Google and Baidu systematically favour their own platforms in AI health answers over primary sources, per arXiv research on generative search engines. SEJ reinforces the break: page-one rank does not translate to AI citations, and corroboration across third-party sources, not brand size, determines who appears in local AI answers. A new EHQ benchmark finds 14 LLMs vary sharply in confabulation risk, making entity accuracy an active liability for brands with weak structured-data footprints. ChatGPT Ads CPCs are already reaching $22 with no auction transparency, leaving paid AI visibility as ungoverned territory.
Proven, dated changes we track across sources. Only confirmed and corroborated signals appear here.
AI Overviews pull from review content and Google Business Profile completeness when constructing local recommendation answers, making recency and review volume active ranking inputs.
LLM citation sources are determined by pre-existing institutional trust and training data authority, not by Google page-one ranking.
A new 3,000-question benchmark (EHQ-3000) quantifies per-model variation in confabulation versus abstention, revealing that epistemic honesty differs substantially across 14 models.
After six months of live advertising, ChatGPT Ads still provides no auction insights, industry benchmarks, or reliable audience data to advertisers.
Same audit finds Baidu AI Overviews route health-query citations to Baidu ecosystem properties rather than diverse external sources.
Audit of 1,920 health queries across 12 countries finds Google AI Overviews systematically cite Google-owned or adjacent properties over independent primary health sources.
LLMs exposed to conflicting entity attributes produce hedged or inaccurate brand answers, regardless of total content volume published by the brand.
ChatGPT's share of AI chatbot web traffic fell from 73.3% to 55.5% year-on-year while Gemini doubled and Claude grew nearly fivefold, per Similarweb.
Latent Space's Astra tracker documents that citation behaviour varies meaningfully across frontier models, with each model favouring different source types and domains.
Artificial Analysis released v4.2 of its Intelligence Index, revising GPT-6 Astra's score upward by four points after public skepticism about the previous methodology.
GPT-6 Astra launched to developers on 5 September 2026, with OpenAI citing improved prompt comprehension and more sophisticated output generation across the board.
OpenAI released GPT-6 Astra with new SOTA performance in coding and computer use, a 2.5x higher token price, lower cost per task, and reduced model monitorability.
ChatGPT lost 18 points of AI traffic share in a year. Brand citation strategy must catch up.