GEO Glossary

Authoritative definitions of key Generative Engine Optimization, AI search, and citation analysis terms.

GEO (Generative Engine Optimization)

The discipline of optimizing web content to be cited by generative search engines like ChatGPT, Claude, Perplexity, and Google AI Overviews.

Unlike classic SEO which targets position #1 on Google, GEO aims to have AI recommend your content when a user asks a question. The practice includes adding explicit definitions, structured schema.org data, direct-answer snippets, and competitive positioning in plain text that AI can extract. For example, a SaaS landing page might rank #1 on Google for "best project management tool" yet receive zero citations when a user asks ChatGPT the same question — because the page lacks a clear one-sentence definition that AI can quote. Research from Princeton and Georgia Tech (2024) showed that adding quotable statistics and authoritative language can increase AI citation rates by up to 40%.

Related: AI Search · Citation Score · Schema.org

AI Search

Search mediated by artificial intelligence — the user asks a chatbot (ChatGPT, Claude, Perplexity) and receives a synthesized response with citations, rather than a list of links.

In 2026, more than 30% of high-intent searches (purchases, comparisons, technical problems) already go through AI search. The key difference from traditional search is that the user doesn't click 10 results — they receive ONE synthesized response that cites 3-5 sources. If your site isn't one of those sources, you don't exist for that user. This shift is especially pronounced in technical and B2B niches: developers asking "what's a good alternative to Firebase?" increasingly use ChatGPT or Perplexity instead of Google. The traffic from AI search tends to be higher-intent and longer-session than organic search, because users arrive with a specific recommendation rather than browsing a list.

Related: AI Overviews · Citation Score

AI Overviews

Generative response blocks that Google shows above organic results when it detects a question. Formerly known as SGE (Search Generative Experience).

They function as an AI layer over Google's SERP: they read the top pages for your query, synthesize a response, and cite sources inline. If your content appears as an AI Overview source, you receive highly qualified traffic. If not, users tend not to scroll down to classic organic results. AI Overviews currently appear on roughly 15-30% of informational queries in the US, with higher rates for "what is," "how to," and comparison queries. Pages with FAQPage schema, clear definitions, and structured data are disproportionately selected as sources. Optimizing for AI Overviews overlaps heavily with GEO best practices: both reward explicit, well-structured answers over keyword-dense marketing copy.

Related: AI Search · Citation Score

Citation Score

A site's visibility metric in AI engines — the percentage of plausible queries where the site appears cited in the responses of the queried AI engines.

At AICite we calculate the score by generating 15 representative queries from the URL's content, running them through each configured AI engine, and measuring the citation ratio. Score 0/100 = never cited. Score 100/100 = cited in every query × every engine. Most indie sites score 0-20/100; established unicorns reach 50-90/100 depending on the engine. The score is broken down per engine, so you can see if ChatGPT ignores you while Claude cites you consistently (cross-engine asymmetry). Tracking your citation score weekly reveals whether content changes, new backlinks, or schema additions are actually improving your AI visibility — or if competitors are overtaking you.

Related: GEO · LLM-as-judge

LLM-as-judge

An AI engineering pattern where one LLM evaluates another LLM's output against structured criteria — useful for classifying, extracting information, or detecting properties in non-deterministic responses.

At AICite we use Claude Haiku 4.5 as judge to extract whether an AI engine's response cites a given site: it reads the text + sources, decides cited (yes/no), position, sentiment (positive/neutral/negative), and lists competitors cited. It's the modern replacement for regex-and-hope when classifying text. It works because models like Haiku are cheap ($1/$5 per million tokens) and consistent at structured tasks. The judge model uses structured outputs (JSON schema enforcement) to guarantee machine-readable results, eliminating parsing errors. This pattern is widely used beyond AICite: content moderation, automated grading, and evaluation benchmarks all rely on LLM-as-judge to assess unstructured text against defined criteria.

Multi-engine orchestration

The pattern of running multiple LLMs in parallel for a single task, aggregating results or using consensus.

AICite queries ChatGPT and Claude (and optionally Perplexity, AI Overviews) in parallel with Promise.allSettled — if one engine fails, the others continue, and we report the successful subset. Each engine has different biases (Claude knows B2B SaaS companies better, ChatGPT has broader general coverage), so running them together gives a more complete view of visibility. This approach also reveals cross-engine asymmetry: a site might be cited 90% of the time by Claude but only 15% by ChatGPT, indicating that the content resonates with one model's training data but not the other. Multi-engine results help prioritize which engine-specific optimizations to tackle first.

Cross-engine asymmetry

The phenomenon where the same site has very different scores across different AI engines (e.g., 93% on Claude but 20% on ChatGPT).

Each AI engine has different training data, prioritizes different types of content, and uses different retrieval strategies. Sites well established in a specific niche (e.g., Linear in project management) can be well known by one engine and nearly invisible in another. Detecting this asymmetry is one of the first things GEO enables — before optimizing, measure per engine separately. For example, AICite's own audit of Staxly showed Claude citing it for developer-tool queries while ChatGPT hallucinated it as a Minecraft proxy. Without per-engine measurement, you might optimize for the wrong problem — adding schema when the real issue is that one engine has stale or incorrect training data about your brand.

Related: GEO · Citation Score

Hallucination

When an LLM generates incorrect information with a tone of confidence — e.g., inventing product features, misattributing a quote, confusing entities with similar names.

Particularly damaging in GEO because your site may be cited but described incorrectly. Real audited example: ChatGPT believes Staxly (a dev platform catalog) is a Minecraft proxy. The typical cause is ambiguous content or missing disambiguation signals — no clear "About" section, or niche terms not defined. GEO fixes include detecting and correcting specific hallucinations. The most effective countermeasure is a disambiguation block in your homepage's first 60 words — a clear, factual statement of what your product is, who it's for, and what it competes with. Schema.org SoftwareApplication markup and explicit "What we are NOT" sections also reduce hallucination risk significantly by giving AI engines structured, unambiguous facts to anchor on.

Related: GEO · Disambiguation

Disambiguation signals

Explicit cues in your content that help AI correctly classify your site: "X is a tool for Y targeting Z users", "X competes with A, B, C", JSON-LD SoftwareApplication.

Without disambiguation signals, AI tends to infer incorrectly — especially for ambiguous brand names or products in new categories. Good signals include: an explicit one-liner definition in the H1 or opening paragraph, schema.org structured data, explicit comparisons with well-known competitors, and "What we're NOT" sections for edge cases. The strongest disambiguation signal is a factual sentence in the first paragraph: "X is a [category] tool that [primary function] for [target audience]." AI models weight early-page content heavily when forming entity associations. Adding JSON-LD SoftwareApplication schema with applicationCategory and operatingSystem fields further anchors the entity classification, reducing the chance of AI confusing your product with similarly-named entities.

Related: Hallucination · Schema.org

Schema.org

A standardized vocabulary of data types (Article, FAQPage, SoftwareApplication, etc.) inserted as JSON-LD in HTML so that AI crawlers and search engines can extract structured information without ambiguity.

In GEO it's particularly useful for: marking visible FAQs (FAQPage), declaring pricing and category (SoftwareApplication), describing comparisons (Product), and providing breadcrumbs. AI engines use it for both retrieval and citations — a FAQ marked with FAQPage is much more likely to be cited than free-form prose with the same information. The recommended implementation is JSON-LD in a script tag, which is invisible to users but machine-readable. Key schema types for GEO include SoftwareApplication (for SaaS products), FAQPage (for Q&A content), HowTo (for tutorials), and Organization (for brand identity). Note that Google restricted FAQPage rich results to government and healthcare sites in August 2023, but the schema still benefits AI citation engines.

Related: Disambiguation signals · GEO

Want to know how visible your site is to AI engines?

Audit your URL for free →