Issue 2 — 14/04/2026

Anthropic's ClaudeBot: Entity Authority and Structured Data Are Primary Citation Signals

Brain Buddy AI · 01/04/2026

A detailed technical guide published in April 2026 documents that Claude selects sources based on entity authority, factual accuracy, structured data, and content clarity — not raw traffic or backlink volume. Claude also supports llms.txt for explicit crawl configuration. Practitioners optimising for Anthropic's platform should prioritise entity disambiguation, JSON-LD markup, and a verifiable author footprint over volume-based signals.

Google AI Overviews Now Appear on 65% of Queries; CTR Drops 40% When Present

LLM Intel · 07/04/2026

According to figures cited by LLM Intel, Google AI Overviews are now triggered on 65% of queries, and pages that appear in traditional SERPs below an AI Overview see an average 40% click-through rate reduction. This confirms the structural shift in value: ranking without citation in the synthesised answer yields materially less traffic. The implication is that GEO and SEO must be run as parallel workstreams, not sequential ones.

ChatGPT Search Exceeds 100 Million Daily Queries as AI Search Normalises

LLM Intel · 07/04/2026

LLM Intel reports ChatGPT Search is now processing more than 100 million daily queries, cementing it as a citation surface practitioners cannot ignore. Unlike Google AI Overviews, ChatGPT Search operates outside the traditional index and favours sources with high factual density and explicit citations within the content itself. Brands relying solely on Google-facing optimisation are likely invisible in a significant share of these interactions.

RAG Architecture Means Real-Time Retrieval — Not Training Data — Drives Most AI Citations

LatentView · 07/04/2026

A LatentView explainer confirms that RAG-enabled AI systems retrieve external content at query time rather than relying solely on baked-in training data. This means content published after a model's training cutoff can still be cited — provided it is crawlable, structured, and indexed. For practitioners, this shifts the crawl and indexation checklist from a one-time SEO hygiene task to an ongoing citation readiness requirement: if ClaudeBot, Perplexity's crawler, or Bing's AI indexer cannot access and parse your content cleanly, recency alone will not help.

Perplexity Hits 780M Monthly Queries; Citation Behaviour Diverges Sharply from Google AI Mode

Surferstack · 02/04/2026

Surferstack's April 2026 analysis — drawing on a 680-million-citation dataset — finds that Perplexity and Google AI Mode reward fundamentally different content signals: Perplexity favours Reddit presence, real-time freshness, and direct-answer formatting, while Google AI Mode requires traditional SEO foundations plus E-E-A-T signals. Perplexity processed 780 million monthly queries as of mid-2025, tripling its volume in one year. The key practitioner implication is explicit: you cannot optimise for one platform and expect lift on the other — platform-specific content strategies are now a baseline requirement.

Perplexity Ships Memory Engine, Scheduled Searches, and Live Document Connectors

Leadership in Change · 04/04/2026

A feature roundup published 4 April 2026 documents ten Perplexity updates shipped since December 2025, including a persistent memory engine, scheduled automated searches, SEC-linked financial data integration, and live connectors to Gmail, Slack, and Google Drive. For ASO and GEO practitioners, the scheduled search feature is directly relevant: Perplexity can now resurface branded or topical queries on a recurring basis, meaning content freshness and consistent entity mentions across authoritative sources become compounding citation signals rather than point-in-time ones.

Building a Semantic Research Assistant: A Production RAG Pipeline Over 120 arXiv Papers

Srikanth Reddy · https://medium.com/@srikanth2314/building-a-semantic-research-assistant-a-production-rag-pipeline-over-120-arxiv-papers-9ed3073da527 · 13/04/2026

A practitioner-authored benchmark comparing three transformer embedding models against BM25 retrieval across 120 arXiv papers found a 7.7× answer quality gap in favour of semantic embeddings over keyword-based retrieval. For GEO practitioners, the finding reinforces why semantic coherence — writing that clusters concepts consistently rather than keyword-stuffing — improves retrieval probability in RAG-based AI search systems: BM25-style matching is increasingly a fallback, not a primary retrieval path. (Note: this is a practitioner blog post, not a peer-reviewed paper.)

Which Papers Do AI Systems Actually Cite? A 100-Query Empirical Study

Academic SEO (author unattributed) · https://academicseo.co.uk/blog/ai-citation-study.html · 07/04/2026

Researchers queried an AI system 100 times with biomedical questions and tracked citations against a matched control group of 172 non-cited papers. The finding is counterintuitive for GEO practitioners: once human citation count is controlled for, no content feature — open access status, structured abstracts, declarative titles, or entity specificity — significantly predicts AI citation. The only reliable signal is how often humans have already cited the source. (Unconfirmed: methodology not peer-reviewed; single AI system tested; findings may not generalise beyond academic or biomedical content.)