How Bench is built

Where the data comes from

Every video's description already lists the brands tested and a link to each one, so the product list and the Amazon links come straight from what Todd published.

The rankings take more work. Todd announces the results out loud, so we turn each video's audio into text, sometimes using the captions YouTube generates and sometimes transcribing it ourselves. An AI model then reads that text to pick out the overall ranking, the best-value pick and anything he warns against, along with the exact words he used. Where a video shows his results charts and summary table, those are read too and used to check the ranking.

Videos are sorted into categories and each product into its brand, including where the brand is based and, when Todd says so, where the product is made.

Why you might spot a mistake

Turning speech into text is an automated technique, and it isn't perfect. Brand names are the usual casualty: an unfamiliar name can come out as a different word, or even as another brand in the same test. Here's a real one, from a step bit test:

We check rankings against Todd's own charts wherever we can, but some errors can still get through. The video is always the final word, and every result here links back to it.