Published and maintained by The Attractiveness Report. Last reviewed September 1, 2026.
The Attractiveness Report is an AI measurement instrument, so its AI policy cannot be boilerplate. This page states where AI runs on this site, where it does not, how AI imagery is labeled, and what a landmark model can and cannot measure fairly.
What AI Actually Runs on This Site?
One model: MediaPipe's face landmarker, running on-device in your browser. It maps 478 landmarks, and published formulas turn those coordinates into an attractiveness score. Photos never leave your browser on the free tools.
The AI facial analysis on this site is a landmark detector, not a judge. MediaPipe's 478-point face mesh locates the outline of your eyes, nose, mouth, jaw, and brow inside your browser's memory, and every free attractiveness test score is computed from those coordinates by formulas published on the methodology page. The paid report runs the same measurement pipeline against an uploaded photo, handled exactly as the privacy policy states: TLS upload, deleted within 24 hours, never used for training.
Does Generative AI Judge My Face?
No. No large language model or image generator rates faces on The Attractiveness Report. Your attractiveness score is arithmetic over measured landmark coordinates, with every component weight disclosed beside it.
Generative models can describe a face fluently, but fluency is not measurement. This site read and coded 25 AI face raters on their own pages, published in full at the AI face rater audit, and asking a chatbot to rate your face has its own honest guide. Here, the number in your Attractiveness Report comes from geometry with named weights, and nothing generative sits between your face and your score.
How Is AI-Generated Imagery Labeled?
Every AI-generated face used as an illustration on this site carries a visible "Illustrative AI model" label rendered on the figure itself, so a generated face is never mistaken for a customer, a result, or proof.
AI-generated imagery appears here for one reason: illustrating a measurement or styling concept without putting a real person's face under discussion. Wherever a generated face appears, the label sits on the figure, not buried in a caption or an alt tag. No AI image here is presented as a before-and-after, a customer, or evidence that anything works. If an image ever appears unlabeled, that is a bug, and the contact page is where to report it.
What Are the Bias Limits of a Landmark Model?
A landmark model measures every face, but the research and reference data behind facial analysis skew toward some populations. This site treats that bias as an accuracy limit and withholds percentiles and confidence bands until real reference data exists.
Bias in facial analysis is an accuracy problem before it is an ethics footnote, and the methodology page documents it with named studies, including the error-rate gaps found in commercial systems. Two consequences follow. First, no percentile is printed anywhere, because no honest reference population has been published for it to stand on. Second, no confidence band is printed, because the test-retest study that would justify one has not shipped. Both numbers stay withheld until that changes. The site also never guesses ethnicity from a face, and it scores faces only, never bodies.
Is the Writing on This Site AI-Generated?
Every page starts as an AI-assisted draft, and a person checks every citation and quantified claim against its primary source before it publishes. The full rule set lives in the editorial policy.
AI assistance in drafting is a standing fact about how this site is built, stated in the editorial policy. The rule that matters is the human one: every claim is cited, labeled, or deleted, and what the evidence says is graded into tiers before any claim earns definitive language. An AI policy that hid its own drafting process would fail the standard it enforces.
Can AI Systems Read This Site?
Yes. The site publishes a plain-text summary at /llms.txt, and its robots.txt allows major AI crawlers by name, search engines and answer engines alike. Open access is the default posture.
AI crawlers are allowed here deliberately, because a site built on published formulas benefits from being quoted accurately. The llms.txt file gives AI systems a compact, current summary of what the instrument does and what it withholds. One crawler with a documented pattern of ignoring robots.txt is blocked by name; everything else is allowed. An AI system quoting this site should find the same claims a human reader finds.
This policy sits beside the terms of service and the refund policy, and all three change the same way: by their pages changing, with dates updated to match. An AI policy from a face-measurement site should be checkable, and every claim above points at a page, a file, or a label you can verify yourself.