## 1. Executive Summary

Marcelo Ornelas entered the market in 2020 at 17, dropped economics college and an internship for a media-buying course, and after a failed profit-sharing partnership in 2022 one offer, one channel, broken by election-season auction inflation got hired as a junior buyer at Doutor Natura in early 2023. Two years later he left as Head of Native, then moved to a US-focused operation. He has invested more than R$30 million in native traffic since 2022 and calls native the most constant traffic source there is: not the highest ROAS like YouTube, not the easiest to start like Meta, but the one that keeps running for a long time.

This episode is a full Taboola operating manual: self-service versus managed accounts, publisher block lists and white lists, Redtrack setup, organic and paid spying, creative geometry (image times headline), the six-by-six variation test, advertorial structure and its four optimization levers, and the bidding rules he actually uses.

## 2. Alpha / Sauce - what actually moves the needle, and exactly how to run it

**Never depend on a single acquisition channel.** His first partnership ran YouTube only; the 2022 political campaigns inflated the auction (R$1 clicks became R$3), the funnel stopped converting, and the partnership died. One channel is a company risk, not a strategy.

**Get a managed Taboola account by doing outreach, not by buying one.** Self-service caps your spend (around $10,000 before the account locks), cannot take long block lists, and lacks fast approval. Managed accounts have no spending limit, sub-accounts for scale, weekly benchmark reports, chat-based approval in about 20 minutes instead of 24-hour review, and publisher white lists. His route: find Taboola people on LinkedIn with your funnel and your results on other networks account managers earn roughly 1% of your spend, so a serious buyer is welcome.

**On self-service, block publishers from day one, in real time.** You get roughly 600 negative credits against a managed account's 1,200 1,500. Open the Websites tab, set the view to real time, and spend those credits blocking junk as it spends. Junk shows up as non-news sites with abnormally high CTR (around 2 3%+ where news sits far lower) click the site URL button and inspect it; he has seen one site family eat $500,000 of an account's budget.

**Ask your manager for the vertical white list.** Top native managers get a list of the best 500 1,000 publishers for their niche (nutra in his case) and restrict to it instead of blocking forever. That one conversation is the difference between guessing and running on known-good inventory.

**Image times headline is the whole creative game.** His formula: image squared times headline equals success. The image is squared because it stops the scroll it is the hook and tires first, so it needs the most replacement. The headline is what sells and survives longer. If the headline is zero, everything is zero.

**Micro-vary winners instead of replacing them.** He has run the same weight-loss headline since March, only micro-varying it, against completely different images. Micro-variations of a winning image (shirt color, hair, small changes) read as new creative to the algorithm and slow fatigue down.

**Build 36 ads, run 6.** Per angle: six images and six headlines uploaded in Taboola's variation mode, which crosses every headline with every image 36 combinations. He runs one pairing per image (six live, thirty paused), waits for two sales to judge headlines, then activates the winning headline against the paused images. Test at least three angles 108 creatives built but one angle per campaign, never mixed, or the CTR data gets tangled.

**Spy organically first, then pay.** VPN on phone and desktop, scroll the big portals (Yahoo, CBS News, ABC News, ESPN) until the same ad keeps winning auctions that is a working offer. Take that domain into Adplexity (~$200/month) to pull every ad that operator is running, and dig through four years of archive to revive angles nobody runs anymore. He also imports creative styles from Facebook into native.

**Copy validated advertorial structures; do not write from scratch.** The Golden Hippo / Beverly Hills MD pattern: common belief in the niche is true but not the root cause, doctor as authority, one simple home remedy, CTA about 500 600 words. Adapt the body to your niche the way you would adapt a Facebook ad body. His four optimization levers, in his order: layout, headline (push your top ad's headline into the advertorial), the CTA play image, and the lead. Layout is the easiest palliative; the lead is the lever he would keep if he could only keep one. Note the benchmark: advertorial-to-VSL passage sits around 25 30% in the US versus 30 40% in Brazil watch passage, but conversion rate is what pays.

**Bid with data, not with hope.** Two strategies: cost control (you set the CPC and enter the auction) and Max Conversions (budget only, Meta-style). Managed managers send weekly benchmarks average mobile nutra click around $0.25. Your CTR moves your real CPC: strong clickbait creative lets you bid lower, weak creative forces you up to compete.

## 3. Methods & Setup - Step-by-Step

1. **Get the account right before anything else.** Outreach Taboola staff on LinkedIn with your funnel, your niche and your results on other networks; if you must start self-service, accept the limits knowingly; setup: LinkedIn message script and funnel URL ready.

2. **Wire tracking first, with Redtrack.** He does not recommend Voluum for beginners (complex); Taboola itself sends a Redtrack setup guide because that is what everyone uses; setup: Redtrack account, offer + funnel events, postbacks to Taboola, and one person who owns the tracker if you can afford it.

3. **Build the publisher lists.** Managed: request the 500 1,000 publisher white list for your vertical plus the block list. Self-service: pre-build your own negative document if you have one, otherwise block live from day one; setup: Websites tab open in real-time view on launch day, 600 negative credits spent on junk, URL inspection button used before every block.

4. **Spy organically before creating anything.** VPN on phone and desktop, scroll Yahoo, CBS News, ABC News and ESPN until a domain keeps showing in the top of the feed winning the auction means high CPM and a working creative; setup: two devices with VPN, a running note of repeating domains.

5. **Pull the winner's whole book in Adplexity.** Paste each repeating domain to see every ad that operator runs, filter by affiliate network or checkout, and check the archive going back years for angles that worked and are now unsaturated; setup: Adplexity subscription (~$200/month), saved searches per niche.

6. **Choose the angles to test at least three.** For health and nutra he uses food (common foods tied to the problem), expert/doctor, and strange-trick; memory adds the puzzle angle (elderly person mid-daily-activity, AI-generated); setup: an angle list per niche with image concept and headline concept written down for each.

7. **Write headlines against the formula, under 70 characters.** Doctor + problem + solution, curiosity ("has been linked to this daily habit"), or a specific story ("arrested for losing weight too fast"); keep the mechanism or key word at the front because mobile truncates the tail; setup: headline bank of validated winners, plus a GPT prompt fed with your history, competitors' ads and past funnels that outputs headline, avatar, benefit and mechanism.

8. **Generate the six-by-six per angle.** Six images of that angle, six headlines of that angle, upload in variation mode 36 combinations created; setup: one campaign per angle, variation mode creative upload, six pairings activated and thirty left paused.

9. **Launch with proven plus new.** Every new campaign opens with his two all-weather headlines (the ones that sell with almost any image) plus four fresh ones generated in GPT; setup: two bank headlines + four new per campaign, budget at benchmark CPC (~$0.25 mobile nutra) under cost control.

10. **Optimize on two sales, headline first.** Wait for roughly two sales before judging; pause headlines that did not sell, activate the paused images against the headlines that did; images get rotated continuously because they tire first; setup: a written rule two sales = headline read, no sales after a spend threshold = angle killed.

11. **Model the advertorial, then work its four levers.** Adapt a validated structure (common belief not the root cause doctor authority simple home remedy CTA, 500 600 words), keep the micro-lead congruent with the ad and the VSL mechanism; then optimize in order: layout, headline, CTA play image, lead; setup: advertorial built from a named swipe structure, passage and conversion tracked in Redtrack, layout variant split against control.

12. **Match the layout to the publisher.** He has run advertorials styled to the visual language of the site the traffic comes from color, header, font so the click from a news page lands on something that looks like it; setup: top-five publishers identified, a developer task to serve a matching header, split test against generic layout.

13. **Install the weekly intelligence loop.** Managed accounts get weekly benchmark reports; he keeps a LinkedIn feed of native operators posting insights and takes English lessons to sit in communities like Native Hub; setup: Monday check of benchmarks, one community or feed reviewed weekly.

## 4. Tools & Stack - what each one is for here

| Tool | What it is for here |
|---|---|
| Taboola (managed account) | The traffic source: no spend cap, sub-accounts, chat approval in ~20 min, weekly benchmarks |
| Redtrack | Conversion tracking, funnel-stage metrics, postbacks feeding Taboola the default he recommends |
| Voluum | Alternative tracker he used earlier; powerful but too complex for a beginner |
| Adplexity (~$200/mo) | Paid spy: paste a winning domain, see all of that operator's ads, dig years of archive |
| VPN (phone + desktop) | Organic spy on Yahoo, CBS News, ABC News, ESPN to find which domains keep winning |
| Swipe files (his released native swipe + Golden Hippo-style advertorials) | Winning ads and validated advertorial structures to model, not write from scratch |
| GPT headline prompt ("headline killer") | Generates headlines from avatar, benefit, problem and mechanism over your historical bank |
| LinkedIn | Outreach to Taboola reps for managed accounts, and a feed of native operators' insights |

## 5. Actionable Checklist

- [ ] Outreach Taboola reps on LinkedIn with funnel, niche and results instead of buying an account
- [ ] Set up Redtrack with postbacks before spending a dollar
- [ ] Ask your manager for the vertical white list (500 1,000 top publishers) and the block list
- [ ] On self-service, block junk publishers live from day one using the real-time Websites tab
- [ ] VPN both devices and scroll the big portals until the same domains repeat
- [ ] Paste repeating domains into Adplexity and mine the archive for unsaturated angles
- [ ] Define at least three angles per test with image concept + headline concept each
- [ ] Write headlines under 70 characters, mechanism first; keep a validated headline bank
- [ ] Upload 6 images 6 headlines in variation mode, run only 6, pause 30
- [ ] Judge headlines after ~2 sales; pause losers, activate winner against paused images
- [ ] Model advertorials from validated structures, then test layout headline CTA lead
- [ ] Check weekly benchmarks, hold CPC near the niche average, adjust bid by CTR strength