Prompt Engineering Glossary: 50 Terms Defined

How many of these terms could you actually define on the spot: zero-shot, temperature, chain-of-thought, context window? If you use AI tools regularly but picked up the vocabulary piecemeal from scattered blog posts and Twitter threads, you probably know what half of these mean intuitively without being able to explain them clearly to a teammate. This prompt engineering glossary is built to fix that — not as a dictionary to memorize, but as a reference to actually use while you're writing prompts and talking to your team about them.

Quick-Start (Copy This Right Now)

Here are the core terms you'll run into constantly, grouped by category so you're not scanning fifty entries to find the one you need.

Foundational concepts

Techniques

What this does: grouping terms by function rather than alphabetically means you can scan the category relevant to what you're actually trying to do — reach for "Techniques" when you're building a new prompt strategy, "Foundational concepts" when you're troubleshooting unexpected behavior.

⚡ Pro tip: bookmark or save this glossary somewhere you'll actually reopen it. Vocabulary like this sticks fastest when you look a term up in the moment you need it for a real task, not when you read the whole list once and move on.

Understanding Why Terminology Actually Matters

This isn't pedantry for its own sake. Precise terminology prevents a specific, recurring failure: two people using the same word to mean different things, and neither realizing it until something goes wrong.

Real-world scenario — marketing team miscommunication: a marketing coordinator asked a colleague to "use few-shot prompting" for a client email template, but the colleague — new to the terminology — interpreted this as simply writing a shorter prompt, not providing multiple examples. The resulting prompt technically followed the instruction as understood, but produced inconsistent formatting the actual few-shot technique would have prevented. The confusion cost about an hour of back-and-forth that a shared, precise vocabulary would have avoided entirely.

⚠️ Common mistake: assuming a colleague shares your exact definition of a term just because they nod along in a meeting. Terms like "prompt chaining" and "prompt patterns" sound similar but mean genuinely different things — confirm shared understanding explicitly on anything that affects how work actually gets done.

Step-by-Step: Using This Glossary in Practice

Step 1 — When writing a prompt, name your technique explicitly in your documentation. Instead of just writing the prompt text, note which named technique you're using: "This uses few-shot prompting with 3 examples" tells a future editor exactly what to preserve if they modify it later.

Prompt (documented):
Technique: Few-shot prompting, 3 examples
Purpose: Guide consistent formatting for social media captions

[Example 1] → [desired output 1]
[Example 2] → [desired output 2]
[Example 3] → [desired output 3]
Now generate a caption for: [new input]

What this does: naming the technique explicitly turns an opaque wall of examples into something a teammate can recognize and reason about — they know this is few-shot prompting, they know the pattern, and they know changing the example count changes the technique's effectiveness.

Step 2 — Use precise terms when asking AI to help with meta-tasks. If you want the model to critique or improve your own prompt, naming the specific technique gets more useful feedback than a vague request.

Review this prompt and suggest whether adding chain-of-thought reasoning or a few-shot example would improve its consistency: [paste prompt]

What this does: naming specific techniques gives the model concrete options to evaluate against, rather than a generic "make this better" that produces equally generic suggestions.

Real-world scenario — freelance consultant explaining techniques to a client: an independent AI consultant working with a non-technical client used the term "chain-of-thought prompting" during a strategy call, then immediately followed with a plain-language explanation ("basically, we ask it to show its work before giving a final answer") — pairing precise terminology with an accessible translation kept the client engaged instead of lost in jargon, while still building their vocabulary over time.

⚡ Pro tip: when introducing a technical term to someone unfamiliar with it, always follow immediately with a one-sentence plain-language translation. This builds their vocabulary gradually instead of either overwhelming them with jargon or oversimplifying to the point of being useless later.

Pro-Level Variations

For technical teams, use precise terms in code comments and documentation exactly as defined here — "this function implements few-shot prompting with dynamically selected examples" is unambiguous in a way that "gives the AI some examples" isn't, especially months later when someone else maintains the code.

For cross-functional meetings, keep a shared glossary link handy (this one, or your own internal version) and reference it directly when jargon comes up, rather than assuming everyone already knows the term or stopping to explain it verbally every single time.

⚡ Pro tip: if your team uses AI tools daily, consider building a short internal glossary of your own team-specific shorthand alongside the general terms here — abbreviations and nicknames specific to your workflows benefit from the same clarity as the industry-standard vocabulary.

Troubleshooting Common Confusions

"Few-shot" vs. "prompt patterns": few-shot refers specifically to providing examples within a single prompt. A prompt pattern is a broader reusable structure — few-shot can be one ingredient within a larger pattern, but they're not interchangeable terms.

"Temperature" vs. "creativity": temperature is a specific numerical setting; "creativity" is a vague description of the effect. Referring to "turning up the creativity" is fine in casual conversation, but if you're actually adjusting a setting in an API call, use "temperature" so there's no ambiguity about what's being changed.

"Context window" vs. "memory": a model's context window is the amount of text it can process in a single exchange — it's not the same as persistent memory across separate conversations, which is a different feature entirely and not universally available across tools.

⚡ Pro tip: whenever you're unsure whether you and a colleague mean the same thing by a term, just ask directly — "when you say prompt chaining, do you mean sequential prompts or something else?" A 10-second clarifying question is cheaper than an hour of work built on a misunderstanding.

Keeping Your Glossary Growing

The 16 terms above cover the ones you'll hit constantly, but a genuinely useful prompt engineering glossary keeps growing as the field does — new techniques get named, existing ones get refined, and your own team develops shorthand that deserves the same clarity as industry-standard terms.

A few more worth knowing as you go deeper: prompt injection (malicious or unintended instructions embedded in input text that hijack a model's behavior), hallucination (confident output that's factually incorrect or fabricated), grounding (anchoring a model's output to a specific provided source rather than its general training knowledge), prompt drift (a saved prompt's output quality changing over time due to model updates, without the prompt text itself changing), and structured output (responses constrained to a specific parseable format like JSON rather than free-form prose).

Ask AI to help expand your own glossary:
"I use these AI techniques regularly: [list]. Are there 3-5 related terms I should know that I'm likely missing?"

What this does: using the model itself to surface adjacent terminology you haven't encountered yet is a fast way to fill gaps in your own vocabulary, especially for techniques that are newer or less commonly written about than the foundational ones.

⚡ Pro tip: revisit your glossary every few months rather than treating it as a one-time reference. Terminology in this space shifts quickly enough that a glossary from a year ago may already be missing terms that are now in everyday use on your team.

Your Turn

Vocabulary isn't the interesting part of prompt engineering — technique and judgment are. But shared, precise vocabulary is what lets a team actually build on each other's work instead of re-explaining the same concepts with slightly different words every time.

Keep this prompt engineering glossary handy the next time you're documenting a prompt, training a new team member, or just trying to remember what "chain-of-thought" actually means mid-conversation. A tool like PromptABCD lets you tag your saved prompts with the specific technique they use, so the vocabulary from this glossary and your actual prompt library stay connected instead of living in two separate places.