Insights & how-to

How to Write Better GPT Prompts: 5 Principles With Examples

By the daglo team · October 8, 2026

Better GPT Prompts, With Real Examples

Do you still type "write a report" into ChatGPT, then resend the same request over and over when the answer misses the mark? Each round makes the answer longer while the part you actually needed is still missing, and you end up writing it yourself anyway.

A GPT prompt is the request you give an AI like ChatGPT when you hand it a task. OpenAI's official documentation recommends building a good prompt from four parts: identity (role), instructions, examples, and context. This guide turns the core of OpenAI's prompting guidance, as of October 2026, into five principles, then compares before-and-after prompts for reports, emails, and summaries. Finally, you'll see how to run the same prompt through several AI models side by side in daglo.

Why does ChatGPT give you the wrong answer?

Office worker holding their head in frustration in front of a laptop

The most common reason is that the AI doesn't know your goal or situation. "Write a report" leaves out who will read it, what decision it supports, and what format you need, so the model fills the gaps with a generic answer.

OpenAI explains the difference with a workplace analogy. A reasoning model is like a senior coworker who can figure out the steps once you share the goal, while a standard GPT model is more like a junior coworker who does best with explicit steps and criteria. Either way, the goal and the standard for success need to come first.

5 GPT prompt principles from OpenAI's guidance

Hand writing in a notebook with checkboxes

A strong prompt states the goal, success criteria, constraints, and context first, leaves the path to the model, and then checks the result against examples.

  1. Lead with the goal. Say what the output is for, such as "so my manager can decide whether to raise next week's ad budget."
  2. Define success. Criteria like "conclusion in the first sentence, no more than three supporting numbers" let the model check its own work.
  3. Add constraints and context. Include length, tone, and words to avoid, and paste in source material such as meeting notes or last week's report.
  4. Start short and test with examples. Begin with a brief prompt, show one or two examples of the answer you want, check the result, and refine from there.
  5. Specify the output format. Ask for a table, bullet points, or a ready-to-send email so you spend less time reformatting.

Before-and-after prompts for reports, emails, and summaries

Focused office worker typing on a laptop at a desk

The same task produces very different results once you add a goal, criteria, and format. Here are three everyday examples.

TaskBeforeAfter
Weekly reportWrite this week's reportWrite a weekly report that helps my manager decide whether to raise next week's ad budget. Put the conclusion in the first sentence and use no more than three numbers, taken only from the attached data. Format: bullet points under Results, Issues, and Next week.
Work emailWrite an email asking the vendor to push back the dateWrite an email asking our vendor contact to move the delivery date back one week because of our internal review schedule. Apologize only once. Keep it polite and under five sentences, and suggest a subject line.
Lecture or meeting summarySummarize thisSummarize the meeting notes below for a teammate who missed the meeting. Use a table with decisions, action items with owners, and open questions. Don't guess anything that isn't in the notes.

Every improved prompt answers three questions: who will read it, what it should help decide, and what format it needs. Adding just those three makes a visible difference right away.

Compare the same prompt across AI models in daglo

Coworkers working on laptops together in a bright office

The same prompt can produce answers with different length and tone depending on the model. daglo's all-in-one AI chat gives you ChatGPT, Claude, and Gemini in one app, so you can run one prompt and see which answer fits your work.

  • Compare answers by model. Paste in the improved prompts from the table above and see which model handles reports best and which writes your emails best.
  • Use transcripts as context. Transcribe a meeting or lecture with daglo, then paste the transcript into the context part of your prompt so the AI answers from what was actually said.
  • Keep asking with Board Chat. Ask follow-up questions in Board Chat on a board that holds your transcripts, and get answers grounded in the whole set of records.

Always get consent before recording, and remove personal or sensitive details before adding anything to a prompt.

Start writing better GPT prompts today

Pick the task you do most often and add three lines in front of your prompt: the goal, the success criteria, and the format. If the answer still misses, show one example and add one more rule at a time. Model names and detailed recommendations change often, so check OpenAI's official documentation before using a prompt for important work.

#GPTPrompts #ChatGPTPrompts #PromptWriting #PromptEngineering #AIProductivity #WorkAutomation #ReportWriting #daglo

👉 Compare AI answers free

← Back to blog

Where talk becomes work.

Get started with 1,000 credits a month — no credit card required.