I’ve been using GPT-5 for a week, and the biggest lesson?

The biggest improvements don’t just come from the AI itself — they come from how you communicate with it.

Most of the time, when people think AI isn’t delivering, the real issue isn’t the model… it’s the prompt.

Here are the approaches that have made the biggest difference for me:

1. Start with one crystal-clear goalKick things off with a single, specific objective. Instead of something vague like “Summarise this and make it engaging,” say: “Create a 5-point summary of this article for a busy CFO.”

   
   

2. Give it a role and contextTell GPT-5 who it should be, who it’s talking to, and the situation. The same request given “as a marketing copywriter for a law firm” will produce a completely different tone than “as a healthcare data analyst.”

   
   

3. Set the rules upfrontDefine format, word count, tone, and any constraints before it starts. Example: “Max 120 words, UK spelling, plain language, no em dashes.”

   
   

4. Ask for a quick plan firstHave it outline 2–3 steps before doing the full task. This prevents wasted time and missed expectations.

   
   

5. Control the depth of thinkingThis one’s key — tell it whether you want a quick, surface-level answer or a deep, detailed one. Without guidance, it’ll usually default to somewhere in the middle.

   
   

6. Decide how hands-on you want to beYou can have GPT-5 run the whole process without checking in, or ask it to pause for your approval at any risky or creative decision points.

   
   

7. Make it check itselfGet it to review its own output for clarity, completeness, and alignment with your goal before it gives you the final result.

   
   

Using these steps has made my results sharper, faster, and much closer to what I actually need on the first try.

I’ve also put together my full GPT-5 Master Prompt Template — if you want it, it’s in the comments.