Issue 16 — 21/07/2026

AEO Periodic Table V4: Brand Visibility Factors for AI Search

Goodie · 23/06/2026

Drawn from 1.13 million prompts across ChatGPT, Claude, Perplexity, Grok, Gemini, and Google AI Mode, Goodie's fourth-edition study finds that AI citation is shaped as much by off-site signals — Wikipedia, Reddit, review platforms, news coverage — as by a brand's own pages, yet most organisations have no shared analytics layer across SEO, PR, and social to manage those surfaces. The framework maps visibility factors into a periodic-table format to help teams identify which levers sit inside versus outside their direct control. For practitioners, the key implication is that citation audits must extend well beyond the brand's own domain.

How AI Engines Choose and Cite Sources: A 7-Month Analysis

Conductor · 07/05/2026

Tracking citation behaviour across seven AI engines from September 2025 through March 2026 — producing 1,056 engine-intent-month data points — Conductor found that each engine maintains a persistent editorial identity: ChatGPT and ChatGPT Search are the only engines to surface Wikipedia; Perplexity and Google Gemini both default to YouTube across most intents; Google AI Mode preferentially routes users back to Google properties. Only 11 per cent of domains (per a parallel GoGoChimp analysis) are cited by more than one engine, and citation volume varies 615x between platforms. Practitioners cannot rely on a single content strategy to cover the full AI search ecosystem.

Seer Interactive AIO CTR Study v3: 2026 Update

Seer Interactive · 24/04/2026

Covering 53 brands, 5.47 million tracked queries, and 2.43 billion organic impressions across full-year 2025 plus Q1 2026 actuals, Seer's third CTR study finds that the predicted continued decline in click-through rates did not materialise — the downward trend reversed in early 2026. The data are directional rather than causal, and the study notes that AI Overview frequency and source selection continue to fluctuate by query type. For practitioners, this signals that the window to build citation share while CPCs and competition for AIO slots remain lower than traditional position-one may be narrowing.

Google AI Mode Begins Rolling Out Third-Party App Integrations in the U.S.

Google Search Blog · 16/07/2026

From 16 July 2026, U.S. users can securely link services such as Instacart, Canva, and Spotify directly inside Google AI Mode, enabling actions — adding items to a cart, generating design templates, curating playlists — without leaving the Search interface. Google frames the capability under 'Personal Intelligence,' meaning AI Mode responses will become increasingly tailored to individual connected-account data. For GEO practitioners, this deepens AI Mode's role as a transactional endpoint rather than a discovery layer: brands in commerce, productivity, and media categories that are not cited within AI Mode answers risk being bypassed at the point of action, not merely at the point of research.

AI Overviews Now Fire on ~65% of Question-Based Searches; First-Citation Slot Correlates Strongly with Standard SEO Signals

Seer Interactive · 28/05/2026

Seer's study of 8,500 stratified keywords across 30 industries found that AI Overviews now appear on roughly 65% of question-based queries, and that the first-citation slot correlates primarily with established SEO fundamentals — page authority, relevance, and snippet eligibility — rather than any AI-specific schema or special file. Google's own published guidance (cited in the study) confirms there are no additional technical requirements beyond normal SEO best practices for AIO eligibility. Practitioners should focus optimisation effort on content clarity and snippet structure rather than proprietary AIO-specific tactics.

Google Introduces 'How This Ad Was Made' AI Transparency Panel Across Search, YouTube, and Discover

Google Ads Blog · 09/07/2026

Google is rolling out a 'How this ad was made' section inside the My Ad Center panel, disclosing whether an ad was created or edited with generative AI; advertisers using Google's own AI tools will have disclosures added automatically, while those using third-party AI tools can self-declare. Where local regulations require it, a label will appear directly on the ad creative. For GEO and content teams, the disclosure infrastructure signals that AI-provenance labelling is becoming a platform-level standard — organic content credibility signals (authorship, sourcing, editorial standards) are likely to carry increasing differentiation weight as AI-generated content becomes visibly flagged.

EY Develops Multimodal RAG Framework Built Around Knowledge Graphs to Improve Answer Accuracy

SiliconANGLE · 17/07/2026

EY has published details of a multimodal RAG architecture that retrieves charts, tables, engineering diagrams, and images alongside text, connecting them through a knowledge graph before supplying evidence to the LLM at inference time. The approach does not modify the underlying model but changes how enterprise content is indexed and related, producing answers EY describes as more complete and verifiable. For GEO practitioners, the implication is directional: as enterprise AI systems adopt richer retrieval pipelines, content that exists only as narrative prose — without structured tables, clearly labelled figures, or machine-readable entity relationships — is at greater risk of being overlooked in favour of content that can be cleanly extracted in multiple modalities.

Google Expands Managed Agents in Gemini API with Background Execution and Remote MCP Integration

Google DeepMind Blog · 07/07/2026

Google DeepMind has announced background (async) execution, remote MCP server integration, and custom function-calling support for Managed Agents in the Gemini API, allowing agents to complete long-running tasks without a persistent user session. The update targets production-grade agentic deployments where reliability and credential management across interactions are blockers. For ASO/GEO practitioners, the relevance is structural: as Gemini-powered agents increasingly perform multi-step research tasks autonomously, content that is crawlable, answer-extractable, and structured for machine retrieval will be surfaced in agent-generated outputs — not just in direct user queries.

MultiSynt/MT: ~4.8 Trillion Target-Language Tokens Released as Open Multilingual Pre-Training Corpus Across 36 European Languages

arXiv (via arXiv.gg) · 01/07/2026

Researchers have released MultiSynt/MT, an open synthetic parallel corpus of approximately 4.8 trillion tokens across 36 European languages, produced by machine-translating 100 billion high-quality Nemotron-CC tokens. Reference LLMs trained on MultiSynt/MT reach the performance of a native-data baseline (HPLT 2.0) using roughly 72% fewer pre-training tokens. For GEO practitioners targeting non-English markets, this corpus release signals that LLM knowledge of medium- and lower-resource European languages will improve substantially in the next generation of open models — raising the urgency of publishing authoritative, citable content in those languages before model knowledge gaps close and competition for citation share intensifies. (Pre-publication / arXiv)

Deduplication in LLM Pre-Training: How Duplicate Content Distorts Model Weights and What It Means for Citation

LLM Pioneer Hub · N/A

A technical guide to production deduplication of LLM pre-training corpora identifies four downstream effects of duplicate content: wasted training compute, increased memorisation risk, validation-set leakage, and unexplainable data versioning. Web mirrors, reposted news, templated product pages, and site farms are cited as common duplicate sources. The practitioner implication is direct: highly templated or widely syndicated content may be deduplicated out of training corpora entirely, reducing the probability that such pages shape model knowledge; original, distinctively structured content has a structural advantage in training-data retention.

DuoFlow-KG: A Dual-Modal Evidence Retrieval Framework for High-Density LLM-Augmented KGQA

Authors not specified in source · https://www.nature.com/articles/s41598-026-59526-3 · N/A

DuoFlow-KG proposes a unified retrieval framework that constructs compact evidence subgraphs by jointly modelling semantic relevance and structural dependencies within knowledge graphs, addressing fragmented evidence and search-space explosion in multi-hop reasoning. The dual-modal fusion module projects semantic and topological signals into a shared embedding space, improving answer coherence on Knowledge Graph Question Answering benchmarks. For GEO practitioners, the paper is evidence that retrieval systems underpinning enterprise and API-served LLMs are moving toward entity-relationship structures — content that exposes clear entity definitions, typed relationships, and verifiable claims is better positioned to be surfaced by next-generation retrieval pipelines.

Multi-Turn Agentic Scientific Literature Search via Workflow Induction

Jisen Li, Bingxuan Li, Nanyi Jiang et al. · https://arxiv.gg/abs/2607.00597 · 01/07/2026

PaperPilot frames scientific literature search as 'workflow induction': given an anchor paper and an underspecified user intent, the agent iteratively refines its retrieval strategy across multiple turns rather than issuing a single query. The approach makes search strategies inspectable and refineable, outperforming fixed-pipeline agents on multi-turn benchmarks. For GEO practitioners, the relevance is in the multi-turn trajectory: as AI agents issue successive retrieval calls rather than single queries, content that answers follow-up questions and covers a topic at depth — not just a single keyword intent — is more likely to be cited across the full agent session. (Pre-publication / arXiv)