What you'll be able to do
- Explain how Meta's ad system decides who sees your ad in 2026 - to a client, in one minute, without notes
- Stop wasting hours on tactics that quietly stopped working
- Name the three levers that actually move performance now, and spend your time there
The short version: In late 2025, Meta finished rolling out a new ad delivery engine called Andromeda. It reads your ad creative to decide who should see it - your creative IS your targeting now. Interest stacks, lookalike layers, and elaborately partitioned campaign structures matter far less than they did in 2023, and as of February 2026 your detailed targeting selections are officially treated as suggestions. The three levers that matter: creative diversity, simplified structure, clean conversion signal. Most of what the ads-guru internet teaches is a beautifully preserved museum exhibit. Welcome to the current decade.
The room changed while everyone was looking at the decks
Every scene has that person still handing out physical flyers for a party that moved venues two years ago. The Meta ads education industry is substantially that person. This module is the new address.
If this is your first rodeo: good news - you're skipping the museum entirely. You never have to learn interest stacking, so you never have to unlearn it. Veterans reading this module are doing the harder job: letting go of skills that used to be the whole craft. Be nice to them in the group chat.
What happened
Meta announced Andromeda on December 2, 2024 through its engineering blog, describing a next-generation ads retrieval engine built to power Advantage+ automation (Meta Engineering, Dec 2024). It rolled out in waves through 2025 - which is why performance that year felt episodic, unstable-then-fine, over and over (Uproas, Feb 2026) - reached broad availability across most objectives and placements by October 2025, and became simply how the platform behaves by Q1 2026 (Weber Media, June 2026 · Jetfuel, July 2026).
Here's the piece that makes everything click: Andromeda is not the auction. It's the step BEFORE the auction - the retrieval system deciding which ads even get considered for a given person. Door policy before the dance floor. If your ad is never retrieved, your bid, your budget, your optimization settings - all irrelevant. You can't win a room you never entered (Uproas, Feb 2026).
›Go deeper: why Meta rebuilt the engine at all
Two pressures converged. First, volume: Advantage+ Creative and AI generation tools exploded ad variation counts - by global rollout, Meta was processing over 15 million new ad creatives per month, and the legacy retrieval system had no path to that scale (Denote, 2026). Second, signal: post-privacy-era interest and demographic data got weaker every year. So Meta rebuilt delivery around the one signal getting RICHER instead of poorer - the creative itself, which deep neural networks can now evaluate across thousands of variants in milliseconds. The detail most coverage misses: Andromeda wasn't built just to improve targeting; it was built because the old system was drowning (Denote, 2026).
What this means for you, specifically
Your creative does the targeting. Andromeda reads the ad - visuals, copy, format, mood - and predicts who it fits from behavioral signals (Jetfuel, July 2026). An ad that looks and sounds like a techno warehouse at 2am goes and finds techno warehouse people. A generic "EVENT · SATURDAY · TICKETS" graphic finds approximately no one, because it describes no one.
Detailed targeting became advisory - officially. Meta's February 2026 API update formalized what had been quietly rolling out: your interest and demographic selections are treated as suggestions, not hard constraints. The system can and will deliver beyond them when it predicts better performance (Denote, Feb 2026). The hours you'd have spent building audience architectures now belong to creative production. Not a loophole - the design.
Old structures actively hurt. Fragmented accounts - many campaigns, many ad sets, tightly sliced audiences - chop your conversion signal into pieces too small to learn from. Consolidated, broadly-delivering campaigns are the doctrine, and at small budgets it matters most (Weber Media, June 2026). Full structural build in Module 4.
Volume expectations moved. Meta removed its old ~6-ads-per-ad-set guidance; strong accounts now test 10 to 50 ads per ad set - with the emphasis on genuinely different concepts, not fifty crops of one flyer (Dataslayer, Apr 2026). What "genuinely different" means is Module 6's whole job.
Your job description changed. The old craft: control the campaign. The new craft: feed the system good inputs - creative quality and variety, clean signal, sensible structure. The advertisers struggling most in 2026 are the ones still trying to micromanage; the ones winning accepted the shift and redirected the energy (Dataslayer, Apr 2026). Less puppeteer, more booking agent: you don't control the crowd, you control what's on the bill.
›Go deeper: "episodic" 2025, explained for the veterans
If your accounts felt haunted through 2025 - stretches of chaos, then normal, then chaos - that wasn't you. Meta integrated Andromeda deliberately and incrementally across objectives and placements, tuning in the background rather than flipping one switch, so its effects arrived in uneven waves (Uproas, Feb 2026). Useful client-conversation ammunition when someone asks why last year's playbook "suddenly" stopped working: it didn't suddenly do anything. The floor moved in stages, and now it's done moving. This is the floor.
From the field
A venue client's account, inherited: 9 campaigns, 30+ ad sets, interest-stacked to the rafters, $75/day spread across all of it like too little butter on too much toast. Nothing ever exited learning. And here's the thing - it wasn't built by a fool. It was built by someone faithfully following genuinely excellent 2019 advice. Consolidated to 2 campaigns with a real creative library at the same spend: the account finally had concentrated signal, learning stabilized, costs dropped. Nothing about the venue changed. The calendar did, and the account hadn't been told. (This is also why every page of this school wears a verification date. Undated advice isn't advice; it's archaeology.)
Common mistakes
- Following a 2022 tutorial that spends 40 loving minutes on audience building - archaeology
- Testing 12 versions of one flyer and logging it as creative diversity
- Blaming targeting when the honest diagnosis is that the ad is boring
- Restructuring the account weekly, resetting learning every time, then describing the results as "inconsistent"
- Explaining the platform to clients with 2023 vocabulary - you now have the one-minute 2026 version; use it
Take-home
"What Changed: 2023 vs 2026" comparison chart - one page, old world vs new world, lever by lever. Built to be forwarded to the co-promoter who still wants to talk about lookalike percentages.
Sources & dates
Meta Engineering Blog (Dec 2024) · Weber Media (June 2026) · Jetfuel (July 2026) · Denote (Feb 2026) · Uproas (Feb 2026) · Dataslayer (Apr 2026)
Learn more
Optional extra credit. Nothing below is required for the quiz, the certificate, or the job - the full lesson is above. This is for the sickos who want more.
- Meta's engineering blog post announcing Andromeda (Dec 2024) - the primary source, surprisingly readable
- Denote's Andromeda analysis (updated 2026) - the best outside deep-dive, including the February API change
- Uproas on the rollout timeline and refresh strategy (Feb 2026)