How Meta Ads Actually Work in 2026 (And Why Everything You Learned Before Is Outdated)

If you started running Meta ads a few years ago, you probably learned a version of this: build your audiences carefully, set up separate campaigns for top, middle, and bottom of funnel, test new interest groups to find the ones that spike your ROAS, and tinker with your targeting until something sticks.

That approach is not just outdated. In some ways it is actively working against you. The platform has changed more in the last two to three years than it did in the previous decade, and the advertisers who are getting consistent results are the ones who understand what actually drives performance now.

This post covers the fundamental shift in how Meta's algorithm works, what that means for how you set up and manage your account, and the creative strategy that sits behind every high-performing ad account in 2026.

The Old Way vs. The New Way

From roughly 2018 through to about 2022, you were in control of the targeting. You told Meta who to show your ads to through:

• Interest groups

• Lookalike audiences

• Custom audience segments

The game was finding the right pockets of people and serving them the right message, and success often came from discovering a new interest combination that nobody else had found yet. It worked, but it was manual, time-consuming, and not sustainable. When a targeting hack stopped working, you had to find another one.

That era is over.

Meta's algorithm has become sophisticated enough that your manually defined audiences are largely being ignored. Even when you set them up, Meta is increasingly doing its own thing underneath. The platform is pushing hard toward a model where you supply the creative inputs and the algorithm handles everything else.

This is not a bad thing, once you understand it. It means your energy is better spent in completely different places.

Meet the Two Systems Changing Everything: Andromeda and GEM

The reason Meta's algorithm has become so much more capable comes down to two systems working together in the background. You did not get a notification when they switched on. There was no setting to enable. They are simply the new engine running underneath every ad on the platform.

Andromeda

Andromeda is the system that decides which ads are even eligible to be shown to a given person at a given moment.

In the past, Meta could evaluate thousands of candidate ads for a user. Now, thanks to Andromeda, it processes millions, and it does so by reading real-time behaviour signals rather than static demographic categories.

Think of it as an extremely sophisticated matchmaker that is watching what people actually do, second by second, rather than making guesses based on their age or interests.

GEM

GEM is the underlying model that teaches Andromeda. It has been trained across billions of data points and it learns not just what gets clicked, but what ultimately converts.

If Andromeda is the matchmaker who brings the right person to your ad, GEM is the teacher who trained that matchmaker on everything that has worked before and everything that has not.

What This Means Practically

The algorithm is now matching ads to people based on actual intent signals, updated in real time.

If someone visits a candle website, adds something to their cart, and then opens Instagram, they are likely to start seeing candle ads within minutes. Their behaviour told the system what they were interested in, and the system responded.

You cannot replicate that level of real-time intelligence with manual audience targeting. The algorithm simply has access to more data than you do.

What the Algorithm Actually Rewards Now

Understanding that Meta runs on these AI systems changes what you should be optimising for.

The algorithm rewards three things above all else:

1. Clean, quality signals

2. Positive user experience

3. Creative diversity

1. Clean, Quality Signals

This means giving Meta an accurate picture of who your customer actually is, not a vague demographic statement, but a specific person with specific behaviours and desires.

This also means:

• Your pixel is set up correctly

• Your conversions API is functioning properly

If your tracking is broken or incomplete, you are feeding the algorithm bad information, and it will charge you a premium for the privilege.

Higher CPMs for poor user experience is not a coincidence. It is the algorithm reflecting back the quality of what you are giving it.

2. Positive User Experience

GEM tracks what happens after someone clicks your ad.

A low conversion rate on your website is not just a website problem. It signals to Meta that its users are having a poor experience with your brand, and it will deprioritise you accordingly.

Reducing refund rates, improving page speed, and giving people a trustworthy checkout process are all, in a very real sense, ad account decisions.

3. Creative Diversity

This is the big shift.

Where targeting used to be the primary lever, creative is now the targeting.

When you show Meta a range of well-constructed ads speaking to different angles, different emotions, and different customer personas, you are giving the algorithm the raw material to find the right match for every type of person on the platform.

The bottleneck is no longer who sees your ads. It is which ads are relevant enough to earn attention.

Why You Should Stop Using the Words "Lookalike" and "Interest Targeting"

This might sound blunt, but if these terms are still central to how you think about Meta ads, it is worth letting them go.

Meta has been phasing them out for years. They still exist as options in the interface, but the algorithm increasingly ignores them in favour of its own signals.

Spending mental energy on building the perfect lookalike audience is effort that would be far better placed in creative strategy.

The same goes for top, middle, and bottom of funnel campaign structures.

The idea that you need separate campaigns for awareness, consideration, and conversion made sense when you controlled the targeting. It makes much less sense when the algorithm is doing the matching dynamically based on where each individual user actually is in their purchase journey at that moment.

Simplicity scales. Complexity fails.

One campaign, broad targeting, and let the algorithm do what it is genuinely very good at doing.

The 3-3-2-2 Creative Strategy: Feeding the Machine

If creative is the new targeting, then you need a system for producing creative consistently and strategically. The 3-3-2-2 method is a framework for doing exactly that.

The idea starts with angles, not ads.

Before you think about what your creative looks like, you choose three different advertising angles. An angle is the core reason someone might buy your product, framed around a specific emotion or motivation.

The three main angle types are:

• Pain

• Desire

• Identity or demographic

Step 1: Three Angles

Pain
What problem does your product solve and how much is that problem costing someone?

Desire
What does life look like after the problem is gone?

Identity
Who is the person you are speaking to, and does your ad reflect their world back at them?

Step 2: Three Creative Executions Per Angle

For each of those three angles, you create three different creative executions.

These might be:

• A video

• A static image

• A graphic with text overlay

The format matters less than the fact that you have multiple variations of the same angle.

Step 3: Two Primary Text Variations Per Creative

For each piece of creative, you write two different pieces of primary text:

• One leans into the negative emotion

• One leans into the positive aspiration

Meta's GEM system is reading the emotional signal of your copy and serving it to the kind of person most likely to respond to that register.

Step 4: Two Headlines Per Primary Text

Finally, for each piece of primary text, you have two headline variations.

These might be:

• Urgency-based

• Question-based

• Identity-based

When multiplied out:

3 angles × 3 creatives × 2 primary texts × 2 headlines

You end up with a large number of combinations from a structured starting point.

This is what the algorithm needs. It is not dozens of random ads thrown at the wall. It is a deliberate variety of well-researched inputs, organised around specific human motivations.

How to Set Up Your Account to Match the Algorithm

The account structure that works in 2026 is considerably simpler than what many people are used to.

Core Testing Campaign

For most Shopify brands:

• One core testing campaign

• Ad sets organised by angle, not audience

Pain angle ads in one ad set.
Desire angle ads in another.
Identity angle ads in another.

This way, when you review performance after seven days, you can clearly see which angle is resonating.

Inside each ad set:

• Run three to six ads

Under $500 per day in spend:

• One to three ad sets at a time

The goal is consolidation.

Splitting a $100 daily budget across five campaigns means none accumulate enough data to learn.

Review Windows

Seven days is the minimum review window before making decisions.

For smaller budgets:

• 14 to 30 days may be more appropriate

The algorithm needs time and data.

Checking daily and reacting to fluctuations is one of the most common ways people prevent their accounts from ever stabilising.

Winners Campaign

Once you identify winning ads from your core testing campaign:

• Graduate them to a Winners Campaign

• One ad set

• Your ten best-performing ads

As new winners emerge:

• Swap out the bottom performers

This gives proven creative more budget and room to scale.

What to Look At When You Review Your Account

Every time you open Ads Manager, focus on four metrics:

• CPM

• Click-through rate

• Conversion rate

• Average order value

Everything else, including ROAS and cost per purchase, is a result of how these four metrics are performing together.

You do not optimise ROAS directly.
You optimise the inputs that produce it.

CPM

Tells you how much Meta trusts your brand and how competitive your space is.

A rising CPM without improvement elsewhere signals investigation into:

• User experience

• Pixel quality

Click-Through Rate

Tells you whether your ad is stopping the scroll.

Low CTR usually means:

• The angle is wrong

• The hook is weak

• The creative is not distinctive

Conversion Rate

Tells you what happens after the click.

Strong CTR + weak conversion rate = website problem.

Average Order Value

Tells you whether the economics work.

Sometimes the lever is not more customers.
It is higher spend per customer through:

• Bundles

• Upsells

• Higher-priced offers

Every performance problem traces back to one of these four numbers.

Creative Is Not Optional: The Three Vs

The Meta advertising environment is more competitive than ever.

AI tools are lowering the barrier to producing creative.
Volume of ads is increasing every year.

Staying visible requires three Vs:

• Volume

• Variety

• Velocity

Volume

Enough creative for the algorithm to test meaningfully.

Variety

Different angles and emotional registers, not slight variations of the same idea.

Velocity

Producing and iterating quickly enough to stay ahead of fatigue.

This does not mean producing low-quality work at speed.

Flooding your account with generic AI content is not a shortcut. It is stagnation.

The goal is a systematic process for generating fresh, research-backed creative that gives the algorithm useful signals.

The advertisers winning on Meta right now are not the ones who found a clever hack or a perfect interest group.

They are the ones who:

• Built a repeatable creative system

• Read their four core metrics honestly

• Made clear-headed decisions about what to scale and what to replace

The algorithm is more capable than it has ever been.

Your job is to give it something worth working with.