A
AEO (Answer Engine Optimization). The practice of structuring content so that search engines extract and display it as a featured snippet or direct answer. Distinct from GEO in that AEO targets traditional search engine answer boxes, not LLM-generated answers. See GEO vs SEO vs AEO. Answer engine. A system that generates a natural-language answer to a user’s question, rather than returning a list of links. Examples include ChatGPT, Claude, Gemini, and Perplexity. Answer engines may be grounded (retrieve and cite) or ungrounded (answer from parametric memory). See How answer engines work.C
Citation. A reference in an answer engine response that attributes a claim to a source. Citations may be inline numbers, URLs, or footnote-style links. A brand that is mentioned without a citation has no traceable source for the claim. See LLM engines. Credit. The unit of metering on the platform, denominated in micro-USD (µUSD). 1,000,000 µUSD = $1.00. Operations reserve credits up front and settle actual cost after. New workspaces receive a free grant of 1,000,000 µUSD. See Credits and billing.G
GEO (Generative Engine Optimization). The practice of improving how a brand appears in AI-generated answers from LLM-based engines. Measured through mention rate, share of voice, position, sentiment, and citation rate. See What is GEO. GEO Score. A versioned 0–100 composite score (geo-score-v1) with five weighted components: visibility (0.25), relative share of voice (0.35), position (0.15), sentiment (0.10), and modeled reach (0.15). Missing components are reported and weights renormalized. See GEO Score.
Grounded engine. An answer engine that retrieves external information (from an index, web search, or database) before generating an answer. Grounded engines can cite sources. See How answer engines work.
H
Hallucination. A confident but incorrect statement generated by an LLM. Hallucination is more common in ungrounded engines that rely on parametric memory. A brand that launched after the engine’s training cutoff may be described with invented facts.I
ICP (Ideal Customer Profile). A description of the customer segments a brand targets. Used by the prompt generator to produce relevant questions. Impression share. The proportion of search impressions a brand receives for a given set of queries. From GSC data. Not directly a GEO metric, but seeds prompt importance scoring. Intent. The category of a prompt question. The platform uses six intents:discovery, problem_solution, use_case, comparison, expert, and brand_research. Each intent tests a different visibility dimension. See Prompt sets.
M
MCP (Model Context Protocol). The protocol ai-cmo.dev uses to expose tools to AI assistants. The packageai-cmo-mcp runs locally (stdio only, no HTTP/SSE endpoint) and proxies to the REST API. Provides ten tools for reading scores, runs, reports, and recommendations. See MCP server.
Micro-USD (µUSD). The denominational unit for credits. 1 µUSD = 0.000001 USD. All credit balances, reservations, and settlements are integer micro-USD amounts.
P
Preset. A named measurement configuration that defines which engines to run, how many prompts, and the budget ceiling. Available presets:smoke, lite, standard, deep, and exa. See Runs.
Prompt set. The curated list of questions sent to answer engines during a measurement run. Each prompt has an intent, topic, language, importance, and lifecycle state. See Prompt sets.
R
RAG (Retrieval-Augmented Generation). An architecture where an LLM retrieves relevant passages from an external index before generating an answer. Most grounded answer engines use some form of RAG. Recommendation. Tracked work produced by the platform, with a lifecycle:proposed → todo → in_review → done → archived. Each recommendation carries evidence references, generated copy, and state history. See Recommendations.
Research artifact. A schema-validated, immutable output from a research run. Carries source references, timestamps, digests, and the operation that produced it.
Retrieval. The process of finding relevant documents or passages from an index or search engine to condition an LLM’s answer. The quality of retrieval determines whether a brand’s content can be cited. See How answer engines work.
Run. A single measurement execution. Records the preset, date, provider responses, extracted mentions, metrics, and cost. Pipeline outcomes: complete, partial, failed. See Runs.