A short document that answers most privacy questions

A model card is a short document stating what a model was trained on, how it was evaluated and what it is meant for, and an honest rating tool would publish one. The format comes from Margaret Mitchell and colleagues (2018), who proposed cards disclosing "the context in which models are intended to be used" alongside evaluation details.

A card exists because a model on its own is a black box, and it is the minimum a builder can publish to make that box legible from outside.

What a proper one covers

A useful card names the training data sources at least in category - scraped images, licensed datasets, user uploads, synthetic generation - without necessarily naming every source. It states the evaluation method and what the reported numbers actually measured. It lists known limitations, which is the section most likely to be thin or missing, because limitations are the part least flattering to publish. And it states an intended use, which is the sentence that tells you whether the tool you are using was built for the purpose you are using it for.

Why almost no consumer rating tool has one

Model cards are common in research releases and increasingly common for enterprise AI products, and close to absent from consumer-facing scoring tools, because a card is a research-culture artefact and most rating tools are products first. Publishing one also means publishing a limitations section, which is a harder sell to a product team than a privacy policy is, since a privacy policy protects the company and a limitations section undercuts the pitch. That asymmetry is most of the reason the practice has not spread here. Regulation now requires a version of this for the largest models. Since 2 August 2025, the EU's obligations for providers of general-purpose AI models have applied, including a public summary of training content. A rating tool built on such a model inherits that disclosure for the base model, not a card for its own scoring layer, which is the part a user actually meets.

What its absence tells you

An absence is not proof of anything specific, but it removes a source of reassurance that would otherwise exist. What "used to improve the model" tends to leave out is exactly the kind of gap a model card would close, since intended use and data provenance are supposed to answer it directly. Where scoring labels come from is the deeper mechanical version of the same question, for anyone who wants to know what a card is even describing.

None of the four satellite sites is a model card, and none claims to be one. Rate Cock publishes its own policy documents rather than a formal card, which is the more common industry pattern outside research labs. For the human-judged alternative, where the "model" is a person and the intended use is stated up front by definition, Rate Penis's coverage of what a commissioned judgment involves sidesteps the whole category. Penis Rater's guidance on choosing a scoring tool is a reasonable place to weigh a tool's disclosure, or lack of it, against everything else that matters in the choice. A measurement method does not need a model card in the same sense, since Measure My Cock's data page is describing a procedure rather than a trained system, and the disclosure question there looks different.