🚫 No AI Slop • This article is human written and researched
AI Attractiveness Score Distribution
Last updated: September 12, 2026
Distribution of AI attractiveness ratings (98,371 ratings)
Lookrank.com
This chart summarizes 98,371 anonymized ratings from the Lookrank looks rating, as of 12 September 2026.
The dashed line is a theoretical population curve centered at 5.0. The bars are Lookrank results, not a random sample of unique people.
See the Real World Outcome: How these statistics impact your career, dating, and life.
Data Privacy & Ethics
We do not use submitted photos to train models, and we do not retain images or facial geometry for this aggregate. The chart is built from 98,371 anonymized scores assigned by the active Lookrank model.
Our primary application database is hosted in the European Union. The service is operated by Intelligent Vision LLC (Wyoming, USA); see our Privacy Policy for the full picture.
We Don't Sell Your Data
Our business is this tool. We're not a data broker. No trackers, no ad cookies, no creepy stuff.
Methodology and Limits
What this dataset contains
98,371 anonymized Lookrank scores as of 12 September 2026, on the 1–10 scale reported to one decimal. Photos and facial geometry are not included. Ten results sit below 1.0.
What the rows are
Each row is a score, not a proven unique person. Repeated analyses by the same person can appear. This is a self-selected product sample, not a population survey.
How to read accuracy figures
The comparison figures below are internal estimates against community votes. Sample size, date range, and exclusions for that comparison are not published here, so they are not independent validation.
Average Score
Weighted mean 5.94. Median 5.9. Population standard deviation 1.30.
Most Common Score
The mode, with 3,195 ratings at exactly 5.2. That is lower than both the mean and the median.
The 9+ Club
456 scores were 9.0 or above. None reached 10.0. The highest recorded score was 9.9 (2 scores).
Upper tail
90.8% scored 7.7 or lower. 9.2% scored 7.8 or higher. 1.7% landed exactly on 7.7, so there is no sharp “top 10%” line.
Why isn't the top 50% exactly 5?
- Self-Selection: People who choose to use a facial-rating tool are not a random population sample, so the mean and median of these scores do not need to equal 5.
- Photo Conditions: Lighting, angle, expression, camera distance, and image quality can move an individual result and influence the aggregate distribution.
- Model Calibration: The scale reflects the active Lookrank model version. Future model updates can shift how scores are distributed.
The rating scale we use
Each score on this chart sits on Lookrank’s published 1–10 scale. See what the bands mean, from rare through extreme, using these 98,371 ratings.
See the 1–10 scaleInternal Accuracy Comparison
Error Margin
This is Lookrank's typical estimated range on the current model, not a guarantee for every photo or person.
Mean Absolute Error (MAE)
In our internal comparison, the AI score differed from community votes by 0.3 points on average.
Worst-Decile MAE
Within the 10% of comparisons with the largest gaps, the average difference was 0.96 points.
Bias and Consistency Notes
⚖️ Gender Bias
Within this internal comparison, we found no significant difference between female and male score gaps. This result does not establish equal performance across all demographic groups.
🤖 Model Consistency and 👥 Human Preference
A model applies the same scoring process repeatedly, while community votes reflect individual and cultural preferences. Consistency does not make an AI score objective or bias-free.
Training data, calibration, demographics, image quality, and background can influence model outputs. Treat the result as a photo-based estimate rather than a universal judgment.
Challenges & Perception Gap
The "Harshness" Paradox
We often receive feedback that our AI is "too harsh," yet our data shows it actually rates users slightly higher than human voters on average. This discrepancy stems from the "Better-Than-Average Effect," where individuals consistently overestimate their own attractiveness.
Self-Rating Bias
Surveys show that both men and women most commonly rate themselves a 7 out of 10 [10]. A lower automated estimate can therefore feel like a penalty even though the model is applying a different reference scale.
High Consensus
Despite individual taste differences, groups can show measurable agreement on attractiveness. Published research [11] and our community data both show areas of consensus, but neither creates a universal standard for an individual face.
The "Averageness" Advantage
This sample clusters around 5.0–7.0, which matches published work on facial averageness [2]. The Lookrank estimate often lands higher on more typical, balanced faces and lower on more atypical ones. That is a model tendency, not a rule about human worth.
Why the upper tail is not a sharp 10% cut
Because scores are reported to one decimal, many results tie. 90.8% of this sample scored 7.7 or lower, while 9.2% scored 7.8 or higher. Saying a 7.7 “beats 90%” would ignore the 1,683 scores sitting exactly on 7.7. The drop-off above 8.0 is still steep: only 0.46% reached 9.0 or above.
Skin Quality vs. Bone Structure
While bone structure is hard to change, our data supports findings that skin homogeneity (evenness) is a major variable. Improvement in skin texture alone can boost scores [7]. This is a likely factor for users who see score variances under different lighting conditions. The AI detects "radiance" signals similar to the human brain [9].
Debunking the Golden Ratio
Lookrank's scoring model does not strictly adhere to the 1.618 "golden ratio" often peddled by basic scanning apps. Modern research finds it a poor predictor of attractiveness [4]. Instead, our attractiveness test prioritizes feature balance, averageness, and skin health, which correlates better with human perception [2]. You can inspect your facial thirds and symmetry directly on our facial measurements tool.
The "Self-Selection" Average (5.94)
Above Average Baseline:
The Lookrank user average of 5.94 is higher than the theoretical population mean of 5.0. This correlates with the concept that people interested in facial analysis already invest in their appearance (grooming, skincare) [8], pushing the baseline up.
Neural Efficiency:
Highly attractive faces are processed with less "neural effort" [9]. Our AI mirrors this: consistent, clear features are easier to "score" with high confidence, while unique or asymmetric features introduce more variability, often resulting in lower, more conservative scores.
References
How Lookrank rates you:
The free Lookrank score is a photo-based 1–10 estimate from an automated analysis. When the reference scale image loads, it is attached so the model can place the result on Lookrank’s 1–10 scale. Community votes on public profiles are a separate human comparison, not ground truth.
What you receive
One overall score, a short written assessment, and a result page. A longer paid report is optional. Processing usually takes tens of seconds. Model version and photo conditions can change the number.
What this is not
It is not a universal geometric formula, a landmark-count measurement, or a prediction of how any specific person will rate you. Consistency across the same photo is not the same as objective attractiveness.
How attractive are you?
Get your 1–10 ratingThe Real World Outcome
Data is just numbers until it hits the real world. Learn how these statistical distributions translate into actual advantages in your career, dating life, and legal outcomes.
Explore the Impact StudyEdited by Alex Alstrom · Questions: [email protected]