OpenAI GPT-5 Prompting Guide
- Patrick Law
- Aug 25
- 2 min read
OpenAI quietly released a GPT-5 Prompting Guide — and it’s a game-changer for anyone using AI in engineering.
Here’s the truth: GPT-5 is powerful out of the box, but bad prompts = bad results.
That’s wasted time, wrong units, and even unsafe calculations.
Here's ho to make GPT-5 work like a reliable engineering assistant.
1. Role + Goal + Guardrails
Tell GPT-5 exactly what role it plays.
Example:
“You are a process engineer. Your goal is to size a centrifugal pump. Use SI units only.”
This structure prevents sloppy, unsafe answers.
2. Layer Your Context
Give info in the right order: background → rules → task. If you throw the task first, GPT-5 might skip important constraints.
3. Put Key Instructions Last
The last line of your prompt gets the most weight. Always end with the exact deliverable you need.
Example:
“Final answer: provide pump power in kW, rounded to 2 decimals.”
4. Use a Chain of Verification
Don’t trust single-step answers. Ask GPT-5 to solve step-by-step and check itself. This is crucial for calculations, reports, and safety cases.
5. Try Dual-Pass Answers
Have GPT-5 draft once, then re-check its own work with a rubric. Think of it as QA/QC inside the prompt.
6. Force “I Don’t Know” Honesty
7. Switch Perspectives
Ask GPT-5 to solve the same problem from different angles, then merge.
Example:
Pass 1: “Solve this pump sizing as a mechanical engineer.”
Pass 2: “Now solve as a chemical engineer.”
Compare → stronger final answer.
8. Control with Delimiters
Use tags or code blocks to control scope.
9. Prime with Examples
Show GPT-5 what “good” looks like before asking.
Example:
Good input → clear datasheet.
Good output → clean calculation table.
Bad output → messy text dump.
GPT-5 learns the format you want.
Without these rules: GPT-5 feels random.
With them: GPT-5 becomes a serious tool for:
Equipment sizing
Safety reviews
Process reports
Daily automation tasks
This is about saving hours and reducing errors in real workflows.
The better your prompt, the more reliable GPT-5 becomes.
Sources: Detailed breakdown from: How to use GPT-5: A Guide to Effective Prompting | Ruben Hassid posted on the topic | LinkedIn
For more AI-for-engineering updates, subscribe here:👉 Singularity Engineering Newsletter
Or, take the next step and learn here:👉 Advance your AI skills with our Udemy course
Comments