5 Ways to Improve Your GPT-5 Prompts

So you’re not getting the results you expected with GPT-5? Here’s why – and how to fix it.

When GPT-5 launched, it promised more precision, deeper reasoning, and fewer hallucinations. And yes, it delivers. But here’s the part most users don’t want to hear: GPT-5 is far less forgiving than its predecessors. Unlike GPT-4, it won’t automatically fill in the gaps in your thinking. If your prompts are vague, contradictory, or lazy, the output will reflect exactly that.

if your results are poor, it’s not GPT-5’s fault. It’s yours.

The good news? Prompting GPT-5 well isn’t rocket science, it’s a discipline. And just like any other discipline in HR, recruitment, or business strategy, it takes structure and intention. Here are five ways (plus a bonus) to get exponential results from GPT-5.

1. Tell it to think more

GPT-5 has an “ultra-deep mode.” You unlock it by explicitly asking the model to think more deeply. Instead of a surface-level response, you’ll get layered reasoning. For example:

❌ “Give me some ideas for sourcing talent.” ✅ “Think more deeply and give me 5 innovative, underutilised sourcing strategies, ranked by potential impact.”

2. Tell it to plan

One of the easiest traps with GPT-5 is ambiguity. When you ask it to “plan before responding,” it validates its own approach before producing an answer. This avoids weak or misaligned results.

Example: HR has asked GPT-5 to create a talent map for emerging skills in AI. The output looks polished, but on review, half the job roles are irrelevant. Why? The team never asked GPT-5 to plan its methodology. The fix was simple: ✅ “Plan your approach to mapping AI skills, then provide the mapping. Show me your validation process.”

The difference? Night and day.

3. Be explicit

Tone, style, and format matter. GPT-5 is literal. If you don’t specify, you’ll get a generic response. Compare:

❌ “Write an email to candidates.”

✅ “Write a concise, professional email to senior executives, persuasive in tone, maximum 200 words, with a call-to-action in the final line.”

Structure is king. Think of your prompt as a job description: the clearer it is, the better the hire (or in this case, the output).

4. Ask it to show its thinking

This isn’t about transparency, it’s about control. When you ask GPT-5 to show its reasoning in bullet points first, you can intervene before it generates the full response.

✅ “Show me your reasoning in 3-5 bullet points, then draft the policy.”

This small step eliminates wasted tokens, saves time, and avoids misdirection.

5. Avoid contradictions

Contradictory prompts are kryptonite. If you ask GPT-5 to “be informal but highly professional” or “keep it short but detailed,” you’re feeding it impossible rules. The result? Confusion, wasted tokens, and poor output.

Instead, prioritise what matters most. Don’t throw the kitchen sink into one prompt.

Bonus: Structure is king

Here’s the truth: GPT-5 is a mirror. If your thinking is messy, your output will be messy. The tighter the structure of your prompt, the better the response.

Here’s a simple template to use:

Try thus propmpt framework:

1. Role → “Act as a…”

2. Objective → “Your goal is to…”

3. Constraints → “Use max 300 words, formal tone, bullet points only.”

4. Process → “Plan your approach first, then produce the answer.”

5. Output → “Deliver the result as a table and executive summary.”

Follow this, and you’ll see results transform overnight.

The uncomfortable truth

GPT-5 is not here to rescue sloppy thinkers. It’s a multiplier. If you prompt well, it will reward you with depth and clarity. If you don’t, it will expose your lack of precision. The real question is: are you willing to raise your game?

Additional thoughts!

1. Are you treating GPT-5 like a colleague to collaborate with, or a vending machine to dispense quick fixes?

2. When was the last time you audited your own prompts for clarity and structure?

3. If GPT-5 exposes your gaps in thinking, what does that say about your approach?