How scoring works

From tells to a number

Every tell has a weight, set by how strongly it reads as generated, and a saturation point, the number of instances per page at which it counts as fully present. For each tell, severity is half coverage (the share of pages it appears on) and half intensity (instances against the saturation point). Its cost is weight times severity.

The site score is 100 × e^{−cost / 10}, summed over all tells, so one strong tell costs about a quarter of the score and a typical generated template lands near 30. Each category uses the same curve with a divisor of 4. Scores of 90 and above have few generated tells, 50 to 89 have some, and below 50 have many.

The weights are opinions, written down. They come from the ai-deslop rules and have not yet been checked against human ratings of real sites. Read the score as a guide to the findings, not a verdict on the work.

It measures the look, not the maker

Many of these tells started on large, carefully designed sites, then spread through templates and AI models trained on them. A famous site can score low, and a site written with AI help can score high after a careful pass. The score says how much of the common generated look a site carries, which is what a visitor reacts to.

What it cannot see

Text inside images and canvas, pages behind a login, and anything that only appears after a click. It audits desktop width only. It cannot know intent: a condensed capital headline may be the brand, and a texture may be the art direction. That is what the keep rule under every finding is for.

Every tell and its weight