GEO glossary

The vocabulary that shows up in every Generative Engine Optimization conversation, defined without the jargon.

GEO (Generative Engine Optimization)

Optimizing for citations inside LLM answers.

The discipline of shaping how ChatGPT, Claude, Perplexity, and Gemini describe and cite your brand when users ask category questions. Sometimes called AI SEO, LLM SEO, or answer-engine optimization.

llms.txt

A markdown file at your root that tells LLM crawlers what your site is about.

A proposed standard (llmstxt.org) modeled on robots.txt and sitemap.xml. It lives at /llms.txt, uses H1 for site name, blockquote for summary, and H2 sections listing pages as markdown links. Crawlers use it to skip parsing your JS shell and read curated content directly.

ai-agents.json

A machine-readable manifest describing your endpoints and crawl policy.

A JSON manifest at /ai-agents.json that names your public API endpoints, MCP server (if any), preferred citation string, allowed crawlers, and license terms. Complements llms.txt for agents that want structured data rather than markdown.

Citation share

The percent of category answers in which an LLM cites your domain.

For a fixed set of seed queries, citation share = (answers citing your domain) / (total answers). Measured per model. It's the closest analog to 'search visibility' but scoped to a query panel you control rather than the whole search index.

Seed query matrix

The fixed list of questions used to measure GEO performance.

A curated set of category questions polled across each LLM on a schedule. The matrix must be public and stable — otherwise the score is unfalsifiable. Solarly's 70+ query matrix is published at /geo/methodology.

Confidence band

The uncertainty range on any single measurement.

LLM answers are non-deterministic. A confidence band widens when repeated polls of the same query produce different citations, and narrows when the model is consistent. Any GEO tool reporting a single number without a band is overstating what's measurable.

Fix Pack

The concrete deliverables that close a GEO gap on a given site.

In Solarly's model: a tailored llms.txt, ai-agents.json, and JSON-LD schema patches shipped to a target domain, with a guaranteed re-audit. Other vendors use different terms — the substance is the same set of artifacts.

SSR (Server-Side Rendering)

Serving fully-rendered HTML at the URL, not a JS shell that hydrates.

Most LLM crawlers do not execute JavaScript. If your title, description, JSON-LD, and body copy are injected client-side by React, most models will never see them. SSR (or static rendering) puts the content in the initial HTML response.

GPTBot / ClaudeBot / PerplexityBot / Google-Extended

The named crawlers that feed each major LLM.

OpenAI ships GPTBot, Anthropic ships ClaudeBot and Anthropic-AI, Perplexity ships PerplexityBot, Google separates its generative-AI crawler under Google-Extended. Robots.txt directives target them individually.

MCP (Model Context Protocol)

An open protocol for exposing tools and data to LLMs directly.

MCP lets a site expose typed tools an LLM client can call (list_case_studies, get_case_study, audit_url, etc.). It's an emerging channel for agent-to-site interaction that bypasses HTML entirely.

Drift

Change in citation share over time, usually after a model update.

LLMs update weights on their own schedule. A site that was consistently cited last month may drop with no site-side change — that's drift. Repeated polling is the only way to detect it.