On 112 major news events, Canadian influencers put a fifth of their posts into U.S. stories; American influencers put 2% into Canada’s. How the 485 most active political influencers in each country split their posts on 112 major news events, 2024–2025, by where the event came from. Derived from the brief’s data. Read the brief ↗
ChatGPT cited the 21 news outlets that block its search crawler just once; Google AI Mode cited them 2,356 times. Citations in AI answers to Québec election questions, 10–11 September 2026, of the 21 French- and English-language outlets whose robots.txt blocked OpenAI’s search crawler. Read the report ↗
Before the referendum, 53% of Albertans had confidence in Elections Alberta, down 18 points from 2023. Albertans with a great deal or quite a lot of confidence in each agency, in weighted MEO surveys from May 2023 and Sept. 24–Oct. 5, 2026. Read the report ↗
After Meta’s news ban, likes on local news fell 89% in rural areas and mid-sized cities, 48% in big cities. Likes on local outlets’ posts across platforms in the six months after Meta’s 1 August 2023 news ban, against the six months before; community size from the 2021 Census. Read the brief ↗
Posts saying the Yandex strikes wiped out Russian bots drew 73% of views; posts disputing it drew 14%. 622 posts on X about the claim in the Alberta referendum conversation, 8–13 October 2026 (retweets excluded), and their 26.6 million views as counted at collection. Read the insight ↗
After three shootings, 11% of posts held a group responsible, but those posts drew 22% of engagement. Posts about the events by Canadian news outlets, politicians and influencers and by U.S. influencers, in the 72 hours after the Charlie Kirk assassination and the Tumbler Ridge and Montreal shootings. Read the report ↗
An AI content-farm network on Facebook, largely run from Vietnam, grew from 117 to 3,231 posts a day in 16 months. Monthly average of daily posts from the 340-page network, January 2025 to May 2026, derived from the report’s data; 89% of pages that disclose an administrator country have administrators in Vietnam. Read the report ↗
Attacks seeking help with non-consensual intimate images succeeded 74% of the time, the most of six harms. Share of 420 multi-turn attacks on four AI chatbots (May–June 2026) that drew a full, specific harmful answer, per an independent judge model. For intimate images, attacks sought help finding tools to make them, not the images themselves. Read the memo ↗
