Module 4 min
Core Vocabulary
Retrieval, grounding, passage, citation, RAG, hallucination: five words that explain almost every GEO decision.
Why it mattersMost GEO mistakes are really vocabulary mistakes: treating an LLM like a keyword index.
Definition & Foundation
What it is, in plain words
Most GEO mistakes are really vocabulary mistakes: people treat a language model like a keyword index and optimize the wrong thing. Seven words fix that. The cards below are the working glossary: each term in plain language, with the decision it changes. Share them before briefing any content.
One pairing underlies all seven: parametric knowledge (what the model absorbed during training; frozen, unattributed) versus retrieved knowledge (what it fetches live and can cite). GEO is almost entirely the fight to move your facts into the retrieved column, because retrieved knowledge is what gets grounded, quoted, and credited. Get these terms right and most GEO tactics stop being mysterious and start being obvious.
The Seven Words
Plain-language definitions, and the decision each one changes
Retrieval
The engine fetches candidate passages that look relevant to the query: a librarian pulling books off the shelf before reading any of them. If your content is never retrieved, nothing downstream can save you. Retrievability comes first.
Passage
The actual chunk the engine pulls: a few sentences, not a whole page. Citation studies favor self-contained passages of roughly 134–167 words. The decision it changes: you optimize passages for prompts, not pages for keywords.
Grounding
The model ties its answer to the fetched sources instead of its memory. A grounded claim can be checked and credited; an ungrounded one is a guess. Clean, sourced statements are what make your facts groundable.
Citation
The visible credit linking a sentence in the answer back to your passage. This is GEO's conversion event, the equivalent of the click. Everything else in this glossary exists in service of earning it.
Hallucination
What happens when the model answers from memory and invents something: a stat, a spec, a feature you never shipped. Your defense is making the true fact easy to retrieve and quote: stated cleanly, sourced, self-contained.
Parametric vs retrieved
Trained-in knowledge versus freshly fetched knowledge. Parametric is slow to change and never credits you; retrieved can change this week and carries your citation. Different goals, different tactics; decide which one a project targets.
Query fan-out
The engine silently splits one question into many sub-queries and builds its answer from sources that keep appearing across them. A key reason only ~38% of AI Overview citations now come from top-10 pages: covering the sub-questions beats ranking for the typed one. See Query Fan-Out.
Putting It to Work
Making the words do work
A glossary only pays off when it changes decisions. Wire these terms into how your team writes briefs and critiques drafts: three moves, no tooling required.
The shared-language playbook
Put the seven terms on the brief
Add a one-line glossary to your content brief template, and make "core passage" a required field. When a writer sees it as a deliverable, "passage" stops being jargon and becomes work product.
Critique drafts in the right vocabulary
Replace vague notes like "make it rank better" with precise ones: "this passage isn't self-contained," or "this claim will be hallucinated unless we ground it with a cited stat." Precise language produces precise edits.
Flag hallucination-risk claims
Train writers to spot claims a model is likely to get wrong from memory (niche stats, recent changes, your own product details) and state those in clean, quotable, well-sourced passages. That is where grounding earns its keep.
Verification Checks
How to know it's really done
The vocabulary is working when it changes edits, not when it's memorized. Click a check to mark it verified:
Works Together With
The nodes this one leans on
Tools for this from Foundations
Go deeper
Voices to follow
Always current
These links resolve live: what you get is generated or filtered the moment you click.
Fresh from the field feed refreshed July 8, 2026
- Jul 7 AI Search: Is Your Content Strategy Accidentally Recommending Your Competitors? Search Engine Journal
- Jul 7 SEO Study: 5 Lessons From Running AI Agents Across Every Search via @sejournal, @lorenbaker Search Engine Journal
- Jul 7 Google Search Console Adds Reports For Social Posts via @sejournal, @MattGSouthern Search Engine Journal
- Jul 7 Used or cited: The two ways brands appear in AI search Search Engine Land