@shauneng 100% ... From a CS standpoint I would rather work for free and build my portfolio under my own conditions and wait to find a solid brand who get it, rather then being a slave $5k a month. Most experienced CS's aren't looking for 5k plus a little, might aswell do ad bounties
Meta doesn’t know who your customer is...but it does know what your customer 𝘥𝘰𝘦𝘴.
When you run broad targeting on Meta, you aren't really bidding on "personas" anymore.
You’re paying the delivery system to test human reactions to your creative in real time.
Here is how the algorithm matches your ads to potential customers 🧵👇
𝗦𝘁𝗮𝗴𝗲 𝟭: 𝗦𝗰𝗮𝗻𝗻𝗶𝗻𝗴 𝘁𝗵𝗲 𝗰𝗿𝗲𝗮𝘁𝗶𝘃𝗲.
Meta doesn’t inspect bios or interest stacks anymore. Instead, it scans your ad’s audio transcript, on-screen text, visuals, and pacing, then simply tracks and predicts in real time who might slow down their scroll to watch it.
𝗦𝘁𝗮𝗴𝗲 𝟮: 𝗣𝗮𝘁𝘁𝗲𝗿𝗻 𝗿𝗲𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻.
If your ad gets a specific reaction from a small cohort (e.g., stopping to read a chart or graph in your ad), the system will then test it on other users who also usually pause to read charts. It then groups them by behavioral 𝘩𝘢𝘣𝘪𝘵𝘴...even across totally unrelated topics. 🤯
𝗦𝘁𝗮𝗴𝗲 𝟯: 𝗔𝗰𝘁𝗶𝗼𝗻 > 𝗗𝗲𝗺𝗼𝗴𝗿𝗮𝗽𝗵𝗶𝗰𝘀.
The auction doesn’t really care if someone is 22 in Miami or 55 in Denver. If both stop to watch a clip of a super in-depth chart or graph, they get tagged with that behavior and share the 𝘦𝘹𝘢𝘤𝘵 𝘴𝘢𝘮𝘦 behavioral bucket. AKA: you get served based on 𝘩𝘰𝘸 𝘺𝘰𝘶 𝘣𝘳𝘰𝘸𝘴𝘦, not who you are on paper.
𝗦𝘁𝗮𝗴𝗲 𝟰: 𝗖𝘂𝗿𝗿𝗲��𝘁 𝗶𝗻𝘁𝗲𝗻𝘁 𝘄𝗶𝗻𝘀.
Something a user does today will outweigh their long-term history (which is bonkers, if you think about it!). Someone watching comedy for 3 months who suddenly spends 20 minutes researching back pain gets categorized for back pain, and their feed changes instantly. The feed delivers for 𝘢𝘤𝘵𝘪𝘷𝘦 intent, not an outdated profile tag.
(I think this is why manual interest targeting is fading so quickly. Trying to target the tag "interested in dogs" isn't going to hold up when the user just binged 10 funny cat videos in a row...)
You don’t select an audience inside Ads Manager; your creative does that for you. And if you start designing for 𝘢𝘤𝘵𝘪𝘰𝘯𝘴 instead of 𝘪𝘥𝘦𝘯𝘵𝘪𝘵𝘪𝘦𝘴, you can tap into entirely unsaturated audiences at cheaper CPMs much faster.
Example: say you're selling a premium ergonomic office chair.
The crowded pool goes after: remote tech workers.
(High overlap, brutal bidding wars, sky-high CPMs).
You're smarter than that, though, because you read this post 😅. Instead of targeting that interest tag, you build creative around 3 different consumer behaviors:
1️⃣ The "spreadsheet" watch behavior
Format: A first-person, giant spreadsheet comparison yapper of 14 chairs tested over 1,000 hrs, measuring foam sag.
2️⃣ The "How It's Made" watch behavior
Format: A pro-produced walkthrough of the entire manufacturing process.
3️⃣ The "bad habit" watch behavior
Format: A real user review: "If you sit on one leg by 2 PM every day, your seat pan depth is misaligned."
These 3 pieces of content all promote the exact same product, but the system indexes those formats independently and routes your ads into fresh buyer pools competitors ignore.
To scale creative under modern delivery models, test formats that prompt specific physical actions:
• Stress Tests (durability, torture tests, failure demos)
• Teardowns (internals, build quality, materials)
• Contrasts (cheap vs. premium, old vs. new)
• Audits (scoring rubrics, grading sheets, ratings)
• Corrections (mistakes, bloopers, "don't do this")
• Borrowing (shop setups, adjacent trade visuals)
• Symptoms (habit callouts, physical quirks)
• Raw Proof (spreadsheets, bench tests, data logs)
• Tool Stacks (software setups, equipment teardowns)
• Before/Afters (process timelapses, real-time transformations)
• Exposés (industry secrets, vendor pricing, markup myths)
• Protocols (step-by-step SOPs, daily routines)
• Debates (provocative questions, unpopular takes, trade lore)
• Stress Audits (calculators, cost-per-use, loss estimators)
📍 TL;DR: Platforms don’t run on manual demographic checkboxes.
They follow interaction sequences.
Your creative is your targeting system. Target the 𝘸𝘢𝘵𝘤𝘩𝘪𝘯𝘨 𝘣𝘦𝘩𝘢𝘷𝘪𝘰𝘳, not the persona.
(END OF YAP. 😅)
*Disclaimer: This is my interpretation of Meta's documentation around Andromeda, GEM, and the new targeting system we're dealing with in paid advertising, what have you guys seen??
@matthewvolkwyn and I brainstormed an easy, completely-overlooked method to source winning ideas for drama story ads…
That has already cranked out MULTIPLE top spenders for 8 & 9-figure brands:
Just look at drama media outlets/tabloids.
That industry’s been around for way longer than the ecom space…
And spent over a century perfecting the art of hooking and retaining attention with diabolical headlines and stories…
So just take those headlines and stories and turn them into drama ads = PRINT
Here’s a list of the biggest ones in the US and UK:
- [Take a Break](https://t.co/f9cyXcXYIr)
- [That’s Life!](https://t.co/ZNxCjJlyu2)
- [Chat](https://t.co/T4TXR68vJW)
- [The Sun](https://t.co/kvmw848nhK)
- [Daily Mirror](https://t.co/jrH29vIK2z)
- [Daily Mail](https://t.co/IXHUZ9xRJQ)
- [New York Post](https://t.co/oFtC1n7yNH)
- [National Enquirer](https://t.co/nt2sSGQYHX)
- [TMZ](https://t.co/jiF62O75sW)
- [E! News] (https://t.co/9YUpYsdySp)
Go try it and let us know your results.
Credit: Michael Kleynhans for sharing the pics below
this TRUE CRIME ad had no human editor
claude code searched a free film library, picked the shots and cut them on the narrator's words
42 s. 25 cuts. no timeline.
1958: a bank mails 60,000 people a credit card nobody asked for
case file -> suspects -> case closed
quick workflow:
> internet archive API = 15 film clips, $0
> claude reads the frames, picks each shot
> every cut lands on a spoken word
> stamps + exhibits = code, timed to the voice
> only 6 of 21 shots generated
most b-roll doesn't need generating
it already exists
RT + reply 'FILE' and i'll send you the full workflow
(must follow so i can dm)
@lorenzo_pravata Interesting, I would imagine the stage of awareness would be consistent; it might also be worth exploring the market sophistication level they are acknowledging or dealing with in the ad, dunno... I would definitely track valence and the intensity of that valence.