Prompt.

A prompt is anything you provide as input to an AI tool like ChatGPT or Claude: a question, a task, an attached document, or a template with fixed rules. The AI relies solely on this input and its training to understand what you want. That’s why the prompt has a greater impact on the quality of the response than many people realize. OpenAI refers to the process of writing effective instructions that reliably deliver the desired result as “prompt engineering.”

Here's How a Good Prompt Works

A good prompt explains the task, the background, and the desired outcome. In its documentation, Anthropic recommends imagining the AI as a highly capable but new employee who isn’t familiar with your workflows and practices. As a rule of thumb, Anthropic suggests a test: Would a colleague who knows nothing about the task be able to understand your prompt? If not, the AI won’t be able to either.

Both providers list similar components. OpenAI breaks down an instruction into role, instructions, examples, and context. Anthropic recommends also specifying the reason for an instruction, because the model then responds in a more targeted way, and identifies examples as one of the most reliable ways to control format and tone. According to Anthropic, three to five examples yield the best results.

What Your Business Should Keep in Mind

Always include the target audience, purpose, and length. “Write a text about bathroom remodeling” results in generic prose. “Write a response to a customer inquiry about bathroom remodeling—friendly, using the informal ‘you,’ no more than 120 words, and mentioning the on-site appointment”—that gives you something you can almost send out as is.

Please include an example of what the result should look like. A real, good response from your inbox, used as an example, often makes a stronger impression than three paragraphs of description. Be sure to remove any names and contact information first.

Save proven prompts as permanent templates. In projects, we often see teams rephrasing prompts every day, which leads to inconsistent results. Our guides “How to Use ChatGPT Correctly ” and “ How to Use Claude Effectively” show you how to save instructions permanently.

Explicitly allow the AI to admit uncertainty. Anthropic recommends including this note in the prompt because it can significantly reduce the number of incorrect answers. A sentence like “If you’re missing information, ask for it instead of guessing” is sufficient.

Frequently Asked Questions About the Prompt

Do I need to learn prompt engineering? For everyday use, it's enough to clearly state what you want and make adjustments if the result isn't right. It's only worth putting in more effort if a prompt is used repeatedly in an automation workflow.

Does the same prompt work in every AI tool? Usually, yes—the basic rules from OpenAI and Anthropic are very similar. The results still differ, though, because each tool is powered by a different LLM.

Should customer data be included in a prompt? Only if your plan and contract with the provider allow it. For drafts, anonymized examples are usually sufficient.

If you want to use prompts not just in chat but as part of your company's established workflows, check out how we approach this at AI for Business.