Reference · Speak the Language

The Dealership AI Glossary

Every term you'll hear in an AI pitch this year — defined in plain English, with a dealership example. Bookmark it, share it with your managers.

By the Learn Claude for Dealers team · Updated July 2026

The basics

LLM (Large Language Model)

The engine under the hood of Claude, ChatGPT, and Gemini. Software trained on enormous amounts of text that predicts language well enough to read, write, and reason. When a vendor says "our AI," ask which LLM it runs on.

Prompt

The instructions you give an AI. The single biggest lever on output quality. A good one includes role, context, constraints, and format — see dealership prompt templates.

Context / Context window

Everything the AI can "see" while working — your instructions, files, and the conversation so far. The window is its working memory size. AI with no context guesses; AI with your store's context performs.

Token

The unit AI reads and bills in — roughly three-quarters of a word. API pricing is per token; think of it as the AI's version of a per-minute phone bill.

Hallucination

When AI states something false with total confidence — a spec that doesn't exist, an incentive that expired. The fix is context and guardrails, not hope: give it the facts and forbid it from inventing. See guardrails.

The building blocks

Markdown (.md) file

A plain-text document with simple formatting, used to hold your store's voice, facts, and rules so AI can load them. The employee handbook for your AI — see knowledge files.

Project (Claude)

A workspace pre-loaded with your knowledge files, so every conversation starts already knowing your store.

Skill

A packaged procedure the AI loads on demand — your merchandising method or review-response SOP, executed the same way every time, by anyone.

Agent

AI given a role, instructions, tools, and a deliverable, executing multi-step work — a digital employee. See building a dealership agent team.

Workflow / Agentic workflow

Multiple skills and agents chained so work runs end to end — data in, decision-ready deliverable out — instead of living in one chat window.

Spec file / PRD

Documents that tell an AI exactly what to build and how to behave (PRD = product requirements document). Plain-English project scoping — the way modern AI work is briefed.

The plumbing

API

How software talks to software. "API access" means AI can be wired directly into tools and websites instead of used through a chat screen.

MCP (Model Context Protocol)

An open standard that lets AI assistants securely connect to other systems — calendars, CRMs, databases. Think of it as a universal adapter between your AI and your tools.

RAG (Retrieval-Augmented Generation)

A setup where AI looks facts up in your documents before answering, instead of relying on memory. How vendors make AI answer from your inventory and policies.

Fine-tuning

Retraining a model on custom data. Expensive and rarely what a dealership needs — knowledge files and skills get you there faster and cheaper. If a vendor leads with fine-tuning, ask why.

Scraping

Automatically pulling data from websites — competitor prices, inventory counts. Paired with AI analysis, it powers the competitive briefs in the agent playbook (taught hands-on at the workshop).

The buzzwords to decode in vendor pitches

"AI-powered"

Means anything from "wraps ChatGPT" to "real proprietary system." Ask: which model, what data of ours does it use, what happens to that data, and what does it do that a Skill we build ourselves can't?

"Automation"

Rules-based software following if-then logic. Useful, but not intelligence. AI handles ambiguity; automation repeats steps. Most great systems combine both.

"Autonomous"

Runs without a human in the loop. Powerful for internal work; for customer-facing work, demand review gates until trust is earned — see guardrails.