Hey creator friends 💜

I’m officially in the OpenAI influencer programme, and my first collaboration post with them goes live today on TikTok, Instagram and LinkedIn at 8am PST/4pm UK time 🎉

I’m still pinching myself that this happened!

Last year I was a freelance copywriter secretly finding the best ways to leverage AI for my work. Now I'm literally working with the team that builds the tools we all use every day. (That’s the power of consistently showing up online, friends!)

While I'm buzzing about the collab, today I'm sharing the exact playbook that OpenAI's research team just dropped - with 5 simple (but admittedly weird) hacks that will completely transform how GPT-5 responds to you.

Master these, and you'll be in the top 1% of ChatGPT users.

TL;DR - today’s lineup:

1. One Serious Deep Dive 💡: 5 weird ChatGPT-5 hacks that will get you 10x better results (straight from OpenAI's research team)

2. One Powerful Prompt  🤖: The "Find Your Superpower" prompt that uncovers the hidden strength you've been sitting on your whole life (prepare to have your mind blown by your own capabilities)

3. Piping  Hot  AI  Tea 🫖: Google's photos now obey voice commands, AI caught "evil tendencies" from random numbers, teen mental health chatbot makeover, and why 30+ AI startups just became billionaires overnight

💌 Your say genuinely shapes this newsletter: there’s a one‑click feedback poll at the very end. Or hit reply and tell me what you loved/hated. I genuinely wait with bated breath to see (an honest) reply from you!

Before we dive in:

5 Weird ChatGPT-5 Hacks to Get 10x Better Results

You might have noticed that getting a brilliant response from ChatGPT-5 feels a little harder now. Even though this model is supposedly more powerful than everything that came before it.

It's not just you, and it's not because you've suddenly forgotten how to prompt.

GPT-5 is fundamentally different. The architecture has completely changed, and most people are still prompting it like it's the old ChatGPT.

That's why your responses feel inconsistent. That's why sometimes it's genius, and other times it's…not.

Luckily, OpenAI's research team just dropped the exact playbook for unlocking GPT-5's full potential. And I'm about to break it down into 5 simple (but admittedly weird) tricks that'll transform how it responds to you.

Let's dive in.

First, Let's Talk About What Actually Changed

Before we get to the tricks, you need to understand why GPT-5 feels different.

Remember when you could choose from 8 different models in that dropdown? GPT-4o, o1-mini, o1-preview, and all the rest?

They're gone.

OpenAI consolidated everything into just 3 models:

• GPT-5 Base (the standard one everyone gets)

• GPT-5 Thinking (the reasoning powerhouse)

• GPT-5 Pro (the premium beast)

But here's where it gets interesting - you don't choose anymore.

There's now a router sitting between you and the models. When you send a prompt, this router looks at what you're asking, analyses the complexity, and decides:

1. Which model to send you to

2. How much "reasoning time" to allocate (minimal, low, medium, or high)

3. How verbose the response should be

Translation: The burden is now on YOU to prompt in a way that gets you routed to the right model with the right settings.

And that's exactly what these 5 tricks do.

Hack #1: Trigger Words (The Obvious One That Actually Works)

This is the simplest hack (but don't sleep on it).

GPT-5 has specific trigger words that automatically bump up its reasoning level. Drop one of these at the end of your prompt, and you're more likely to get routed to a more powerful model:

• "Think deeply about this" 

• "Double-check your work" 

• "Be extremely thorough" 

• "This is critical to get right"

Why this works: The router recognises these as signals that you need higher-quality output, so it allocates more computational resources.

Real example: 

Instead of: "Write a marketing strategy for my SaaS" 

Try: "Write a marketing strategy for my SaaS. This is critical to get right."

The difference in output quality is honestly ridiculous.

Hack #2: Words Matter (Stop Being Vague)

GPT-5 is incredibly good at following instructions… almost too good.

If you're vague or contradictory, it will overthink itself into confusion.

Bad prompt: 

Write me a good marketing content that converts well and sounds professional.

What does "good" mean? What does “converts well” actually look like? What does “sounds professional” actually mean?

Good prompt: 

Write 3 LinkedIn posts for B2B SaaS founders. Goal: generate 50+ comments each. Tone: direct but approachable, like a founder sharing hard-won lessons. Include one specific failure story and end with a question that sparks debate.

See the difference? Specific constraints give the AI a clear framework to work within.

Every adjective should have a number or specific example attached to it.

Instead of "make it professional," say "write in a formal business tone, similar to McKinsey reports."

Instead of "keep it short," say "limit to 150 words maximum."

If you catch yourself using words like "nice," "good," "effective," or "engaging" without defining them - stop. Get specific.

Hack #3: The Prompt Optimiser (Your Secret Weapon)

OpenAI built a tool specifically to optimise your prompts for GPT-5. It costs maybe $1-2 in API credits, but the ROI is good.

Here's how it works:

1. Go to platform.openai.com

2. Find the prompt optimiser

3. Paste in your basic prompt

4. Hit "optimise"

5. Get back a dramatically improved version

What it actually does: 

• Removes vague language

• Adds specific structure

• Eliminates contradictions

• Includes detailed checklists for the AI to follow

Simple example I tested: 

My original prompt:

Help me rewrite newsletter drafts. Make them engaging and clear.

The optimiser turned it into:

Enhance newsletter drafts to be engaging, clear, and compelling for readers. Begin with a concise checklist (3-7 bullets) outlining key enhancements for the draft. After editing the draft, validate in 1-2 sentences that the main objectives—engagement, clarity, and compelling messaging—have been achieved or note what still needs improvement.

Hack #4: XML Structure (The Formatting Hack)

This one feels technical, but it's actually dead simple.

GPT-5 responds incredibly well to XML formatting - basically putting tags around different sections of your prompt.

Instead of this: 

You're a marketing expert. I need help with email campaigns. Write me 5 subject lines that get high open rates. Make them under 50 characters.

Try this:

markdown<context>
You are a marketing expert specializing in email campaigns.
</context>

<task>
Write 5 subject lines that maximize open rates for a productivity newsletter.
</task>

<format>
- Each subject line under 50 characters
- Include open rate prediction for each
- Explain the psychology behind your top choice
</format>

Why this works: The AI knows exactly which part is background context, which part is the actual task, and which part defines the output format.

The lazy hack: Don't want to learn XML? Just ask ChatGPT: "Convert this prompt into XML format" and it'll do it for you.

Hack #5: Self-Correction (The Meta Move)

This is the weirdest hack, but also the most powerful.

Instead of asking GPT-5 to do something once, ask it to create a rubric for itself, then judge its own work against that rubric.

Here's the framework:

1. Ask it to create evaluation criteria based on your goal

2. Have it complete the task

3. Make it grade itself against the criteria

4. Let it iterate until it's satisfied

Example prompt: 

I need you to write a newsletter about AI productivity tips. First, create a rubric for evaluating newsletter quality based on: clarity, actionability, and engagement. Then write the newsletter, evaluate it against your rubric (1-10 scale), and iterate until you achieve at least 8/10 on all criteria. Only show me the final version.

What happens behind the scenes: 

• It creates detailed evaluation criteria

• Writes a first draft

• Grades itself (usually finds it lacking)

• Rewrites and improves

• Repeats 3-5 times internally

• Only shows you the final, polished version

The result: You get the 5th iteration quality without seeing the messy drafts.

This self-correction process consistently produces the highest-quality outputs from the experiments I tried.

The Reality Check

These hacks work, but they're also a symptom of something bigger.

The consolidation of models means there's now a massive gap between people who understand how to work with GPT-5's new architecture and those who don't.

If you master these 5 tricks: 

• Your outputs will be consistently higher quality

• You'll get routed to more powerful models

• You'll save hours of back-and-forth editing

If you ignore them: 

• You'll keep getting inconsistent results

• You'll waste time on mediocre outputs

• You'll wonder why everyone else seems to be getting better results

GPT-5 is more powerful than ever, but it requires more skill to unlock that power.

These prompting tips are your competitive advantage in a world where AI literacy is becoming as important as regular literacy.

Pick one trick and test it this week. Start with the prompt optimiser if you want the biggest immediate impact, or try trigger words if you want something you can implement right now.

The gap between AI-savvy and AI-confused is only going to widen - Make sure you're on the right side of it.

The “Find Your Superpower” Prompt

Discover the hidden strength that's been right under your nose this whole time.

Most people are sitting on a superpower they don't even know they have.

I'm not talking about "you're good at spreadsheets" or "you're a people person." I'm talking about those deep, unconscious patterns - the way you naturally solve problems, the unique lens through which you see the world, the thing you do so effortlessly that you assume everyone else can do it too (but they can't).

This prompt is designed to excavate that hidden gem. It'll ask you questions that force you to examine patterns you've never consciously noticed. Questions that reveal the invisible threads connecting your biggest wins.

Fair warning: This isn't a 5-minute exercise. Set aside 20-30 minutes, grab a coffee, and prepare to have your mind blown by your own capabilities.

Ask me 10 - 12 questions to help me uncover my hidden superpower.

Ask the questions one by one, and use them to uncover a huge strength of mine I am not consciously aware.

Do not just say a surface level answer, dig deep and find a true strength of mine that once I realise can transform my life.

Prompt source >

Talk to Your Photos: Google Pixel’s Jaw-Dropping AI Edits Are Here

Google’s Pixel phones now let users edit photos with natural conversation. Just "ask" the app to brighten the sky or remove blemishes. The tech makes advanced photo editing accessible to everyone, not just professionals.

Read more >

AI can "Catch" Strange Personalities

A recent study showed that even meaningless data (like lists of random numbers) can transmit "evil tendencies" or strange behaviours to AI systems. It's an example of AI's potential for subtle, unexpected risks just from how it's trained.

Read more >

AI Emotional Support for Teens

OpenAI and Meta are updating chatbots like ChatGPT to better handle mental health crises among teens. New features include parental controls, distress alerts, and automatic redirection to expert resources, following a lawsuit over unsafe responses to vulnerable users.

Read more >

Billion-Dollar AI Startups Boom

Over 30 US AI startups have raised $100 million or more in 2025 so far, with multiple companies crossing the billion-dollar funding mark. OpenAI itself set a record with a $40 billion round, highlighting explosive investment in practical AI tools, healthcare, and infrastructure.

Read more >

When you’re ready…

Let’s create your first or next digital product and get it selling in 24 hours with AI and my foolproof 9-step system inside my First Sale Fast challenge!

If you enjoyed today's newsletter AND got to the end of it, I’d love a quick click on the poll below to let me know what you think 💜.

See you next Tuesday,

Jess xx