A quick-reference infographic for marketers and builders: the ten terms showing up in every AI roadmap, vendor deck, and team standup.
June 4, 2026 · 3 min read
If you lead marketing or digital at a firm, you have heard all of these words in the last ninety days. Few teams define them the same way. This is the shared vocabulary I use when I am aligning practice marketers, agencies, and engineering on what we are actually building.
Save it. Argue with it. Make your own version for your org.
Large Language Model. A model trained on massive text (and often multimodal) data to predict and generate language. ChatGPT, Claude, and Copilot are products built on LLMs. When someone says "the model," they usually mean the engine, not the app around it.
A packaged capability inside an AI product. Skills (Claude Skills, custom GPT actions, Copilot extensions) bundle instructions, tools, and guardrails so the model repeats a workflow reliably. Think: how we do competitive briefs here, not a one-off prompt.
Software that plans steps and uses tools toward a goal. Agentic systems break work into subtasks, call APIs, read files, and loop until done or stopped. Marketing use cases: research synthesis, draft-and-route workflows, monitoring. Still needs human judgment on brand and risk.
What the system remembers across sessions. Can mean chat history, a user profile, a project knowledge base, or RAG-retrieved docs. Ask which memory: short-term thread, long-term store, or enterprise corpus. Compliance lives here.
Answer Engine Optimization. Shaping content and structure so AI answer surfaces (ChatGPT search, Perplexity, Google AI Overviews) cite and summarize you accurately. Cousin to SEO, different ranking signals: clarity, authority, extractable facts.
Generative Engine Optimization. Broader than AEO: influencing how generative systems represent your brand across text, images, and recommendations. Still early, still messy. Worth tracking if your category is high-consideration and research-heavy.
Model Context Protocol. An open standard for connecting AI clients to tools and data sources in a consistent way. If your team is wiring CRM, analytics, or internal wikis into assistants, you will hear MCP in the architecture conversation.
Retrieval-Augmented Generation. The model pulls relevant chunks from your documents or database before answering, so responses stay grounded in approved material. Common in enterprise marketing: playbooks, disclaimers, past campaigns, firm policies.
How much text the model can hold in one request. Measured in tokens. Bigger windows mean full briefs, transcripts, and decks in one pass. Still not infinite: prioritize what you paste, and structure inputs with headings.
The units models charge and count. Roughly a word fragment. Pricing, limits, and latency tie to tokens. Practical rule for marketers: shorter, structured inputs often outperform dumping entire libraries without curation.
In vendor meetings, I ask which term they mean before we scope a pilot. In internal roadmaps, I map each initiative to one primary concept so "agent" does not become a bucket for every automation idea.
The definitions will keep shifting. The discipline is naming them clearly so your team ships with the same mental model.
Related: From Basic AI Prompts to Agents, MCP, and a Builder Stack | How I Use Four Different AIs