Looksmaxxing Trends

Am I Pretty? Why AI Face Ratings Overrate You, and What New 2026 Research Shows

Is the score ChatGPT or a face rating app gave you accurate? Not as a number. New research from September and October 2026 shows that popular AI chatbots rate faces much higher than real people do, squeeze everyone into a "slightly above average" band, and that some rating websites score women of different races very differently for the same kind of photo. AI is decent at one thing, though: putting faces in order. Here is what the research says and how to use these tools without fooling yourself.

New research (September and October 2026)

Chatbots rate faces too high. A study posted on September 2, 2026 compared ratings of 102 faces from 2,513 real people with ratings from four popular AI models: ChatGPT, Claude, Gemini, and Grok. On a 1 to 7 scale, the average human rating was 3.02. The average AI rating was 4.71. Not one model ever gave any face the lowest score, and Grok was the most generous at an average of 5.21. The researchers think part of this comes from chatbots being trained to be nice. It is a preprint, so it has not been peer reviewed yet.
But they get the order mostly right. The same study found the AI models matched the human ranking of faces very closely, even more closely than a typical single person did. The lead researcher put it simply: ask an AI to sort 10 faces from least to most attractive and it will look a lot like a human's list. Ask it to rate one face alone, and the agreement disappears.
Rating sites may favor white women. A study published October 21, 2026 in the journal Facial Plastic Surgery ran 120 AI made photos of young women, 30 each from Black, White, Latino, and Asian backgrounds, through five free AI rating websites. With the same pose and neutral face, White women averaged 8.18, Asian women 7.79, Latina women 7.72, and Black women 6.91. The websites also disagreed with each other by almost 2 points for the same faces.
Older research found the same inflation. A 2024 study in a facial plastic surgery journal had doctors in training and five AI websites rate the same 40 faces. Humans averaged 5.0 out of 10. The websites averaged 6.9, and scored 75% of the faces higher than people did.

Why one face can get very different scores

The model was built to please you. General chatbots are tuned to be helpful and polite, so a "7 out of 10" from one of them can be closer to a compliment than a measurement.
The photo matters more than the face. A 2018 study showed that a selfie taken about 12 inches from your face makes the nose look about 30% wider than a photo taken from 5 feet away. Light from the side, a tilted head, or a wide angle lens can move a score a lot with no change to your face.
Different apps, different taste. Each rating model learned from its own set of photos and ratings, often from a small group of raters. The 2026 race study found nearly a 2 point gap between the most and least generous sites.

What people who tried it say

AI said 7 to 8, people said 4 to 6. In June 2026, a 25 year old woman posted her photos for strangers to rate after a chatbot gave her a 7 to 8. Most people put her around 4 to 6, with a few higher. The comments said she looked average to a bit above average and pointed to things like styling and more natural expressions, not her bone structure.
Mixed reports across the board. In a 2026 thread asking people how accurate AI ratings felt, some said the score matched how they see themselves, while others called it inflated, harsh, or different every time they asked.
The £285 report that rated everything. In an October 18, 2026 GQ story, a writer paid about £150 for a detailed AI assisted face report, then £135 more for surgery ideas. He got a 12,000 word report with scores like 73 out of 100 for symmetry, a 16 page plan full of fillers and treatments, and a suggestion to consider a nose job, brow lift, and chin surgery, even though the report said he looked generally good. He also found he had been signed up for a yearly subscription. His takeaway: so much was flagged that none of it felt like it mattered.

Lesser-known points

The researchers who proved AI overrates people sell face reports. The September 2026 study was led by scientists at the same company that sells paid AI assisted face reports. The finding still stands, but it also makes a neat case for paying humans or experts instead of a free chatbot, so keep that in mind.
Use AI to compare, not to score. Since AI is much better at ranking than rating, it is more useful to ask which of three photos looks best than to ask for a number. Even then, use the same light and distance for each photo.
A score can push you toward things you do not need. Many paid apps and reports end with a product, treatment, or subscription. If a number comes with a sales plan, the number is part of the sales plan.
Real people do not agree with each other either. Humans in the 2026 study used the whole scale and often disagreed. Who is rating you matters. A crowd of strangers on a rating board is not the same as the people you would actually date.

Our verdict

An AI face score is a polite guess with a built in bias, not a fact about you. If a chatbot says you are a 7, you are probably not a 7 in the eyes of strangers. If a rating app says you are a 3, that is just as shaky, since the score swings with light, angle, and which app you used. The more useful question is not "how pretty am I" but "which of my photos works best, and for whom." That is something real people answer better than a model. The Face They See lets people of the opposite gender rank your photos side by side for free, which is closer to how dating apps actually judge you.

Related guides

See all guides