The State of Looksmaxxing, October 2026

1. Executive summary

Five top-line findings from the dataset as it stands in October 2026.

• Sample size: 14,734 face scans across 4,814 unique email accounts, of which 8,013 are on the current scoring engine (sufficient for primary reporting).
• Average overall score across the 8,013 scans on the current scoring engine: 49 / 100 (range 29 to 73). This overall figure is the average of the per-metric scores, each placed against the first scans of about 2,100 RealSmile users so that 50 is the median scanner, so an average near 50 is a property of the scale, not a finding about faces. It measures facial geometry; how people rate a face is estimated by the separately validated Face Score. The per-metric figures below are drawn from the 3,000 most recent current-engine scans; metrics the engine does not measure reliably are excluded.
• The most-common weakest metric is Canthal Tilt (10.9% of scored scans). See section 4.
• Total commercial entitlements granted to date: 75 across 6 distinct SKUs. See section 6 for the breakdown.
• Rescans observed: 3049 unique emails have submitted more than one scan to date. Section 5 explains why no rescan score-lift figure is published.

2. How we collected the data

All data are aggregated from face scans submitted voluntarily to the public looksmaxxing test at realsmile.online/looksmaxxing-test. Each scan submits a frontal selfie, which is sent to a server-side landmark detection pipeline (see /research for the reference-range basis of each metric). The pipeline returns an overall 0-100 score and per-metric scores across the 17-metric framework. Scans are stored with an email address (for unlock delivery) and an array of metric objects.

For this report we read only aggregate counts and metric arrays. No emails, photos, or personally identifying information left the database. Aggregation ran on October 7, 2026.

The peer-reviewed studies that anchor each metric (Willis & Todorov 2006, Penton-Voak 2001, Rhodes 2006, Hamermesh & Biddle 1994, and others) are indexed at /research-base. When our internal sample is too thin for a claim, we cite the study rather than report an internal number.

3. Overall score distribution

Overall scores from the 8,013 face scans on the current scoring engine, binned in tens. Scores from earlier engine versions sit on a different scale and are not mixed in.

Reading note: bins are inclusive of the lower bound and exclusive of the upper. The 60-69 bin therefore contains scores 60 through 69. Empty bins represent zero scans in that range, not missing data.

4. Most common weakest and strongest metrics

Per-scan, the lowest-scoring metric is taken as that scan's weakest; the highest as its strongest. Counts below are how often each metric appeared in those positions across 3,000 scans with complete metric arrays.

Per-metric average score (ascending)

5. Cohort variance and rescan score lift

3,049 unique emails have submitted more than one scan. We do not publish a rescan score-lift figure: the same face moves several points between two photos taken minutes apart, so a difference between scans cannot be attributed to a real change. For the published-research view of typical protocol timelines (mewing, skincare, hairline, body composition), see looksmaxxing for beginners.

Cohort source attribution (dating vs LinkedIn vs looksmax) is not yet stored on the Scan row. Sprint 580 (planned) will add a source enum populated from the entry page so future reports can publish per-cohort score baselines.

6. What users buy after a scan

Across all entitlements granted to date (75), the SKU split is:

7. Limitations

• Self-selection bias. The sample is users who chose to take a public face-rating test. Curiosity about one's own appearance is a non-random recruitment filter.
• Sample size. 8,013 scans on the current scoring engine is above our 1,000-scan primary-reporting threshold.
• No demographics. Age and geography are not stored on the Scan row, and gender only from July 30, 2026, so this report does not attempt subgroup analysis.
• Samples. Score figures use the 8,013 scans on the current scoring engine; per-metric sections use the most recent 3,000 of them. Metrics the engine does not measure reliably are excluded, and weakest/strongest metrics are ranked against the population on each metric rather than by raw score.
• Score is one of many measures. The 17-metric framework is anthropometric. It does not measure grooming, style, expression, or context, which the published literature (Willis & Todorov 2006; Ekman FACS) shows materially affect real-world judgments.

8. What journalists should cite

Copy the block below verbatim. We license the aggregates under CC BY 4.0, which means you can reuse them in articles, charts, and reports as long as you attribute RealSmile.

Source: RealSmile (realsmile.online) State of Looksmaxxing 2026 report. Aggregates derived from 14734 face scans submitted to realsmile.online between May 19, 2026 and October 7, 2026. Data through October 7, 2026. Methodology: https://realsmile.online/state-of-looksmaxxing-2026#methodology. License: CC BY 4.0.

Need a chart embed? See section 9 for instructions. Need a quote or a custom cut of the data? Email hello@realsmile.online.

9. Embed and re-use

We do not yet ship a dedicated iframe endpoint. Until we do, the simplest embed is to deep-link an anchored section. Examples:

• Score distribution chart: https://realsmile.online/state-of-looksmaxxing-2026#distribution
• Weakest-metric table: https://realsmile.online/state-of-looksmaxxing-2026#metrics
• Entitlement breakdown: https://realsmile.online/state-of-looksmaxxing-2026#entitlements

For a downloadable PDF, use your browser's Print menu (Cmd+P on macOS, Ctrl+P on Windows) and choose Save as PDF. The print stylesheet on this page renders cleanly in black on white.