## 1. Executive Summary
Anthony Carreiro leads the creative team at Arthur PC's SIP Group. He started roughly two years ago as a local-traffic manager, got pulled into the Institute by Diogo Gomes (of "Lesbian Trick" fame), and hit a US Nutra ad that alone sold over 20 million. He now runs a creative team of eight copywriters split into two squads, and says the company he works with hit nine figures in a single month. His intro claims his ad framework has generated over a billion in sales. The episode is a full teardown of how he makes creatives: two ad types (predictability ads that beat the control in four stages, and scaling ads written from scratch), an ad-library keyword spy method, a 50-50 predictability/scaling split, lateral scaling of winning bodies, and a volume-over-quality production model where eight copywriters pump out roughly 240 ads a day and success is measured only by profit per writer. A large section covers AI: he built a custom vertical GPT agent ("GPTonin") that is fed his direct-response psychology and multi-8-figure ads so it can write scaling ads and "beat the control" on command, and he uses Google's VEO3 to generate demonstration visuals.

## 2. Alpha / Sauce - what actually moves the needle, and exactly how to run it
- **Hook is 50% of the ad.** If the hook fails, the ad is dead no matter how good the rest is. Treat the first seconds as the single highest-leverage slot and always open with a visual hook that stops the scroll.
- **Beat the control in four stages, not one.** Take an ad that already works and stack: (1) benefit expansion (layer benefits 2, 3, 4), (2) a value-based CTA with a real dollar amount plus a "reason why" it is free, (3) two fascination bullets after the first CTA, and (4) a second CTA at the very end. Add rather than replace.
- **Put a first CTA early, even in long ads.** Weight-loss ads run 2.5-5 minutes, so drop an express CTA within the first ~30 seconds for the people who are already sold, then keep the rest of the ad for the stragglers.
- **Split the batch 50-50 predictability vs. scaling (80-20 for juniors and offer tests).** Validated-with-a-twist ads keep a stable baseline; from-scratch ads carry the upside. Never test a new offer on 100% risky ads or you will blame the offer for a bad creative.
- **Find what works in the Facebook Ad Library using blended keywords.** Combine market keywords (trick, little recipe, simple recipe, root cause) with niche keywords (fat, weight loss, pounds; for ED: my husband, blue pill, Viagra) and offer keywords (the hook name, e.g., "banana trick"). Search any two words in any order; the platform matches both anywhere in the transcribed video.
- **Never trust the Ad Library's first screen.** One visible duplicate can hide 500 ads on the page. Check every page, identify the most-scaled ads, and imitate their pace.
- **Lateral scaling is infinite while the original still sells.** Take a winning body and write five new hooks with five different avatars. A winning ad body can last 3-4 months, and the podcast's own EMA ad has run over a year.
- **Volume beats quality in ad copy.** With 8 copywriters at ~30 ads a day, you cannot review anything. Test everything; let a single $200k/day winner cover everyone who went into the red. One ad makes it back.
- **Use a custom vertical GPT, not vanilla ChatGPT.** Feed an agent your DR knowledge, process, and winning ads. Send it the psychology (the idea) and let it handle communication (the words) for both "beat the control" and from-scratch scaling ads.

## 3. Methods & Setup - Step-by-Step
1. Establish the ad type you are writing: a predictability ad (beat the control in four stages) or a scaling ad (built from scratch); setup: decide based on whether the offer is validated and whether the front end already converts.
2. Source the market's working ads through the Facebook Ad Library using blended market + niche keywords (for example "trick" + "fat"); setup: type two keywords with no quotes or special operators and filter to video.
3. Go deep into each ad page instead of trusting the first screen one visible duplicate can reveal hundreds of ads; setup: open every page and note the most-scaled advertisers.
4. Add offer keywords by searching the hook name of the exact offer you are spying (e.g., "banana trick") to pull every ad running that mechanism; setup: use the offer's own hook name, not your offer's name.
5. For YouTube, spy with DeepTube (Hytalo's tool), which he calls the best available option; setup: run YouTube creative research through DeepTube to read competitor scripts.
6. Keep the market's proven ad as your baseline and write from-scratch ads alongside it, so you always hold both security and upside; setup: plan the week's batch as a 50-50 split (80-20 if the writer is junior).
7. Apply the first stage to the baseline: expand the benefit chain. Take "lost 12 pounds in 30 days" and layer benefit 2 (husband can't keep his eyes off her), benefit 3 (friends think she had surgery), and so on; setup: write benefits 2-4 for every validated ad before touching anything else.
8. Apply the second stage: make the CTA value-based by attaching a dollar amount and a reason why it is free (e.g., "my friend paid $27, but he's giving it free for 24 hours because he just hit 300k followers"); setup: rewrite the offer-block CTA with a real dollar value plus the reason why.
9. Apply the third stage: add two fascination bullets right after the first CTA using the transition "plus he is also going to reveal," covering problem mechanism and solution or curiosity; setup: insert two bullets immediately after CTA one in every predictability ad.
10. Apply the fourth stage: put a second CTA at the very end so anyone who did not click the first one gets another chance; setup: duplicate a CTA block at the tail of the ad and freeze it into the template.
11. For long weight-loss ads, insert an express CTA within the first 30 seconds instead of shortening the ad; setup: cut a 20-second express CTA and place it early in any ad over 2 minutes.
12. For ED ads, reduce clickbait (TikTok-girl, sexy visual hooks) and shift to scientific authority (a male health expert plus proof) so the VSL gets more qualified traffic; setup: brief writers to trade some clickbait ads for authority/proof-led ED ads.
13. Lateral-scale each winner by reusing its winning body with five new hooks across five avatars; setup: keep a library of winning bodies and match each to a fresh avatar and hook weekly.
14. Measure copywriters by profit per writer, tracked via their initials in the billing platform (UTMify/RedTrack), not by hit rate, which is impossible at this volume; setup: name/UTM every ad with the writer's initials and review monthly profit.

## 4. Tools & Stack - what each one is for here
| Tool | Role here |
| --- | --- |
| Facebook Ad Library | The 80-20 spy surface; keyword-based, catches DR ads via the video transcript |
| AdSpy | Secondary spy tool used only to highlight ads so none are missed |
| DeepTube (Hytalo) | The YouTube creative-spy tool he considers the best available |
| GPTonin (custom vertical GPT) | His own ad agent: writes scaling ads and "beats the control" from a pasted ad |
| ChatGPT / Claude | Base models; only useful when fed DR-specific psychology through a tuned agent |
| VEO3 (Google) | Generates demonstration visuals (bacteria, mechanism scenes) for VSLs and ads |
| UTMify / RedTrack | Tracks profit per copywriter by initials to measure real performance |
| Editor + media manager | Downstream of the copywriter: edit, launch, and validate at volume without review |

## 5. Actionable Checklist
- [ ] Classify each ad as predictability or scaling before writing a word.
- [ ] Spy only via blended market + niche + offer keywords, never generic searches.
- [ ] Open every Ad Library page; do not trust the first-screen duplicate count.
- [ ] Rewrite CTAs to carry a dollar value plus a reason why it is free.
- [ ] Layer benefits 2-4 onto every validated ad.
- [ ] Add two fascination bullets after the first CTA and a second CTA at the end.
- [ ] Insert an early express CTA in any ad longer than two minutes.
- [ ] Shoot for a 50-50 (or 80-20) predictability-to-scaling batch split.
- [ ] Reuse winning bodies with five new hooks and avatars for lateral scaling.
- [ ] Move ED creatives from clickbait toward authority-and-proof.
- [ ] Feed your own process and winners into a custom vertical GPT agent.
- [ ] Measure writers by monthly profit per initials, not by hit rate.