Meta Ads in 2026: Advantage+, AI Delivery and Creative Testing

Meta's AI now decides who sees your ads. Here is how Advantage+, Andromeda and GEM work in 2026, and a step-by-step creative testing framework to run every week.

SSquarebase
Published 1 Oct 2026 · 9 min read

Quick answer: In 2026, Meta ads are delivered by AI systems (Andromeda for ad retrieval and GEM for ranking) that decide who sees which ad. Advantage+ campaigns automate budget, audience and placements, so your main lever is creative. Run a steady pipeline of genuinely different concepts, test them in a structured way, and feed winners into broad Advantage+ campaigns.

Meta ads in 2026 work very differently from the interest-stacking and lookalike playbooks most brands learnt between 2018 and 2022. Meta Advantage+ now automates most audience, placement and budget decisions, and the delivery system picks the right person for each ad based on the ad itself.

This guide explains how Advantage+ and Meta’s AI-driven ad delivery work now, using Meta’s own engineering and developer documentation, and gives you a practical creative testing framework you can run every week.

Key takeaways

Meta’s Andromeda retrieval engine (announced December 2024) and GEM ranking model (detailed November 2025) now do most of the audience-finding work.
A campaign counts as Advantage+ when budget, audience and placements are all automated, under the unified campaign setup Meta began rolling out in May 2025.
Meta itself says creative diversification is now the best lever to find relevant audiences.
Diversity means different concepts, personas and formats, not ten colour versions of one ad.
Clean conversion data (Pixel plus Conversions API) matters as much as creative, because the AI optimises towards whatever signal you send.

What changed in Meta ad delivery between 2024 and 2026?

Three changes explain almost everything advertisers are seeing in their accounts: a new retrieval engine, a new ranking model and a simplified campaign structure.

Andromeda: the retrieval engine

Before an ad can be ranked, Meta has to pull a shortlist of candidate ads from millions of options. In December 2024, Meta’s engineering team published details of Meta Andromeda, its next-generation ads retrieval system.

Meta reported a 6% recall improvement in retrieval and an 8% ads quality improvement on selected segments. The same post says the system was built partly because generative AI tools were causing exponential growth in the number of ad candidates, with more than a million advertisers creating over 15 million ads in a month using Meta’s generative AI tools.

GEM: the ranking model

Once candidates are retrieved, ranking models decide which ad to show. In November 2025, Meta described its Generative Ads Recommendation Model (GEM) as its most advanced ads foundation model, built on an LLM-inspired approach.

Meta said GEM drove a 5% increase in ad conversions on Instagram and a 3% increase on Facebook Feed in Q2 2025. The practical point: the model reads signals from the ad and the user together, so what your ad shows and says directly shapes who it reaches.

The unified Advantage+ campaign setup

From 29 May 2025, Meta began rolling out a single campaign flow for Sales, App and Leads objectives. Advantage+ shopping campaigns became Advantage+ sales campaigns. According to Meta’s Marketing API documentation, a campaign gets Advantage+ status when three automation levers are on: campaign-level budget, Advantage+ audience (or minimal targeting) and no placement restrictions.

How do Advantage+ sales campaigns work now?

You still choose the objective, the conversion event, the budget, the country and your ads. Meta’s system handles the rest within the limits you set.

Lever	Manual setup (older approach)	Advantage+ setup (2026 default)
Budget	Fixed budget per ad set	Campaign-level budget that Meta moves to the best ad sets and ads
Audience	Stacked interests, lookalikes, narrow age bands	Advantage+ audience, with your audiences used as suggestions
Placements	Hand-picked placements	All eligible placements across Meta apps and the wider network
Creative	Few ads, often one format	Many diverse ads; the system matches each ad to likely buyers
Your main job	Targeting and bid tweaks	Creative supply, conversion data quality and budget decisions

You can still set hard limits such as location, minimum age, excluded audiences and language. If you switch off the audience or placement automation, the campaign simply stops being labelled Advantage+. That is a valid choice for some regulated categories or very small local businesses.

Why is creative now the main targeting lever?

When the platform chooses the audience, the ad itself becomes the strongest signal you control. A video about back pain for desk workers will find different people from a video about gifting for new parents, even inside the same broad campaign.

Meta has said this plainly. In an April 2025 post quoted by Jon Loomer’s analysis of Andromeda, Meta wrote that “the focus has shifted from niche targeting to creative diversification as the best lever to find the most relevant audiences.” Meta recommends varying concepts and angles, formats and personas.

The scale of AI creative use is also growing. On its Q1 2026 earnings call (29 April 2026), Meta said the number of advertisers using its AI creative tools had doubled from 4 million to 8 million in about four months, as reported by PPC Land. More ads in the system means more competition for attention, which makes genuinely distinct creative more valuable, not less.

What “diverse creative” actually means

Different concepts: problem and solution, founder story, customer testimonial, product demo, comparison, offer-led.
Different personas: the same product sold to a first-time buyer, a repeat buyer, a gifter or a business buyer.
Different formats: short vertical video, longer explainer video, static image, carousel, user-generated style content.
Different hooks: the first two to three seconds of a video or the headline of a static ad should change between concepts.

Changing only the background colour, the button text or the font is iteration, not diversification. Those variations tend to look identical to the delivery system and end up competing for the same people.

What is a practical creative testing framework for Meta ads in 2026?

The framework below is the one we use as a working process. The thresholds are starting points that you should adjust to your own average order value, conversion rate and budget, not fixed rules from Meta.

Step 1: Build a concept matrix

Map personas against angles before anyone opens a design tool. Each filled cell is a concept that needs at least one video and one static execution.

Persona (example: premium eyewear brand)	Angle A: Problem	Angle B: Proof	Angle C: Offer
Office professional	Screen fatigue after long workdays	Customer review on lens comfort	Free eye test with first frame
Style-led buyer	Generic frames that look the same	Creator try-on video	New collection launch
Gifter	Hard to buy for someone who has everything	Unboxing and packaging close-up	Gift card or exchange promise

Step 2: Separate testing from scaling

Run one testing campaign with a fixed share of total Meta spend, commonly 10% to 20%, and keep your main Advantage+ sales campaign for scaling. New concepts go into the testing campaign first so that a weak batch does not disrupt a stable scaling campaign.

Step 3: Launch in batches

Launch three to six new concepts at a time, each as its own ad or small group of ads. Keep the offer, landing page and conversion event the same within a batch so that the concept is the main variable.

Step 4: Judge on the right metrics, at the right time

Early signals (first 2 to 3 days): hook rate (3-second video views divided by impressions), click-through rate and cost per landing page view.
Decision signals (after enough conversions): cost per purchase or lead, and ROAS against your break-even ROAS.
Spend share: if Meta refuses to spend on an ad even at a reasonable budget, the system has already judged it weaker than its siblings.

A useful rule of thumb is to let an ad spend at least two to three times your target cost per purchase before calling it a loser, unless its early signals are clearly poor.

Step 5: Graduate winners and iterate

Move winning ads into the scaling campaign using the existing post ID where possible, so you keep social proof such as comments and reactions. Then brief the next batch: two or three iterations of the winning concept (new hook, new opening shot, new format) plus two or three completely new concepts.

Step 6: Refresh on a schedule

Set a weekly or fortnightly creative cadence. Watch for rising frequency combined with falling click-through rate and rising cost per result, which usually signals fatigue for that concept.

What should you still control manually?

Automation does not remove the need for judgement. These are the inputs that most affect results in Advantage+ campaigns:

Conversion signal quality: send purchase and lead events through both the Pixel and Meta’s Conversions API, with deduplication set up correctly.
The optimisation event: optimise for the event that matches revenue, such as a purchase or a qualified lead, not a cheap proxy like add to cart.
Hard limits: location, minimum age, language and genuine exclusions such as existing employees.
Existing customer measurement: track how much spend goes to people who already bought from you, so that new customer acquisition cost is not hidden.
Landing page match: the page should continue the promise made in the ad. A great ad sent to a generic home page wastes the delivery system’s work.
Budget and bid strategy: highest volume, cost per result goal or ROAS goal, chosen to fit your margin.

What are the most common Meta ads mistakes in 2026?

Splitting budget across many small ad sets with near-identical audiences, which starves each one of learning data.
Uploading many lookalike variations of one ad and calling it creative testing.
Judging new ads within 24 hours on ROAS alone.
Optimising for a top-of-funnel event when the business needs purchases or qualified leads.
Reading Meta’s attributed ROAS without checking it against store revenue and blended marketing efficiency.

The last point matters most for growing brands. Platform-reported ROAS is useful for comparing ads inside Meta, but business decisions should be checked against total revenue and total marketing spend.

How Squarebase can help

Squarebase plans and runs Meta campaigns with a weekly creative testing process, clean conversion tracking and reporting tied to your store or CRM data. If you want a second opinion on your account structure or creative pipeline, see our performance marketing services or talk to our team.

Frequently asked questions

What is Meta Advantage+ in 2026?

Meta Advantage+ is Meta’s set of AI automation features for ads. A campaign is labelled Advantage+ when its budget is set at campaign level, its audience uses Advantage+ audience or minimal targeting, and placements are not restricted. Meta then decides who sees each ad and where.

What is Meta Andromeda?

Andromeda is Meta’s ads retrieval engine, described by Meta’s engineering team in December 2024. It selects a shortlist of candidate ads for each person before ranking happens. Meta reported a 6% recall improvement and an 8% ads quality improvement on selected segments.

Why is creative called the new targeting on Meta?

When Meta’s AI chooses the audience, the content of the ad becomes the strongest signal you control. Different concepts, personas and formats reach different people inside the same broad campaign. Meta has said creative diversification is now the best lever to find relevant audiences.

How many ads should I test at once on Meta?

A practical starting point is three to six genuinely different concepts per batch, inside a testing campaign that takes around 10% to 20% of Meta spend. Keep the offer and landing page constant within a batch so the concept is the main variable you are measuring.

Do I still need the Conversions API with Advantage+?

Yes. Advantage+ optimises towards the conversion events you send, so missing or duplicated events lead to poor decisions. Meta’s Conversions API sends events from your server, website platform, app or CRM, and works alongside the Pixel with deduplication.

Sources

Written by Squarebase, Squarebase

Squarebase is a 360° marketing agency, founded in 2017, working on performance marketing, Shopify, websites and SEO, AEO and GEO for brands in India, the USA, the UK, Germany and Australia.