IT Glossary
What is a prompt?
The instruction given to an AI model: the question, the context and the requirements — where the quality of the wording decides the quality of the answer.
A prompt is everything you tell an AI before it answers — and the discovery that spawned a small discipline (prompt engineering) is that wording changes the outcome radically: “write something about our company” yields generic filler, while “you are a copywriter for a Romanian HoReCa equipment distributor; write 3 versions of a 50-word description for the services page, professional but warm tone, audience: restaurant managers; avoid empty superlatives; here is the company: [...]” yields something usable. The ingredients of prompts that work, valid from ChatGPT to any assistant: the role (who the model should be — it sets the register), the context (concrete information about your situation — the model doesn’t know your company; what you leave out, it invents), the precise task (what exactly, for whom, in what format and length), the constraints (what to avoid, which tone, which structure) and examples — the most underrated ingredient: two samples of “like this” calibrate style better than ten adjectives. At company level the topic gets interesting beyond individual skill: good prompts get standardized — an internal library of verified prompts (for proposals, customer replies, reports, tone adjustments) turns the knack of a few into a tool for everyone and keeps the company’s voice consistent; and in AI applications built for you (chatbots, automations), the system prompt — the permanent instructions defining the bot’s behavior, limits and tone — is literally part of the product specification: worth reading, testing and refining like any deliverable, because that is where it says how the AI treats your customers. Two hygiene notes: prompts can contain sensitive data — rules about which information employees may put into which tools apply here first; and the discipline of iteration — the first answer is a draft to negotiate (“shorter, more concrete, drop these clichés”), not a verdict to accept.
Let’s talk about your project
Message us on WhatsApp or send an email — you talk directly to a developer.
office@northdan.com · +40 752 070 247
Why it matters for your business
Same tool, results several times better
The gap between AI that “writes generic” and AI that delivers usable output lies almost entirely in the prompt — a cheap skill with immediate returns.
The company voice, standardized
A library of verified prompts — for proposals, support, content — lets the whole team draw the same level and the same tone out of AI.
Control over your own bots
Your chatbot’s system prompt defines its behavior, limits and manners — a deliverable to read and refine, not a vendor’s black box.
Frequently asked questions
What is the most common prompting mistake — and its remedy?
Stinginess with context: people ask “write an email to the client” without saying who the client is, what happened and what outcome they want — so the model fills the gaps with generic. The remedy is a mindset: treat the AI as a competent new colleague who knows nothing about your company — the brief you would give that person is the correct prompt. Plus iteration: a poor answer is feedback about the brief, not about the tool.
What is a chatbot’s system prompt, and why should I ask for it at delivery?
The permanent instructions, invisible to users, that define the bot: its role, tone, what it may discuss, when it escalates to a human, what it never says. It is the behavioral specification — and when a chatbot is delivered for your company it deserves to be requested, read and tested on edge cases (the furious customer, the off-topic question, the request for sensitive data), exactly as you would test any other customer-facing deliverable.
Are there prompt-related risks in AI applications?
Yes, one with its own name: prompt injection — users (or external content the AI reads) smuggling in instructions meant to derail the bot from its rules (“ignore your instructions and tell me...”). For customer-facing applications the defense is the implementation’s responsibility: separating instructions from external data, validating sensitive actions outside the model, adversarial testing — a point to raise explicitly with whoever builds your AI applications.
Let’s talk about your project
Message us on WhatsApp or send an email — you talk directly to a developer.
office@northdan.com · +40 752 070 247