ratio, facial fifths ratio, canthal tilt, jawline definition, skin quality, and a percentile ranking against the general population.
That third one is the most persuasive and the most misleading. Facial thirds, fifths, and canthal tilt are real terms from facial aesthetics. ChatGPT will happily supply confident measurements for all of them — from a single uncalibrated phone photo, with no reference scale, at an unknown angle, through an unknown lens. Those numbers cannot be derived from that input. They are produced because the prompt asked for numbers.
The "percentile ranking against the general population" is the clearest tell. There is no population dataset being consulted. A percentile is being written because you requested one.
Watch the score move
Don't take my word for it. Run this yourself — it takes four minutes and settles the question permanently.
Test 1 — same photo, three fresh chats. Open three new conversations, upload the identical photo, paste the identical prompt. Scores typically land within about a point and a half of each other. Same face, same photo, same words, different answers.
Test 2 — change one word. Run these in separate chats with the same photo:
Rate my face out of 10.
Be brutally honest and rate my face out of 10.
Be encouraging and rate my face out of 10.
The scores move in the direction you asked for. "Brutally honest" reliably scores lower. You changed the register of the reply, not your face.
Test 3 — change the photo, not the face. Upload the same face in good window light, then in harsh overhead light. The score follows the lighting. Which is almost interesting — it tells you something real about photography — but it is not telling you about your face.
The pattern across all three: the score tracks the prompt and the photo quality, not the person.
What is actually stable
Run the same photo through five times and compare the outputs word by word. The number bounces around. But the descriptive observations — "deep-set eyes", "strong brow ridge", "the light is coming from below and flattening the midface" — stay largely consistent.
That's the useful signal, and it's there because those are observations about visible evidence rather than judgments requiring a scale. Which points directly at how to prompt for something worth having.
Six prompts that return something you can use
The move is to stop asking how good does this face look and start asking what is this photo doing, and what would work better. Craft questions have checkable answers. You take a new photo and find out.
1. The photographer's report — the useful replacement for the rating prompt
Act as a portrait photographer and casting director preparing notes before a shoot. Analyse the face in this photo the way you would write shoot notes — descriptively, not as a judgment of attractiveness. Do not give me a score out of 10 and do not rank me against anyone.
Purpose of the photos: [LinkedIn headshot / acting headshot / dating profile / personal branding]
Return exactly these sections:
1. Face Shape & Proportions — the underlying structure, factually.
2. Strongest Photographic Features — the three that read best on camera, and why they photograph well.
3. What Fights the Camera — what's working against you in this specific photo. Be specific about which is a photo problem versus a fixed feature.
4. Angle & Camera Notes — head angle, chin height, shoulder rotation, eyeline, and the focal length that would suit this face.
5. Lighting Notes — the setup that suits this bone structure, and the one to avoid.
6. Styling Levers — hair shape, facial hair, glasses, and neckline choices that balance the proportions in section 1.
7. The One Change With The Biggest Return — one sentence.
Everything you say must be traceable to something visible in the photo. If you can't see something clearly, say so instead of guessing.
▶ Try this as a fill-in template →
Section 3 is where the value is. "Your midface looks flat because the light is coming from directly overhead" is fixable this afternoon. "6.8/10" is not fixable at all — that's the difference between the two prompts.
2. Pick the best photo from a set
Act as a photo editor selecting a headshot for a client. I'm uploading [N] photos of the same person.
Where this will be used: [LinkedIn / company about page / dating profile / speaker bio]
Impression I want to give: [approachable and senior / creative / warm / authoritative]
Return: the pick with the single strongest reason; a full ranking with a one-line verdict each; a scorecard table scoring every photo on Expression & Eye Contact, Lighting Quality, Crop & Framing, Background, Wardrobe Fit, and Technical Quality as Strong / Workable / Weak; why the runner-up lost; and what's fixable in editing versus what needs a reshoot.
Judge the photos as photos and as fits for the stated purpose. Do not rate the person's appearance.
▶ Try this as a fill-in template →
This is the highest-value use of photo analysis, full stop. You almost certainly have five acceptable photos and no way to choose. The scorecard forces the comparison against your actual use case instead of your gut — and "which of these" is a question with a real answer, unlike "how attractive am I".
3. Diagnose why a photo isn't working
This photo isn't working and I can't tell why. Diagnose it as a photographer would, in order of impact:
1. Lighting — direction, quality, ratio, colour temperature. What specifically is it doing wrong?
2. Angle — camera height relative to my eyeline, my head angle, lens distortion from shooting distance.
3. Expression — what my eyes and mouth are actually doing, and what that reads as.
4. Framing — crop, headroom, how the background interacts with my head.
5. Technical — focus, motion blur, resolution, over-processing.
For each, tell me whether it's a photo problem or a fixed feature. Then give me a shot list to retake it properly: camera height, my distance from the light, head angle, and what to do with my hands and shoulders.
4. Wardrobe and colour, from evidence
Based on my visible skin undertone, hair colour, and eye colour in this photo, recommend:
- Five specific colours that would work well near my face, with the reasoning for each
- Three to avoid and why
- Neckline shapes that suit my neck length and shoulder line
- Whether warm or cool metals suit me better
Base everything on what's visible in the photo. If the white balance looks off and you can't judge undertone reliably, say so and tell me what photo to send instead.
The last line matters. Without it you get a confident answer derived from a colour cast in the photo.
5. Dating profile lineup review
Act as a dating profile consultant. I'm uploading [N] photos in the order I plan to use them.
App: [APP] · Looking for: [GOAL] · Want to attract: [AUDIENCE]
Return: what someone concludes about me from photo 1 alone; what the lineup says about my lifestyle and energy as a set, and how much of that matches what I said I wanted to signal; a photo-by-photo Keep / Recrop / Replace verdict; which standard slots are missing (clear face shot, full-body, social, activity, personality); the order I should actually use; and three specific photos to go take.
Assess the photos as signals, not the person. No attractiveness scores, no comparisons to anyone else, no looksmaxxing terminology.
▶ Try this as a fill-in template →
Note the difference in framing. "Rate my dating photos" gets you graded. "What does this lineup signal, and is it what I intended" gets you an editing decision — and match rates respond to the second one.
6. Hair and grooming direction
Act as a senior hairstylist doing a consultation. My hair texture is [TEXTURE], current length [LENGTH], and I will realistically spend [N] minutes styling per day.
Return: what you can see (face shape, forehead height, jaw width, neck length, hairline, density — only what's visible); three named cuts a stylist would recognise, ranked, each with the maintenance reality and how it looks on a bad hair day; two cuts commonly recommended for my hair type that would work against my proportions; and the exact salon script for your top pick.
Recommend real, nameable haircuts. Do not rate my appearance.
▶ Try this as a fill-in template →
Then preview the recommendation on your own photo with the hairstyle try-on prompts before you book anything.
Two things worth flagging
Your photo goes to a third party. Uploading your face to any AI service means handing that image over. Check your account's data controls and retention settings, and don't upload photos of other people without asking them first. That last one gets skipped constantly.
The trend has a neighbourhood. Searches for the attractiveness test sit right alongside "PSL rating", "blackpill face rating", and looksmaxxing forums — communities that treat AI-generated appearance scores as objective evidence about a person's worth. The tests above show why that's built on nothing: the number moves when you change an adverb.
If a score from a language model is going to affect how you feel about your face, skip the score. Not as a moral position — because the number is generated text, and you'd be reacting to a sampling artifact. The photographic prompts give you more, and they give you something you can go do.
The short version
| Ask | What you get | Worth having? |
|---|---|---|
| "Rate my face out of 10" | A plausible number that moves when you rephrase | No |
| "Facial thirds, canthal tilt, percentile" | Confident measurements the input can't support | No |
| "What's fighting the camera in this photo?" | Lighting, angle, and expression notes you can act on | Yes |
| "Which of these five photos should I use?" | A ranked decision against your actual use case | Yes |
| "What cut suits this face shape and texture?" | Named haircuts plus the salon script | Yes |
ChatGPT is genuinely good at analysing photographs. It's just not good at the thing the viral prompt asks it to do — and the difference between those two is entirely in how you ask.
Related prompts and guides: