New rules for Meta advertising: creative > targeting

My name is Łukasz, and at 4822 I run advertising campaigns - mainly on Meta, meaning Facebook and Instagram, as well as on Google. On a daily basis, I set up, optimize, and analyze campaigns for our clients. I am writing this text because over the past five years, Meta’s advertising system has undergone a change that affects everyone who spends money on ads on Facebook and Instagram. The change happened gradually. It did not have one specific date or one official press release, so it was easy to miss. But its impact is very real: what worked in 2021 works differently today - or does not work at all.

Who is this text for? It is for people who make decisions about advertising budgets: business owners, marketing directors, and people responsible for ensuring that the money spent on ads delivers results. You do not need to know how to set up campaigns. But you do need to understand what has changed and why it matters for your business.

How should you read this text? The text is divided into sections. If you are not interested in the history of these changes, skip ahead to “What this means in practice.” If you already know the topic, go straight to “What this means for your budget.” I will not be offended if you read it selectively.

April 2021: Apple changes the rules

Apple released iOS 14.5 with App Tracking Transparency, or ATT. In practice, this meant that every app had to ask users for permission to track them. Most people said no.

For Meta, this was a fundamental problem. The Facebook Pixel - a small piece of code placed on advertisers’ websites — had spent years collecting data about what users did after clicking an ad. That data powered targeting: interests, lookalike audiences, and conversion optimization. After ATT, a significant part of that data stopped coming in.

Many of our clients suddenly stopped seeing most conversions in the dashboard. Campaigns that had been steadily generating purchases for months suddenly “stopped working” - because the system was no longer receiving the signal that someone had bought anything at all. The results did not disappear, but they became invisible.

That was the moment when we stopped treating dashboard data as the only source of truth. We started looking at MER — Marketing Efficiency Ratio, meaning the ratio of revenue to total ad spend - because this was a metric that ATT could not distort. We wrote about this more extensively at the end of 2021.

2022–2023: Meta starts rebuilding

Instead of patching the old system, Meta moved toward a new one — based on AI. Advantage+ appeared: a set of tools for automating campaigns. At first, it was an option. Over time, the system increasingly favored campaigns that gave it more freedom in selecting audiences.

For someone setting up campaigns, the change was gradual. Manual targeting still worked, but more and more often, Advantage+ delivered better results. The system learned faster than a human could manually test audience groups.

June 2025: the end of detailed targeting

Meta consolidated interest categories. Sport, music, food, car models — hundreds of detailed categories were merged into broader groups. Targeting, which for years had been an everyday working tool, officially became a “suggestion.” The system could ignore it if it decided it could find better audiences on its own.

From January 15, 2026, campaigns using old, deprecated interest categories that had not been updated simply stopped displaying.

We had several clients whose campaigns had to be updated in December ahead of that deadline. One of them — a company offering B2B training — had been targeting very specific professional categories for years, and those categories disappeared from the system. When we updated the structure in advance and switched the campaign to Advantage+, the results were comparable to what had worked before. Another client, a smaller clothing brand, learned about the changes only in mid-January — and for almost two weeks, their campaigns simply did not display. Zero reach, zero results. That was an expensive lack of attention.

Late 2024–2026: a new engine underneath

This is where a change happened that you cannot see in the ad dashboard, but that changes everything.

Meta rebuilt the system in three layers, each implemented separately.

Andromeda (introduced in late 2024) is the first layer - the engine that initially selects which ads can even be shown to a given user. It processes data about user behavior, such as what they watch, what they like, and how they scroll, and based on that narrows a pool of tens of millions of ads down to a few thousand candidates.

GEM - Generative Ads Model (introduced in mid-2025) is the central intelligence of the entire system. GEM learns from data from ads and organic content across Meta’s ecosystem - Facebook, Instagram, Messenger - and uses that knowledge to power real-time ranking models. It does not make decisions directly on every single impression. Instead, it helps the entire system understand more deeply what works, for whom, and when.

Adaptive Ranking Model (introduced in 2026) is the third layer - the final real-time decision about which ad wins the auction for a specific user at a specific moment. This is where the signals from Andromeda and the knowledge from GEM come together.

Three layers, each smarter than the last. Andromeda decides what may appear. GEM teaches the whole system what is worth showing. Adaptive Ranking Model decides what actually appears.

The effect of these five years of change is simple to describe: Meta’s system no longer waits for instructions from the advertiser. In the past, setting up a campaign looked like this: you selected a target group (age, interests, behaviors) uploaded the creative, and set the budget. The system showed the ad to the people you had chosen.

Now it works differently. You give the system a creative. The system analyzes it - what is shown in the graphic, what the text is about, what message it carries - and then selects the people it will show it to. The creative has become the main targeting signal. An ad aimed at restaurant owners will reach restaurant owners not because you selected the “food service” category, but because the system has recognized what your creative is about and matched it with people who respond to this type of message.

Three years ago, a large part of my work involved testing audience groups - different combinations of interests, lookalikes from various sources, and exclusions. Today, I spend much more time analyzing creatives and planning what will be fed into the system. Instead of asking, “Who are we showing this to?”, I increasingly ask, “What does this creative say about who it is for?” That is a fundamentally different way of working.

If creative is the signal that determines who sees an ad, the question becomes: what makes a creative “work” in the new system?

It is not about making the graphic “pretty.” It is about making it specific. Meta’s system analyzes the content of the creative - the message, the context, the emotions - and uses that to select the audience. A generic stock image with the text “Check out our offer” gives the system almost no information. A creative that speaks specifically to a specific person about a specific problem gives it a lot.

In practice, this means that creative variety has become more important than ever. Different creatives reach different people because the system matches the message to the audience. One campaign with one graphic is not enough. The system needs variants so it has something to choose from.

Three things.

First: creative is not an aesthetic cost, it is an operating cost. If until now you have treated ad creatives as “something you need to have in order to launch a campaign,” it is worth rethinking that. In Meta’s new system, the quality and variety of creatives directly affect who your ad reaches and how it performs.

Second: campaigns need time. the system learns from data. Every campaign rebuild, objective change, or reset erases what it has learned. Campaigns that perform well are often the ones that have had enough time and stability for the system to “get going.” Patience has become more profitable than constant tinkering.

Third: the data on your side matters too. Meta reports that companies with more accurately configured conversion events (Pixel, Conversions API) and better customer data give the system more material to learn from. The system works better when it knows what it is actually supposed to optimize for.

Honestly: Meta controls both the system and the narrative around it. The data on GEM’s effectiveness ( a 5% increase in conversions on Instagram and a 3% increase on Facebook) comes from Meta. Independent verification is limited. It is a black box we do not have access to.

We see the effects in the campaigns we run. But we do not see the mechanism. This is a limitation we work with — and one worth being aware of before treating these changes as a guarantee of better results.

Meta’s advertising system in 2026 works fundamentally differently than it did five years ago. The change was not sudden - it unfolded over several years, which made it easy to miss. But its effect is very real: creative determines who sees your ad. If you want to talk about how this affects your campaigns, get in touch. We will be happy to discuss it.

→ Book a call