Airings, not spend. A machine classifier marks each 30-second window of broadcast as an advert or not and names the brand. We count the windows it marked as gambling advertising with at least 70% confidence over the last 28 days. A sponsorship line inside a sports show and a paid spot in a break are the same airing. We do not know what any read cost, who heard it, or whether it was paid for at all.
Promotions and messages are counted apart. A read that sells a bet, a ticket or a game is a promotion. A spot for the South African Responsible Gambling Foundation's counselling line is a message. The counselling line read at the end of a promotion, as the law requires, stays inside the promotion. The Foundation is funded by the industry.
We monitor six stations. 702, Power FM, Kaya 959 and 947 (Gauteng), Cape Talk (Western Cape) and SAfm (national), all in English. An operator that advertises on the stations we do not listen to is absent here, and that is a fact about our coverage, not about the operator.
Brand names come from speech recognition. The classifier spells an operator as it heard it, and we fold the spellings we know about onto one name. A read keyed on a product we have not met yet lands in "Other" on the Ads page until we add it; the ledger here can therefore undercount, never overcount.
The debate rows are keyword matches over the last 7 days of non-advertising windows, folded into conversations. Transcripts are machine-generated and quoted as heard. Allegations made on air are attributed to whoever made them.
The law section is written by hand from the sources it links, with the day each was checked. Bills move; if an entry is out of date, the source link and the date are there so you can see how far.
Something wrong? Tell us and we will correct it. Updated 15 Oct, 18:11 SAST.