Every week we count every ad the trade has running on Facebook and Instagram. What they say, what they promise, how many each company runs, how long they have been up. We name no company. It is about the trade, and about where you stand in it.
The bath ad census, edition 2026-W36: 118 companies, 3,225 live creatives, 4 September 2026.
The window ad census, edition 2026-W36: 76 companies, 3,227 live creatives, 4 September 2026.
Meta's public ad library shows what is running now. After an advertiser stops an ad, it stops being browsable. So each week we query every verified page in a trade by its own id. We keep what came back, word for word, and store nothing derived. Every number on an edition comes from those raw rows, and we publish the rows with it.
A keyword search cannot count an advertiser. In the bath edition the busiest page returned 53 creatives to a phrase. Asked directly, the same page returned 260, on the same day. So we sample nothing by phrase.
A census counts what it can find. These queries find the national brands and the franchise networks with ease. The local company a different phrase would have found is the one we miss. Each edition prints its own recall, in counts.
The bath edition ran 57 queries and counts 215 verified pages, 147 of which one query alone surfaced. The window edition ran 58 queries and counts 144 verified pages, 92 of which one query alone surfaced.
One page lists every number this site published and later changed. It gives the cause and the size of each change. Read what the census got wrong.
Expenety, the [trade] ad census, edition [week], with the edition URL. Every edition carries a schema.org Dataset record and a row file you can download. Both sit under CC BY 4.0. Use the rows, cite the edition.