The GPT-5 model can rebuild your creative funnel if you know what to feed it (and I'm showing you how).

Table of Contents

• Well it is GPT-5’s launch week. Big vibes, unclear …

  • Founder transcripts with GPT5

• GPT5 projects are a massive unlock

• White-labeling is still free money (and so easy to …

  • That’s it for this week

Remember when Facebook dropped Ads Manager and everyone lost their minds?

No? Me neither.

I was still in college listening to Rather Be and smoking reefer from Lewiston, Maine (you know, that real good good). #gobobcats

Well it is GPT-5’s launch week. Big vibes, unclear outcome.

OpenAI says it’s the smartest model ever.
Betting markets say you could’ve 5x’d your money betting against it.

The biggest thing is that o3 and 4o are one mind now. But now you just have to whisper sweet nothings into your prompt to make it think big brain style.

Other times it just spits out a 7th grade book report in soft pink Comic Sans (not a joke, that’s a real feature, you can now change the color of your chat bubble, we’re officially in the AGI era).

Also, and I can’t stress this enough,
THE MODEL DOESN’T KNOW HOW TO GRAPH.
(Like... the bars are lying. They visually misrepresent the data. Bruh)

The biggest kicker, the em dash is still alive and well. So glad it gets to perpetually grace us with its presence—whether we want it to or not.

Jokes aside, one big win I see thus far with GPT-5:

You can finally upload a transcript and build ad creative off of it without the model hallucinating 90% of the time.

This used to be a nightmare. You’d drop in a founder interview, ask for hooks or copy ideas and GPT-4 would start inventing fake quotes, imaginary backstories, and somehow turn “we started in 2021” into “after surviving a volcanic eruption, the founders knew their mission was clear.”

GPT-5? Much tighter leash.
It stays closer to source, respects tone, and actually quotes things you've given it.
(Still not perfect, it’ll occasionally paraphrase, but we’re moving in the right direction.)

I still like to get it negative prompts to avoid hallucination, but this is the biggest win I’ve seen so far.

An Actual Massive Performance Unlock: Mining and storing insights in GPT

I may be late to the game here, but low key the projects in GPT make creative strategy 1000x easier.

Here’s the workflow:

1. Create a dedicated GPT project for your ad account.
This becomes your creative war room. Drop in old ads, reports, screenshots from Triple Whale, Motion, Atria, Google Drive folders full of random UGC: the whole mess.

2. Upload performance data and creative assets.
You don’t need to clean it all. GPT can handle ugly spreadsheets, half-baked Airtable exports, and screenshots with red circles drawn in Figma.

3. Ask GPT to tag every ad by PAM:

• Persona: who this ad was for

• Angle: what it leaned on (education, social proof, emotional trigger)

• Motivation: why that person should care

This forms the DNA of every creative you’ve run.

4. Generate your PAM Cluster Map (this may be a little extra, and I would say this is optional).
Now you can visualize your entire creative strategy at a glance.

• X-axis = creative similarity (how tightly each ad aligns on persona, angle, motivation)

• Y-axis = CPA (lower is better)

• Bubble size = spend

• Faded dots = outliers

• Dark clusters = tested ideas with density

• Gold rings = current top performers

Instead of guessing what to scale, you can see it. You know where you're over-testing, where you're underinvested, and where new angles are starting to rise.

5. Layer in trend data.
This is where things get interesting.

• Upload a TikTok trend report from Trendpop

• Or just scrape popular sounds, formats, and content themes from your feed

• Feed this into the same GPT project and prompt it like this:
  “Take my top-performing PAM clusters and remix them into native TikTok creative ideas based on current trends.”

GPT can now turn your ADHD burnout explainer into a “Day in the Life” voiceover.
It can wrap your highest-performing hook into the structure of a trending meme.
It can even match soundbites from existing testimonials to current audio.

6. Keep feeding the loop.
Every time you drop in new performance results or test results, GPT gets smarter. Your cluster map updates. Your remix suggestions get sharper. Your strategy compounds.

I am still experimenting more and more on this, but I will continue sharing my findings. I genuinely believe this is a massive unlock.

White-labeling is still free money (and so easy to do)

Harry Delmege said it best: white-label pages are outperforming everything right now. If you’re not running ads from one yet, make it your priority this week.

Here’s what we mean by white-labeling:

You spin up a new Facebook Page or Instagram account that’s not tied to your brand name, but is aligned to one of your core personas. Then you run ads from that page.

You don’t need a huge organic following (or any following for that matter). You just need:

• A relevant name

• A decent profile photo

• And one good ad to start running that is relevant to that page

That’s it. You now have a parallel media channel targeting the same customer from a fresh angle, one that often converts better because it feels more native.

Examples:

• Selling non-alcoholic beer? Create a page called “BeerSnobs” or “Sober Curious Club”

• Selling wellness supplements to moms? Try “Healthy Mom Hacks”

• Selling CGM tech? Go with “Biohack Nation”, “Glucose Diaries”, or “Data-Driven Dad”

Each one maps to a proven Persona + Angle + Motivation combo… aka, a PAM cluster.

And yes, you can create multiple.
If you’ve got three top-performing PAM clusters, you can spin up three whitelabel pages and test each as its own creative container.

Now here’s where GPT-5 makes it stupid simple:

1. Feed GPT your product, audience, and top-performing creative clusters

2. Ask it to generate 20–50 whitelabel page ideas, aligned to those clusters

3. Pick a few

4. Do yo thang

That’s it for this week

Between the GPT-5 launch, hallucinated bar charts, trend-chasing, and vending-machine prompts, it’s been a weird one. But under the noise, the tools are getting better, and the workflows are getting faster.

If you only take one thing from this email:

Use GPT-5 like a strategist, not a novelty.
Feed it the good stuff. Tag what worked. Remix what’s winning.
Then run it through a whitelabel page and let the market decide.

And if you’re sitting on great creative but still guessing which persona it’s actually speaking to, go set up your PAM clusters. Let GPT label it for you. It’s boring, but it’s the cheat code.

As always, reply and tell me what’s working, what’s not, and what you want a teardown of next.

Till then: launch weird things, test fast, and stay slightly unhinged.

Have a great weekend!

Will