GEO / AI Visibility Expert

How to Use This Skill

• Without arguments — load Core Frameworks below for reference.
• With a topic — ask about llms.txt, structured data, tooling, or another indexed topic; I find and read the relevant chapter.
• With "audit <url or file>" — I follow the procedure in patterns.md to score the content against cheatsheet.md and return prioritized fixes.
• Browse — ask "what chapters do you have?" to see the full index below.

When you ask about a topic not covered in Core Frameworks, I read the relevant chapter file before answering.

Chapter Index

• chapters/01-definizioni-storia.md — cos'è GEO, il paper fondativo, GEO vs SEO
• chapters/02-motori-generativi-tecnico.md — come funzionano ChatGPT Search, Perplexity, Google AIO/AI Mode, Copilot, Gemini
• chapters/03-fattori-ottimizzazione.md — tecniche concrete di ottimizzazione, llms.txt
• chapters/04-tooling-misurazione.md — tool di misurazione AI visibility
• chapters/05-case-study-dati.md — dati, statistiche, casi studio
• chapters/06-checklist-best-practice.md — checklist e anti-pattern
• chapters/07-video-insights-consenso.md — consenso e divergenze cross-fonte da ricerca video
• chapters/08-entity-geo-locale-caso-agenzie.md — entity GEO per business locali di servizi: caso Magnet vs Bliss, perché un LLM consiglia un'agenzia e non un'altra, gap analysis sul concorrente

See also cheatsheet.md (audit checklist), glossary.md (terminologia), patterns.md (procedura di audit).

Core Frameworks & Mental Models

GEO vs SEO shift: SEO optimizes a whole page to rank in a list of 10 clickable links; GEO optimizes the semantic chunk to be selected as a citation inside one synthesized answer. Click-through is often absent entirely (zero-click) — the win is being mentioned even without a click — and a single answer can cite 3-10 sources at once, so there's no exclusive "position 1." A solid technical SEO base (crawlability, speed, no robots.txt blocks on AI crawlers) remains a prerequisite: GEO builds on top of SEO, it doesn't replace it (chapters/01-definizioni-storia.md).

Princeton technique-impact ranking: the founding GEO paper (arXiv:2311.09735) tested 9 techniques on Position-Adjusted Word Count (PAWC) and Subjective Impression (SI). Quotation Addition is the top performer at +41% PAWC / +28% SI; Statistics Addition follows at +33% PAWC / +22% SI (validated on real Perplexity at +37% SI). Keyword Stuffing is the one negative result, at -8% PAWC in the paper (down to -10% on real Perplexity) — it's measurably counterproductive, contradicting classic SEO practice (chapters/01-definizioni-storia.md).

Per-platform divergence: an analysis of 680M+ citations found only 11% of domains are cited by both ChatGPT and Perplexity. Implication: no single-platform strategy is sufficient — GEO work has to be measured and optimized per engine, which is also why multi-platform AI-visibility tooling exists (chapters/02-motori-generativi-tecnico.md).

Answer-first / BLUF structuring: the most-repeated tactical pattern across the 9 practitioner videos surveyed — start every section with the direct answer before any backstory (Bottom Line Up Front), because both humans and AI models weigh the start of a passage more heavily than the middle. Explicit in Kaliber, Ahrefs, Niko, Hostinger, and Surfer Academy (chapters/07-video-insights-consenso.md).

Branded mentions as top factor: off-site brand mentions are the single strongest AI-visibility factor identified in the video material — an Ahrefs study of 75,000 brands found branded web mentions correlate more strongly with AI Overview visibility than backlinks, Domain Rating, or referring domains. Silicon Valley Girl independently reports that 82% of what AI cites is "earned media" (third-party articles, listicles, Reddit threads), not self-published content. Earned mentions outweigh backlinks (chapters/07-video-insights-consenso.md).

SEO-as-prerequisite (with a caveat): classic Google ranking still correlates with AI Overview citation, but that correlation is weakening — only 38% of AI Overview citations now come from top-10 Google pages (2026 data), down sharply from 76% in mid-2025, as query fan-out decouples AI citation from classic ranking (chapters/05-case-study-dati.md). The 76% figure is also the most-repeated statistic across the 9 practitioner videos surveyed, but it traces to a single Ahrefs study rather than independent confirmation — attribute it as "Ahrefs data," not consensus (chapters/07-video-insights-consenso.md).

llms.txt reality check: llms.txt is a plain-markdown file at a site's root giving AI agents a machine-readable summary of the site — an emerging bottom-up convention, not a W3C standard. Adoption has passed 844,000 sites, including Stripe, Cloudflare, Vercel, and Anthropic. It only enables reading, not action — it's a convention, not a citation guarantee (chapters/03-fattori-ottimizzazione.md).

Citation concentration: a realistic-expectations note — an aggregated study of 680M+ citations found the top 1% of cited domains (~12 sites: Wikipedia, Reddit, Forbes, Healthline, Investopedia, NYT, and large .gov/.edu domains) capture 47% of all AI citations (chapters/05-case-study-dati.md).

Retrieval before persuasion (local services): an LLM recommends the business its retrieval returns with enough corroboration, not the "best" one. In the Magnet vs Bliss case, an agency with a stronger portfolio never appeared in "agency + service + Milan" answers while a competitor appeared even when the city changed, because the competitor had a literal page or section for every "service + city" phrasing, the same entity definition in every footer, textual case studies, author pages, and fresh informational content. The excluded agency had an ambiguous name, a brand-first hero with no who/where/what sentence, gallery-style portfolio, and almost no third-party sources. Fix order: disambiguate the entity and align NAP and description everywhere, add third-party corroboration, add schema with a shared @id, then rework the service pages, and lead with a vertical niche where you have real credentials. Avoid doorway pages and self-ranking listicles (chapters/08-entity-geo-locale-caso-agenzie.md).

Supporting Files

• cheatsheet.md — audit checklist by category
• glossary.md — terminology (GEO, AEO, LLMO, RAG, PAWC, etc.)
• patterns.md — content-audit procedure

Scope & Limits

This skill covers GEO/AI-visibility strategy and research: how generative answer engines retrieve and cite sources, concrete content/technical optimization techniques, AI-visibility measurement tooling, and cross-source consensus on best practice. It does not cover hands-on implementation of schema markup, CMS-specific publishing workflows, or general SEO practice beyond what's needed to understand GEO as a layer on top of it — check a dedicated SEO skill or tool documentation for that. Chapter 08 does list the schema properties and NAP fields that matter for local entity disambiguation, as a checklist rather than an implementation guide.