AI Prompting Glossary
Every term you need to know to write better prompts for ChatGPT, Claude, Gemini, and any AI tool. Save the best ones directly to PromptMan with ⌘⇧S.
A
AI model
A large language model (LLM) trained on vast amounts of text data to understand and generate human language. Current leading examples include OpenAI's latest models, Claude Sonnet 4.6 and Opus 4.7 (Anthropic), and Gemini 3.1 Pro (Google).
AI prompt
The text input you give to an AI model to generate a response. A well-structured prompt typically includes a role, context, task, format, and examples. PromptMan helps you save and reuse your best prompts.
Anchoring
A prompting technique where you provide an initial reference point, a number, example, or framing, that the model uses as a baseline for its response. Useful for tone, length, and style calibration.
C
Chain-of-thought (CoT) prompting
A technique where you instruct the AI to reason through a problem step by step before giving a final answer. Adding "think step by step" to your prompt significantly improves accuracy on math, logic, and complex reasoning tasks. OpenAI's latest reasoning research continues to build on this foundational technique.
Context window
The maximum amount of text (measured in tokens) that an AI model can process at once, including both your input and the model's output. OpenAI's current models support context windows of 128,000 tokens or more; Claude Sonnet 4.6 and Opus 4.7 support up to 200,000 tokens. Longer context windows let you include more background information in your prompt.
Constraints
Rules you add to a prompt to limit or shape the model's output. Examples: "respond in 3 bullet points", "do not mention competitors", "use a professional tone". Constraints are one of the most powerful prompting tools and should be saved as reusable templates in PromptMan.
F
Few-shot prompting
Including 2 to 5 examples in your prompt to show the AI the exact output format or style you want. The model learns the pattern from the examples and applies it to your actual question. Highly effective for formatting tasks, classification, and structured data extraction.
Fine-tuning
The process of further training a pre-trained AI model on a specific dataset to improve its performance on a particular task or domain. Unlike prompt engineering, fine-tuning modifies the model's weights and requires technical expertise and labelled training data.
Follow-up prompt
A second message sent to the AI after its initial response, used to refine, expand, or redirect the output. Follow-up prompts are part of multi-turn conversations and are most effective when they reference the previous response explicitly.
G
Grounding
Providing factual context in a prompt to keep the model's response anchored to real information. For example, pasting in a document or article and asking the model to answer based only on that content reduces hallucinations and improves accuracy.
H
Hallucination
When an AI model generates confident-sounding but factually incorrect information. Hallucinations occur because LLMs predict statistically likely text rather than retrieving verified facts. Strategies to reduce them include grounding prompts with real source material, using RAG, and asking the model to say "I don't know" when uncertain.
I
In-context learning
The ability of an LLM to learn a task from examples provided directly in the prompt, without any changes to the model's weights. Few-shot and zero-shot prompting are both forms of in-context learning.
Instruction following
A model's ability to understand and execute explicit instructions in a prompt. Modern instruction-tuned models, like Claude Sonnet 4.6, Opus 4.7, and current OpenAI models, are specifically trained to follow natural language instructions precisely, making them more reliable for structured prompt workflows.
J
Jailbreak
A prompt designed to bypass an AI model's safety guidelines or content filters, tricking it into generating content it would normally refuse. AI providers continuously update their models to resist known jailbreaks.
K
Knowledge cutoff
The date after which an AI model has no training data. For example, if a model's cutoff is January 2025, it won't know about events after that date unless you provide that information in the prompt.
L
LLM (Large Language Model)
A large language model (LLM) trained on vast amounts of text data to understand and generate human language. Current leading examples include OpenAI's latest models, Claude Sonnet 4.6 and Opus 4.7 (Anthropic), and Gemini 3.1 Pro (Google). The quality of your prompts directly determines the quality of the output you get from any of these models.
Prompt library
A collection of saved, reusable prompts organized for quick retrieval. PromptMan is a native macOS app that gives you a personal prompt library accessible from any application via ⌘⇧O.
M
Multi-turn conversation
A conversation with an AI model that spans multiple messages, where the model remembers context from earlier in the thread. Effective multi-turn prompting involves building on previous responses rather than starting fresh with each message.
O
One-shot prompting
Providing exactly one example in your prompt to guide the model's output format. It sits between zero-shot prompting, with no examples, and few-shot prompting, with 2 to 5 examples.
Output format
The structure you request for the AI's response, e.g., JSON, markdown table, numbered list, plain paragraph, or bullet points. Specifying the output format in your prompt is one of the easiest ways to get more consistent, usable results.
P
Persona prompting
Assigning a specific role or persona to the AI in your prompt, for example, "You are a senior copywriter at a tech startup" or "Act as a Socratic tutor who only asks questions." Persona prompting shapes tone, depth, and style of the response.
Prompt chaining
Breaking a complex task into a sequence of smaller prompts, where the output of each step becomes the input of the next. Prompt chaining is more reliable than asking for everything in one long prompt, especially for complex workflows.
Prompt engineering
The practice of designing, testing, and refining prompts to get better outputs from AI models. Good prompt engineering involves clarity, context, examples, and iteration. OpenAI's official prompt engineering guide is the most up-to-date reference.
Prompt injection
A security vulnerability where malicious text in user input overrides the intended system prompt instructions. A concern in AI-powered applications that process untrusted user input before passing it to an LLM.
Prompt template
A reusable prompt structure with placeholders for variable content. For example: "Write a [tone] email to [recipient] about [topic] in under [word count] words." PromptMan lets you save templates and fill in the variables before pasting.
R
RAG (Retrieval-Augmented Generation)
A technique where an AI system retrieves relevant information from an external knowledge base and includes it in the prompt before generating a response. RAG dramatically reduces hallucinations and keeps answers grounded in real, up-to-date data. Used widely in enterprise AI assistants.
Role prompting
Telling the AI what role to play before asking your question. Similar to persona prompting but often more specific to a professional context, e.g., "You are an expert tax accountant in Germany." Role prompts prime the model's vocabulary and reasoning style.
S
System prompt
An instruction given to an AI model before the conversation begins, typically invisible to end users in deployed applications. It sets the model's persona, rules, tone, and constraints. Example: "You are a helpful customer support agent for PromptMan. Always be concise and never discuss competitors." System prompts are among the most valuable templates to store in PromptMan.
T
Temperature
A parameter that controls the randomness of an AI model's output. Low temperature, from 0 to 0.3, produces consistent, deterministic responses that are ideal for factual tasks. High temperature, from 0.7 to 1.0, produces more creative and varied outputs that are better for brainstorming and creative writing.
Token
The basic unit of text that AI models process. A token is roughly 4 characters or 0.75 words in English. "Hello, world!" is 4 tokens. Tokens determine both context window limits and API pricing. More detailed prompts use more tokens but often produce better results.
Tree of thought
An advanced prompting technique that asks the AI to explore multiple reasoning paths simultaneously, evaluate them, and select the best one. More powerful than chain-of-thought for problems with multiple valid solution approaches. Explore Anthropic's 2026 research on tracing model reasoning for a deep dive into how modern models think.
Z
Zero-shot prompting
Asking the AI to perform a task without giving any examples, relying entirely on what the model already knows from training. Zero-shot works well for simple, well-defined tasks. For complex or format-specific tasks, few-shot or chain-of-thought prompting performs better.
Save your best prompts. Use them anywhere.
PromptMan stores every prompt you write so you never have to retype from scratch. Hit ⌘⇧O from anywhere on your Mac.
Join the Waitlist