Wiki
Encyclopedia entries for GEO concepts, terms, standards, companies, people, and events.
Foundations
- AEO vs GEO Concept AEO vs GEO — the same construct in practice. AEO is the older umbrella term from the Featured-Snippet and voice-answer era of extractive single-passage answers; GEO is the term that stuck once generative synthesis became the default answer surface.
- AI Search Timeline (2022–present) Concept AI search arrived sharply on 2022-11-30 with ChatGPT and has changed shape four times since: Shock (Bing Chat / Bard), Fragmentation (SGE / Claude / Perplexity, Aggarwal coined 'GEO'), Mainstream (AI Overviews / ChatGPT Search), Stabilization (AI Mode 1B MAU).
- Answer Loop Concept A generative engine answers every query by running the same four-step runtime loop — query → retrieval → grounding → answer. GEO is not an abstract 'optimize the engine' move; it is intervening at each step, where each step has one lever you can push and one way you can fail.
- Citation vs Mention vs Link Concept An AI answer credits you three non-equivalent ways: a citation (content credited with a reference), a mention (named, no link), or a link (a source, maybe not even used). Being grounded on and being credited are decoupled — each form maps to a different metric and lever.
- Generative Engine Concept A generative engine answers a query by retrieving sources and synthesizing a written answer with an LLM, optionally crediting them. The break from a search engine is the output unit: returning documents vs. composing an answer — and it is the object GEO optimizes for.
- Generative Engine Optimization Concept Generative Engine Optimization (GEO): getting your content retrieved, grounded on, and cited or mentioned in AI answers (ChatGPT, Perplexity, Google AI Overviews, Gemini). An extension of SEO, not a replacement — and GEO Wiki's front door to every GEO sub-topic.
- GEO ROI Models Concept Under generative search the click stops attaching cleanly to value — and traditional ROI math breaks with it. Three currencies (citation, substituted traffic, brand authority) × three industry models (B2B SaaS, B2C e-commerce, media) is the frame this entry argues for.
- LLMO vs GEO Concept LLMO vs GEO — mostly the same thing. 'LLM Optimization' shares GEO's goal and, in practice, near-identical trade definitions. 'Model layer' is a framing emphasis, not a separate discipline; the one genuinely distinct reading, training-corpus inclusion, is what GEO excludes.
- SEO vs GEO Concept SEO vs GEO — same plumbing, different finish line. Both need crawlability, real expertise, clean structure and authoritative mentions; they split only at the success surface: a clicked rank vs a cited/mentioned answer. GEO is a layer on top of SEO, not a replacement.
- Zero-click Search Concept The user gets the answer on the results page and never clicks a source. Predates AI — most Google searches were already zero-click before generative answers escalated it. The precondition for GEO: value moves from being clicked to being cited or mentioned.
Signals
- AI Content Detection Concept An anti-signal entry — how AI engines down-weight content patterns associated with low-effort or scaled production, independent of whether AI tools were involved, and why classifier-based 'AI detection' is not the lever it is sold as.
- Brand Mentions Concept An unlinked brand mention is a load-bearing GEO signal: being named across the web — with no link — feeds the model's entity prior and compounds into future answers. A mention is a different currency from a link, not a weaker one.
- Citability Concept Citability is the structural property that decides whether a retrieved passage can be lifted, intact, into an AI-generated answer — independent of whether the source is trusted enough to be used at all. It is the shape half of grounding; E-E-A-T is the trust half.
- Content Freshness Concept Content freshness is two properties in one word: recency (how old a page is) and currency (whether its claims are still true). AI engines favour recent content — but the average AI-cited page is ~2.9 years old, so freshness is query-conditional, not a mandate to publish weekly.
- E-E-A-T Concept E-E-A-T — Experience, Expertise, Authoritativeness, Trustworthiness — is Google's quality-rater framework, not an algorithmic score. In GEO it is the trust half of grounding: whether a retrieved, liftable passage's source is worth using. Orthogonal to citability, the shape half.
- Entity Recognition Concept Entity recognition is the layer where an AI engine decides which known entity a name refers to. It is the join that attaches a mention, a citation, or a markup assertion to the right node — if you do not resolve, credit leaks, lands on the wrong entity, or is dropped.
- GEO Spam & Manipulation Concept Adversarial GEO is a documented research field and a live commercial market. Every published attack works in a single-attacker lab; every study that adds a second attacker finds the advantage shrinks or inverts. Two of the four attack surfaces are not on your website at all.
- Knowledge Graph Presence Concept Knowledge graph presence is having a structured node an AI engine already trusts — Wikipedia, Wikidata, the Google Knowledge Graph. It is an amplifier, not a cause: it lifts the model prior and gives resolution a destination, but cannot be self-declared into existence.
- Multilingual GEO Concept Multilingual GEO is what changes when a query, page, or citation crosses a language boundary. The GEO loop's shape is invariant; four things vary — source pool, entity binding, chunk shape, trust pool. Per-language source-pool difference is the load-bearing fact.
- Multimodal Signals Concept Multimodal signals are what AI engines read on non-text assets — images, video, audio, charts. The text channel (alt, caption, transcript, schema) still dominates over pixel vision in 2026 web-retrieval pipelines for both index-integrated and live-fetch AI.
Infrastructure
- AI Crawlers Concept AI crawlers split into three categories with opposite access consequences — training, retrieval, and user-triggered. The decision is per-category, not per-bot; the costliest GEO mistake is blocking the citation category to stop the training one.
- ChatGPT-User Concept OpenAI's user-triggered fetcher — it retrieves a page because someone is asking about it right now. OpenAI documents that robots.txt may not apply to it, and that it has no effect on ChatGPT Search inclusion. Blocking it costs a reader's answer and buys almost nothing.
- Core Web Vitals (LCP/INP/CLS) Concept Core Web Vitals (LCP / INP / CLS) is a Google ranking signal, not an AI-engine signal. The GEO effect is bounded — direct on Google AI Overviews, partial on Bing Copilot, negligible on ChatGPT Search, Perplexity, and Claude. AI crawler perf is a separate problem with its own fix.
- Google-Extended Concept Google-Extended is a robots.txt control token, not a crawler. It governs listed Gemini training and grounding uses without changing Google Search inclusion or ranking; Search AI appearance and grounding use a separate Search Console control now rolling out.
- GPTBot Concept OpenAI's training crawler: it collects content that may train future models, and nothing else. It does not feed ChatGPT Search, so blocking it costs no citation. What blocking costs is unmeasurable and delayed — the only honest reason to think twice before disallowing it.
- JSON-LD Standard JSON-LD is one of three Schema.org serializations — Google recommends it because it lives in a script tag, doesn't touch visible HTML, and is easy to maintain. Index-integrated AI parses it as structured data; live-fetch chatbots read it as plain text on the page.
- llms.txt Standard llms.txt is a proposed convention (Answer.AI, 2024): one curated markdown file at /llms.txt naming the pages an LLM should read first. Supply-side adoption is real; demand-side consumption is unconfirmed. A forward-compatible bet, not a citation channel.
- robots.txt Standard robots.txt is RFC 9309 — a voluntary request, not access control. Compliant AI crawlers honor it as documented; non-compliant or spoofed ones do not. Write the policy per category (training / retrieval / user-triggered) and verify the rest at the network layer.
- Schema.org for AI Concept Schema.org markup is not a ranking or citation signal. For AI it is infrastructure — it disambiguates who/what you are and makes a page parseable. Index-integrated AI uses it; live-fetch chatbots read JSON-LD as plain text. It makes an entity resolvable, not a passage liftable.
- Sitemap & IndexNow Standard Sitemap.xml (2005, pull) and IndexNow (2021, push, Bing/Yandex only) are two submission protocols. Their AI effect transits only via host search indexes — AIO via Google, Bing Copilot via Bing. ChatGPT, Perplexity, and Claude consume neither directly.
Practice
- GEO Metrics Concept The 10 core KPIs for measuring GEO — definitions, formulas, SEO equivalents, and how each major vendor (Profound, Otterly, Ahrefs, BrightEdge, Similarweb) actually defines them. A GEO Wiki synthesis, not an industry standard.