AI-Generated Ad Creative in Meta Ads Manager: Does It Actually Work?
META ADS · 9 MIN READ

AI-Generated Ad Creative in Meta Ads Manager: Does It Actually Work?

I ran the generative tools against my own creative for six months. They win at variation and volume, and they have not once produced my best-performing ad.

By Tariq Sallam·September 2026

Meta's creative tools now generate backgrounds, expand images, write copy variants, produce video from stills and automatically restyle assets per placement. All of it is inside Ads Manager and most of it is free.

Having used them across several client accounts for six months, my summary is that they are genuinely useful for the boring half of creative production and consistently unremarkable at the interesting half. The distinction is worth understanding before you hand over your creative pipeline.

Quick Info

Genuinely useful
Resizing, reformatting, background expansion, copy variants
Occasionally useful
Text overlay variants, music and pacing suggestions
Not useful
Generating the core concept or the hero asset
Effect on performance
Neutral to positive on volume, negative if used alone
Rule
Real footage, AI-assisted variation
Watch for
Generated visuals that look generated

What the tools actually do

Image expansion: extends your photograph to fill vertical and square placements. The most useful feature in the set.
Background generation: replaces or creates a setting behind a product.
Text variation: generates headline and primary text alternatives from your original.
Image to video: adds motion to a still, usually a slow pan or zoom.
Automatic adjustments: per-placement cropping, brightness and template overlays.
Catalogue enhancements for e-commerce: automated overlays, frames and price flags.

Note what is on this list: transformation of assets you supply. The tools are better at adapting your material than at inventing material, which matches where the technology actually is.

Where they earn their place

Reformatting

The single best use. You shot one asset; the platform needs it in four aspect ratios. Image expansion does this well and it eliminates the most common creative failure, which is running a square asset in a full-screen placement.

Copy variation at volume

Generating twenty headline variants to choose three from is faster than writing twenty. The output is fluent and generic, and picking the three least generic is a reasonable workflow.

Product photography backgrounds

For e-commerce, placing a clean product shot in a plausible setting is quick and often good enough. This is a real cost saving on catalogue work.

Adding motion to statics

Mild, and it does measurably lift performance in Reels placements where pure statics underperform. Cheap to apply.

Where they fail

The core concept. The tools vary what you give them; they do not decide what argument to make. That decision is the creative work.
Anything with a real person. Generated or heavily altered faces read as artificial and I have seen consistently worse performance.
Your actual product in use. If the ad needs to show the real thing working, generation is not an option.
Copy with a point of view. Generated text averages towards the consensus phrasing, which is precisely what makes an ad forgettable.
Brand consistency at any depth. It matches colours; it does not understand what you sound like.

The tools are extremely good at making forty versions of an average ad and no help at all in making one good one.

What my testing showed

Across several accounts over six months, comparing generated variants against my own creative on the same audiences and budgets.

AI-expanded and reformatted versions of my real assets performed the same as manually adapted versions, at a fraction of the effort. Clear win.
Fully generated visuals performed worse than real footage in every account, with the gap widest in service businesses.
Generated copy variants performed within normal variance of my own for short headlines, and worse for anything requiring a specific claim.
Adding motion to statics improved Reels performance modestly and consistently.
Accounts using only generated creative plateaued faster, because every variant was a variation of the same average idea.

The pattern is consistent: use it on the mechanical steps, keep the concept and the raw material human.

How I use them in practice

01Decide the angle myself. Problem, outcome, objection, proof, price, demonstration.
02Shoot or gather the real asset: phone video, customer footage, a photograph of the actual work.
03Use image expansion to produce every required aspect ratio from that asset.
04Generate fifteen headline variants and pick three, rewriting them so they sound like a person.
05Add motion to statics for Reels placements.
06Upload three to six genuinely different concepts and let delivery choose.
07Keep the raw footage, because next month's variants come from it.

That is perhaps two hours of tooling saved per creative round, which over a year is real. It is not a replacement for having something to say.

Risks worth naming

Homogenisation. If everyone uses the same generator on the same brief, every ad in the feed converges. Distinctiveness becomes the scarce asset.
Automatic adjustments applied without review, producing crops that cut off your product or your text.
Generated backgrounds that misrepresent the product, which is both a performance and a compliance problem.
Losing the habit of making real assets, which is the input everything else depends on.
Disclosure requirements, which vary by market and are tightening. Check the rules where you advertise.

Turn the automatic adjustments on and then actually look at the previews. I have seen headline text cropped in half in a Stories placement because nobody checked.

Should you use them?

Yes, for reformatting, variation and volume. That is genuine leverage and it costs nothing.

No, as your creative strategy. The tools multiply what you give them, and multiplying nothing gives you nothing. The accounts that improve are still the ones producing three to five real concepts a month; the generator just makes each concept go further.

The creative trends piece covers what those concepts should look like.

Frequently Asked Questions

Does Meta's AI creative actually improve performance?

It improves throughput, which improves results indirectly. Fully generated creative has performed worse than real footage in every account I have tested it in.

Should I let Meta automatically adjust my creative?

Yes, with review. The adjustments generally help delivery, and unreviewed cropping regularly cuts off text and products.

Is AI-generated copy good enough for ads?

For short headline variants, close enough to choose from. For specific claims and anything requiring a point of view, it needs rewriting.

Can I use AI creative instead of hiring a photographer?

For product-on-background e-commerce shots, often yes. For anything involving people, premises or real work, no.

Do I need to disclose AI-generated ad content?

Requirements vary by market and are tightening. Check the rules where you advertise, and label synthetic content where required.

What is the best use of these tools?

Turning one real asset into every aspect ratio you need, and generating variants to choose from. Mechanical work, done fast.

Before You Go

Use the generative tools for the mechanical half: resizing, expanding, varying, adding motion. Keep the concept, the argument and the raw footage human, because that is where the difference between ads lives.

The routine that beats everything else is still five real concepts a month. The creative guide covers how to choose them.

AI for the variations. You for the idea.

WRITTEN BY TARIQ SALLAM
Marketing Consultant. Entrepreneur. Content Creator.

I'm a marketing consultant, entrepreneur and content creator. I help businesses grow through practical marketing, websites, SEO, content and AI.

More About Tariq →

Keep reading