What makes a good AI assistant prompt — and why most companies get it wrong

The quality of your assistant's outputs is directly tied to how well it's instructed.

4 min read

What makes a good AI assistant prompt — and why most companies get it wrong

When an AI assistant gives a bad answer, most people assume it's a model problem. The AI wasn't smart enough. The technology isn't ready yet.

In our experience, nine times out of ten it's a prompt problem.

The way an AI assistant is instructed determines the quality of its outputs more than almost any other factor. And most companies haven't thought carefully about this.

What a prompt actually is

The system prompt is the set of instructions that defines how an AI assistant behaves. It's not a query — it's a constitution. It tells the assistant what it knows, how it should respond, what tone to take, what it should do when it's uncertain, and what it should never do.

A poorly written prompt produces outputs that are generic, inconsistent, or confidently wrong. A well-written prompt produces outputs that feel like they came from someone who actually understands the domain.

The most common mistakes

Being too vague — "You are a helpful assistant for our company" tells the AI almost nothing. Vague instructions produce vague outputs.

Not defining the boundaries — every assistant needs to know what it should not do. Without explicit boundaries, it will try to answer everything, including things it shouldn't.

Ignoring tone — how an assistant communicates is as important as what it communicates. If you haven't given it a clear sense of tone, it will default to a generic middle ground that doesn't feel like your brand.

Not updating it — prompts aren't set-and-forget. As the assistant is used in production, you'll discover scenarios it handles poorly. Updating the prompt regularly is part of maintaining a good assistant.

What good prompt design looks like

A good prompt includes: a clear description of who the assistant is and what it's for; the specific knowledge base it draws from; explicit instructions for handling uncertainty; a defined tone and communication style; and examples of good and bad responses in key scenarios.

Writing this well takes time. It requires understanding the workflow deeply enough to anticipate what users will ask. It's not a technical skill — it's a design skill.

At Tailo, prompt design is one of the things we spend the most time on. It's also one of the things clients are most surprised to learn about. The prompt is often where the real quality is made or lost.