Prompt Engineering Tricks: 5 Ways to Make GPT-5 Work Dramatically Better

Most people are using GPT-5 completely wrong, and they don't even know it.

Since its launch, countless users have complained about getting worse results from GPT-5 – even though they're using the exact same prompts that worked perfectly with earlier versions. Sound familiar?

Here's the thing: GPT-5 is definitely more powerful than older models. So why are people struggling with it?

The answer is simple. OpenAI changed how GPT-5 works under the hood. The old tricks don't work anymore. But once you understand these changes, you can make GPT-5 work way better than before.

I'm going to show you five simple techniques that will transform your GPT-5 results. These aren't complicated - you can start using them today.

Why Your Old Prompts Aren't Working Anymore

Before we jump into the solutions, you need to understand what changed.

The Router Problem

GPT-5 now uses an invisible system that decides which model handles your request. Think of it like calling customer service – you explain your problem once, and they're supposed to route you to the right department.

Except GPT-5's router isn't very smart.

Sometimes you get the powerful model. Sometimes you get the basic one. And since the advanced models cost more to run, OpenAI's system tends to default to the cheaper, less capable option whenever possible.

The Precision Trap

GPT-5 is much better at following instructions exactly. This sounds great, but it's actually a double-edged sword.

Previous versions were pretty good at guessing what you wanted, even if your prompt was vague. GPT-5? Not so much. It follows your instructions to the letter – which means if your prompt is unclear or incomplete, you're getting subpar results.

The solution isn't to go back to GPT-4. It's to adapt your prompt engineering approach and learn modern prompting techniques that work with GPT-5's strengths.

Trick 1: Force Better Models with Router Nudge Phrases

Effort Level: Super Easy

This is probably the simplest trick you'll learn today, but it's incredibly powerful.

By adding just four words to the end of your prompts, you can force GPT-5 to use its more advanced reasoning models. This prompt engineering technique works by triggering the model's router system.

The magic phrase? "Think hard about this."

How It Works

Let's say you want investment advice. Instead of just asking:

"What are the pros and cons of putting my cash in a low-cost index fund versus a money market account?"

Try this instead:

"What are the pros and cons of putting my cash in a low-cost index fund versus a money market account? Think hard about this."

The Difference Is Huge

With the nudge phrase, you'll notice:

• A "thinking" indicator showing GPT-5 is processing longer
• More detailed, nuanced responses
• Second-order effects you hadn't considered
• Better structured answers with clear recommendations

Other Phrases That Work

Through testing, I found these phrases reliably trigger deeper reasoning:

• "Think hard about this"
• "Think deeply about this"
• "Think carefully about this"

Trick 2: Master Verbosity Control

Effort Level: Easy

GPT-5's router also controls how long or short your responses are. Instead of hoping you get the right length, you can control this directly with specific phrases.

Low Verbosity: When You Need Just the Essentials

Perfect for quick updates, Slack messages, or executive summaries.

Power phrase: "Give me the bottom line in 100 words or less. Use markdown for clarity and structure."

Example: Drafting a project update for your CEO? This phrase will give you a concise, well-formatted response that won't waste anyone's time.

Medium Verbosity: When You Need Context + Action

Great for team meetings, project updates, or explaining complex situations.

Power phrase: "Aim for a concise 3-5 paragraph explanation."

This gives you enough detail to understand what's happening and what to do about it, without losing your audience's attention.

High Verbosity: When You Need Comprehensive Information

Perfect for project briefs, research summaries, or reference materials.

Power phrase: "Provide a comprehensive and detailed breakdown, 600-800 words."

Trick 3: Use OpenAI's Secret Prompt Optimizer

Effort Level: Medium

Here's something 95% of users don't know: OpenAI has an official prompt optimizer tool↗ that rewrites your prompts specifically for GPT-5.

The Official Way

You can access this through OpenAI's developer platform. Just paste your prompt, click optimize, and get an improved version plus explanations for each change.

The Free Workaround

Don't want to set up a developer account? Use this meta-prompt with ChatGPT-5:

You are an expert prompt engineer specializing in creating prompts for ChatGPT-5. Your task is to take my prompt and optimize it for better results.
Please:

Add clear structure and organization
Eliminate vagueness and ambiguity
Include error handling for missing information
Explain your reasoning for each change
Here's my initial prompt:
[PASTE YOUR PROMPT HERE]

What the Optimizer Actually Does

After running hundreds of prompts through this system, I noticed three consistent improvements:

1. Adds structure – Breaks wall-of-text prompts into logical sections
2. Eliminates vagueness – Makes instructions more specific and actionable
3. Includes error handling – Tells GPT-5 what to do when information is missing

Trick 4: Create XML Sandwiches

Effort Level: Medium

If you've ever used the prompt optimizer, you might have noticed weird angle brackets in the results. These are called XML tags, and they're OpenAI's recommended prompt engineering method for structuring advanced AI prompts for GPT-5.

Why Structure Matters Now

With GPT-5's laser focus on following instructions precisely, clear organization isn't just helpful – it's critical.

Think of XML tags as labeled boxes. Instead of dumping everything into one paragraph and hoping GPT-5 figures it out, you're explicitly telling it: "This is background information, this is the task, this is the output format."

Before and After Example

Instead of this:

Help me prepare for a product manager interview. I'm applying for a senior PM role at a fintech startup. I have 5 years of experience in e-commerce and 2 years in financial services. The job requires expertise in payment systems, user onboarding, and A/B testing. Ask me some practice questions.

Try this:

<task>
Act as a hiring manager and ask me 3 realistic interview questions based on my background and the job requirements.
</task>

<background>
I'm applying for a senior product manager role at a fintech startup. I have 5 years of experience in e-commerce and 2 years in financial services.
</background>

<job_requirements>

Expertise in payment systems
User onboarding optimization
A/B testing experience
</job_requirements>

<tone>
Professional but conversational
</tone>

The Results Speak for Themselves

The structured version produces much more targeted, relevant questions that actually test the skills mentioned in the job posting.

Trick 5: The Perfection Loop

Effort Level: High (But Worth It)

This is the most advanced prompt engineering technique, but it's also the most powerful for complex AI tasks and professional prompt optimization.

Remember how GPT-5 is great at following instructions and critiquing itself? The perfection loop exploits both of these strengths.

How It Works

Instead of accepting GPT-5's first response and manually asking for improvements, you tell it upfront to:

1. Create its own definition of excellence
2. Grade its own work
3. Keep iterating internally until it achieves the best result

Real-World Examples

Example 1:

Write a market analysis report on the enterprise AI industry. Before you begin, develop an internal rubric for what constitutes a world-class market analysis report. Internally iterate and refine the draft until it scores top marks against your rubric.

Example 2:

Draft an outline for my quarterly business review presentation. Before you begin, create an internal rubric with 5 criteria for a perfect QBR. Then use that rubric to internally iterate the outline until your response scores 10 out of 10.

Universal Perfection Loop

Don't want to write custom iteration instructions every time? Add this to the end of any complex prompt:

Before providing your final response, create an internal quality rubric, evaluate your initial draft, and iterate until you achieve excellence. Show me only the final, polished version.

When to Use This

The perfection loop works best for:

• Creating finished documents from scratch
• Writing production-ready code
• Complex analysis or strategic planning
• Any zero-to-one creative task

My recent experience: Product Launch Strategy

I recently used the perfection loop to create a go-to-market strategy for a client's new AI tool. Here's what happened:

Simple prompt results: The first attempt gave me a generic 5-page strategy with the usual suspects – social media, content marketing, partnerships. It was fine, but felt like something any consultant could have written in 30 minutes.

Perfection loop results: The same request with the perfection loop produced something completely different. GPT-5 created its own rubric focusing on market timing, competitive differentiation, and measurable outcomes. Then it iterated internally, eventually delivering a 12-page strategy with:

• Specific launch sequences tied to competitor product cycles
• Detailed customer journey maps with conversion metrics
• Three alternative scenarios based on different market conditions
• Ready-to-use email templates and social media copy

The client said it was the most actionable strategy document they'd ever received. The difference wasn't just length – it was depth, specificity, and strategic thinking that actually considered real-world constraints.

Putting It All Together

Here's the beautiful thing about these prompt engineering methods: these AI optimization techniques aren't mutually exclusive. You can stack multiple prompting strategies for even better results.

For example, a prompt using multiple techniques might look like:

<task>
Write a comprehensive competitive analysis of our top 3 competitors in the SaaS project management space. Think hard about this.
</task>

<background>
[Your company context]
</background>

<output_format>
Provide a comprehensive and detailed breakdown, 800-1000 words. Use markdown formatting for clarity.
</output_format>

<tone>
Professional and analytical
</tone>

Before providing your final response, create an internal quality rubric for competitive analysis reports, evaluate your initial draft, and iterate until you achieve excellence.

This prompt combines:

• Router nudge phrases ("Think hard about this")
• Verbosity control (800-1000 words)
• XML structure
• Perfection loop

Quick Reference Guide

Low Effort, High Impact

1. Router nudge phrases – Add "Think hard about this" to important prompts
2. Verbosity control – Specify exactly how long you want responses

Medium Effort, Big Payoff

3. Prompt optimizer – Use the meta-prompt to improve your prompts
4. XML sandwiches – Structure complex prompts with clear tags

High Effort, Maximum Results

5. Perfection loop – For complex tasks that need multiple iterations

The Bottom Line

GPT-5 is incredibly powerful, but only if you know how to talk to it properly. The old "cross your fingers and hope for the best" approach doesn't work anymore.

These five prompt engineering strategies will help you get consistently better AI results from GPT-5. Start with the easy prompting techniques (router nudges and response control), then gradually work your way up to the more advanced prompt optimization methods.

The key is understanding that GPT-5 rewards precision and structure. Give it clear instructions in a well-organized format, and you'll be amazed at what it can do.

Remember: the goal of effective prompt engineering isn't just to get GPT-5 to respond – it's to get it to respond with exactly what you need, every single time through strategic AI communication.

After months of testing these techniques, I have to be honest: the router nudge phrases have been the most game-changing for my daily workflow.

Here's why this surprised me. I expected the perfection loop to be the winner – it produces the highest quality results, after all. But here's the thing: I use GPT-5 dozens of times per day for everything from email responses to research questions to brainstorming sessions. The perfection loop is incredible for big projects, but it's overkill for 80% of my interactions.

Router nudge phrases, on the other hand, work everywhere. Adding "Think hard about this" to a quick research question transforms a surface-level response into something genuinely insightful. It's the difference between getting Wikipedia-level information and getting analysis that actually helps me make decisions.

The compound effect is huge. When every interaction with GPT-5 is 30% more thoughtful and thorough, it changes how I work. I find myself relying on AI for more complex thinking tasks because I trust the quality of the output.

That said, XML structuring is a close second for anyone doing regular content creation or complex analysis. Once you get used to thinking in terms of <task>, <background>, and <output_format>, your prompts become so much clearer that you rarely need to iterate.

You can test all these techniques easily using Kerlig, and once you find prompts that work well for your specific use cases, save them as custom actions for instant reuse.