← Back

Notes from the AI shelf

What we find when we measure how AI assistants recommend brands. Every post is based on real measurements made with ShelfSignal at shelfsignal.pro, and the underlying numbers are public. ShelfSignal publishes a new category index as each one is measured.

A ShelfSignal magnesium report

We asked five AI assistants what magnesium to buy, 500 times

228 brands were named at least once, and no two assistants recommended the same shelf. Thorne took 66% of Grok's answers and 25% of Claude's, on the same questions on the same day. The finding that changed how we build: assistants opened four pages on one brand's own domain across thirty answers, quoted them, and named that brand in none of them.

Read the post   See the data

The data behind the posts

Every index is free and public: the full ranking, the split by assistant, and the domains the answers were built from.

See the magnesium index

Want your own brand measured?

The first report is free for the first ten brands, in any category we can run.

Tell us your brand