What is a prompt in AI? A prompt is the text you give an AI model so it knows what to produce: your question or instruction, plus any context and examples you add. OpenAI's prompt engineering guide describes using a large language model (an AI model trained on vast amounts of text) "to generate text from a prompt, as you might using ChatGPT."
What is a prompt in AI? Your instruction, plus the context and examples you add
Anthropic's glossary defines large language models (LLMs) as AI models "trained on vast amounts of text data" that "can generate human-like text, answer questions, summarize information, and more." The prompt tells one of these models which job you want and what to work with, unlike a search query, which looks for pages that already exist. When you type "Summarize this email in three bullet points," that sentence is your prompt.
The model reads your prompt inside its context window, which Anthropic describes as the amount of text a model "can look back on and reference when generating new text," a "working memory" separate from what it learned in training. Your text is encoded into tokens, the small units that Anthropic says "can correspond to words, subwords, characters, or even bytes." Anthropic adds that a larger context window lets a model "process and respond to more complex and lengthy prompts."
To see what the model does with your prompt next, read how AI makes decisions, explained simply.
An AI assistant can run on two prompts: the builder's and yours
When you chat with an AI tool, your message can be only part of what the model reads. OpenAI's guide explains that "developer messages are instructions provided by the application developer, prioritized ahead of user messages," while "user messages are instructions provided by an end user, prioritized behind developer messages." In OpenAI's words, developer messages "provide the system's rules and business logic," and user messages provide the inputs those rules are applied to.
Anthropic calls the developer's base instructions the system prompt. Its prompting guide says "setting a role in the system prompt focuses Claude's behavior and tone for your use case. Even a single sentence makes a difference." So an assistant that answers a business's customers works from instructions the customer doesn't write: who the assistant is, what it knows and where it stops.
A good prompt is clear, explains why and shows examples
Anthropic's guide asks you to "think of Claude as a brilliant but new employee who lacks context on your norms and workflows," and adds that "the more precisely you explain what you want, the better the result." Its key test is to show your prompt to a colleague with minimal context on the task: "If they'd be confused, Claude will be too." The same guide tells you to be specific about the output format and constraints you want.
Context helps too. Anthropic says that explaining the reason behind an instruction "can give Claude more information about your goals and help it deliver more targeted responses." OpenAI says it is often useful to include extra context within the prompt, for example "to give the model access to proprietary data, or any other data outside the data set the model was trained on," such as your own business information.
Examples are the third tool. Anthropic calls them "one of the most reliable ways to steer Claude's output format, tone, and structure" and suggests including 3 to 5 examples for best results. OpenAI calls this few-shot learning, which steers a model toward a new task "by including a handful of input/output examples in the prompt, rather than fine-tuning the model," meaning without retraining it.
Here is the difference in practice:
- Vague: "Write a reply to a customer."
- Clear: "You answer messages for a dental clinic. Reply in two friendly sentences and offer the two open slots, Tuesday at 10:00 or Thursday at 16:00. If the customer asks for a treatment price that is not on the list, say you will check with the dentist."
The clear version gives the assistant a role, a format, two real slots and a way out when it doesn't know. Role, format and a way out are the same prompt techniques Anthropic describes.
Limits in the prompt reduce made-up answers, and sensitive queries go to you
OpenAI's guide says "the content generated from a model is non-deterministic," meaning the same prompt can return different answers, so prompting for the output you want "is a mix of art and science," although "you can apply techniques and best practices to get good results consistently." Anthropic's guide on hallucinations (answers a model makes up) says even the most advanced language models "can sometimes generate text that is factually incorrect or inconsistent with the given context."
Anthropic's first fix for made-up answers is part of the prompt itself: explicitly give the model permission to admit uncertainty, a technique it says "can drastically reduce false information." It also suggests telling the model "to only use information from provided documents and not its general knowledge," while noting that these techniques "don't eliminate them entirely" and that you should always validate critical information, especially for high-stakes decisions.
In MAGIA Solo, Catalizadora sets up the assistant to stay within what it knows about your business and not invent prices or promise things you don't offer. When a query gets sensitive or falls outside its scope, the assistant hands it to you instead of improvising.
MAGIA Solo turns these basics into an assistant running in 15 days
If what you want is an assistant answering your clients, MAGIA Solo is Catalizadora's package for professionals and small businesses, such as doctors, lawyers, coaches, studios and clinics. In 15 days Catalizadora hands you a system that answers, books and gets you paid. It is not an experiment: Catalizadora has built more than 500 projects end to end, and systems are running today in 5 countries.
The prompt work this guide describes is done for you. In Phase 0 you give Catalizadora your information and examples of how you speak (services, prices, how you answer today), and Catalizadora trains the assistant's tone to sound like your brand, not a generic robot. Before delivery you test it, and Catalizadora adjusts it until it sounds right.
The assistant handles WhatsApp day and night with your voice and your prices. It offers your slots, books them and sends the reminder, and it shares a payment link right in the chat. You don't need to be technical, because Catalizadora delivers the system running with three training sessions and a simple manual, and during the 6 months by your side it adjusts any answer you don't like.
Code, data, domain, accounts and even the keys are in your name from day one. MAGIA Solo is a one-time payment of 4,500 USD, and monthly operations (keeping the system online and the AI that answers) are paid at cost, directly to the providers and in your name. If you want the system to remain hosted and operated by Catalizadora, that requires an active plan, from 99 USD a month.
Now you know what a prompt is. If you'd rather have an assistant answering your clients in 15 days, with the prompt work done for you, see everything that is included and how the 15 days work on the MAGIA Solo page.