Skip to main content
Once the basics work, a few techniques make answers more reliable and easier to repeat. This page covers structured prompts, examples, step-by-step work, and how to turn a good prompt into something your team can reuse in 48°.

Structure long prompts

For longer prompts, separate instructions from material with headings or tags. The AI then knows what to follow and what to work on.

Show an example

If you need a specific style or format, show one example of a good result. One or two examples are usually more effective than a long description.

Let the model reason

For analysis, planning or calculations, ask the AI to work step by step, or to list its assumptions before answering. Models that support thinking can also reason before they answer. Raise the Thinking level in the chat, or set a default in Chat settings. Higher levels are more thorough but slower.

Split big tasks

Break large work into steps and check each one before moving on:
  1. “Draft an outline for the report.”
  2. “Expand section 2. Keep the tone of section 1.”
  3. “Now check the whole report for contradictions.”
For research across many sources, use Deep Research. You approve a plan before it starts, and it returns a cited report.

Ask for questions first

When the task depends on details you haven’t thought through, ask the AI to interview you. With Ask Human turned on, it asks structured questions with answer options. See Questionnaires.

Turn good prompts into reusable building blocks

When a prompt works, keep it. 48° offers several ways, depending on how you want to reuse it:

Keep the context focused

Everything in a chat counts toward the model’s context window: the conversation, attached files, instructions and tool descriptions. When a chat gets long, answers can lose track of early details.
  • Start a new chat for a new topic.
  • Use Summarize & continue in new chat to carry the essentials forward.
  • Fork from the message where the relevant part begins.
See Context window.

Prompting basics

The four parts of a good prompt.

Assistant configuration

Write strong assistant instructions.

Creating skills

Package knowledge as a skill.

Model guide

Which model suits which task.