40 items tagged citation patterns. Page 1 of 2.
RAG architecture choices, not content quality alone, determine which sources get cited in LLM answers.
Chatbots cited an anti-abortion group in 17% of pregnancy queries, with no disclosure, per AlgorithmWatch.
Pew: a third of post-2022 web pages show AI text. LLM citation credibility now turns on domain provenance.
Reddit's ChatGPT citation share is falling. Institutional content, properly structured, stands to gain.
Reddit's ChatGPT citation share has dropped sharply. No one knows why, and that opacity is the real story.
ChatGPT's fan-out queries mirror E-E-A-T logic. Unnamed institutional content is invisible to AI answers.
GSEs favour a narrow publisher set. Authoritative institutions outside it vanish from AI answers, regardless of content quality.
Competing GSEs face a collective action trap that structurally erodes the web corpus their answers depend on.
Star ratings move LLM recommendations by 31 points. Structured reputation signals now determine AI visibility.
ChatGPT cites small unlicensed sites as often as major licensed publishers, new data shows.
Web search mode changes ChatGPT's citation behaviour systematically, an arXiv audit of 4,812 responses finds.
LLMs attend to authority tokens, not facts, when giving sycophantic answers. Brand credentials must be explicit in source text.
Perplexity crawls live on every query. Fresh, well-structured pages beat established domains.
All Big Four firms now linked to AI hallucinations in published reports, raising citation risks for brands that cite them.
ChatGPT's citation rates vary sharply by topic, creating visibility gaps for policy, finance, and multilateral brands.
Google cites its own properties #1 and #2 in AI Mode, crowding out third-party brand content.
Chinese generative search surfaces only 8.3% of cited brands in final answers, per a new arXiv study of 160,000+ records.
Where source records are thin, LLMs hallucinate most and GEO manipulation costs least.
RAG systems score clean on hallucination checks while citing the wrong entity's evidence, at rates up to 87%.
RAG's two-stage mechanics decide which pages AI search cites. Most B2B content teams are optimising for the wrong stage.
Cloudflare's three-tier AI crawler policy, live September 15, changes which LLMs can see your content.
35% of web-search LLM answers flagged for untrustworthy citations in University of Iceland pre-launch audit.
Hallucinated citations are passing peer review at top AI venues, corrupting the corpora that LLMs learn from.
LLM groupthink rewards high-volume publishers. Institutional knowledge in PDFs barely registers.