Why does ChatGPT rate everyone a 7 or 8?
Three things push chat models toward the same few numbers. First, they are trained to be helpful and agreeable, and a low score feels unhelpful or unkind, so they drift upward. Second, a 10-point scale with no stated reference set has no anchor: the model is not comparing your photo with a defined group, so “7” just means “pleasant”. Third, the answer is generated, not computed, so it settles on common, safe-sounding values. The result is a narrow band of scores that does not separate one photo from another very well. A useful scale needs an anchor. FaceTwin scores five measured proportions and then reports where you sit among 1,196 reference faces, whose median is 7.2, with the 10th percentile at 6.4 and the 90th at 8.0. Notice that its median is also a seven. The difference is that you can see the distribution behind it.
- Chat scores are generated, not computed
- No reference set means no anchor
- FaceTwin: median 7.2, p10 6.4, p90 8.0
- 1,196 reference faces
How to use it
- Upload a selfie or take one: a front-facing, well-lit photo of yourself, no glasses, neutral expression.
- Wait a few seconds: the face model finds 68 landmarks in your browser; the photo is not uploaded.
- Read your 1–10 score and the six explained measurements.
- Try two more photos and average them for a steadier result.
Related
Why chat models drift toward 7 and 8
Language models learn from human feedback that rewards answers people rate as helpful and kind. Telling someone their face is a 4 rarely earns that reward, so the polite middle-to-high band becomes the safest output. The effect is stronger when the prompt is casual, and weaker when you ask for specifics.
There is also a statistical reason. A “7” is a common value in everyday ratings, so the model has seen it used as a friendly default. Without a defined population to compare against, it has no way to say how rare a feature is, only that it sounds good.
A scale needs a reference set
A score only means something relative to something. In FaceTwin’s case the reference set is 1,196 frontal photos of public figures (540 women, 646 men), measured with the same formula you use. Your score is then placed on that distribution.
| Reference point | FaceTwin 1–10 score |
|---|---|
| 10th percentile | 6.4 |
| Median | 7.2 |
| 90th percentile | 8.0 |
That the middle is 7.2 tells you something specific: a score of 7 is ordinary within that set, and the gap between the 10th and 90th percentile is only about 1.6 points. Small differences in a score are therefore small differences in proportions, not large differences in a person. The set is public figures photographed under good conditions, so it is not a population sample, and we say so.
How to get a more useful number from ChatGPT
- Ask for a percentile or a comparison with a described group instead of a bare score, and ask it to state its assumptions.
- Ask it to critique the photo (angle, lens distance, light) before any rating.
- Request a range with uncertainty and the three things that would move it most.
- Run the same photo in a new chat and a different photo of yourself. If the scores barely move across very different photos, the number carries little information.
- Switch the goal from a number to descriptions and tips; those survive the “polite inflation” better.
Measured scoring in one paragraph
FaceTwin places 68 landmarks in your browser and scores symmetry, seven golden-ratio checks, facial thirds, canthal tilt and jaw angle with fixed weights. It then reports the percentile against the reference set. The photo is not uploaded and the same photo gives the same score. That makes it better for comparing two haircuts or two angles, which is the only question a score can answer reliably. It is entertainment, for adults, on your own photo, and not a measure of worth.
Frequently asked questions
Is it true ChatGPT rates everyone an 8?
Many people report scores clustering in the 7 to 8 range. We have not measured ChatGPT, so treat that as a common experience rather than a statistic. The explanation above is why it is plausible.
Why is my FaceTwin score also around 7?
Because the middle of our reference set is 7.2. A 7 means ordinary within that set, not impressive and not poor.
Can I make ChatGPT more critical?
You can ask for blunt, specific feedback, but a harsher tone does not make the number a measurement. Ask for reasons and a range.
Are low scores more accurate?
No. A score is only as good as its method and reference set, whether high or low.
Which should I trust, ChatGPT or FaceTwin?
For tips and tone, ChatGPT. For a repeatable proportions score with a stated reference set, FaceTwin. Neither measures attractiveness itself.
Is a 7.2 good?
It is the median of the reference set. Scores are entertainment, not a measure of worth.
Is my photo uploaded?
Not for local lookalikes or face analysis: the model runs in your browser, so your selfie, landmarks and descriptor stay on your device. AI portraits, including the first free one, and full reports send your photo through our server to the AI provider after a free first-portrait request or a credit purchase, and separate upload consent. Reports also send your numeric measurements. FaceTwin does not save the original photo; it is discarded from request memory after processing.
More from the same selfie
Related by topic
Asking AI to rate your face
The data behind the scores: FaceTwin facts · face shape statistics
Last updated 2026-10-18. FaceTwin is for entertainment and is not identity recognition, medical or cosmetic advice.