The short answer
Meta's generative AI ad tools adapt creative you already have: they expand images to new aspect ratios, generate backgrounds and text variations, and animate photos into video. They don't supply the concept. On our board, ads whose production is AI-generated reach the top tier at 2.6%, against about 12% for other tagged production formats.
What the tools do
Meta's generative AI ad features sit mostly inside Advantage+ creative in Ads Manager (Meta's overview). They take an ad you have built and produce versions of it. The main ones:
| Tool | What it does | Where it helps |
|---|---|---|
| Text variations | Writes alternative primary text and headlines from your copy | More copy angles to test without writing them |
| Image expansion | Extends an image to fit other aspect ratios (for example 1:1 to 9:16) | Reels and Stories placements without a reshoot |
| Background generation | Creates new backgrounds behind a product image | Seasonal or setting variations of a product shot |
| Image animation / video expansion | Adds motion to a still or adapts a video's frame | Getting motion placements from static assets |
| Video generation | Turns several product images into a multi-scene video with text and music | Video for brands without a video budget |
| Virtual try-on | Shows clothing on generated models of different shapes and sizes | Apparel catalogs |
The last two were among the tools Meta announced at Cannes Lions in June 2025, alongside video highlights and brand-consistent text and image generation (PPC Land summary).
Adoption is large. Meta said nearly 2 million advertisers were using its video generation features (Image Animation and Video Expansion) in Q2 2025 (Marketing Dive). And Meta reports results: in its Andromeda engineering post, advertisers using Advantage+ creative features saw a 22% increase in return on ad spend, and image generation was associated with about a 7% increase in conversions. Those are Meta's own averages across advertisers, not controlled results for any one account.
One rule to know: since early 2025, Meta labels ads when its in-house tools make a significant edit, and shows the label next to "Sponsored" when the tools generate a photorealistic human.
What they don't do
Every tool in that table starts from an ad you made. None of them decides:
- The concept. What the ad argues and why someone should care.
- The reader. Whether the ad speaks to someone who has never heard of the problem or someone comparing two products.
- The angle. Root cause, story, expert, cautionary tale.
- The proof. A demonstration, a comparison, a named expert.
- The offer. What the person gets and why now.
Those are the decisions that separate the ads on our board. The tools multiply an idea; they don't replace one. That matters more since Andromeda: ten AI variations of one concept are still one concept to the delivery system, as covered in creative similarity.
How AI-made ads fare on the board
We tag the production format of every ad, and one of the formats is AI-generated: ads whose visuals or presenters are clearly made with generative AI. We can't tell which tool made them (most are likely third-party video tools, not Meta's), and Meta's own enhancements like crops and text variations are invisible in the Ad Library. So this is a read on AI-made creative in general, from our October 9, 2026 snapshot of 2,443 live Meta ads with $50k+ in estimated spend.
| Production format | Ads | Share reaching top tier | Median days live |
|---|---|---|---|
| AI-generated | 77 | 2.6% | 97 |
| All other tagged production formats | 533 | 11.8% | 140 |
77
AI-generated winners, from 44 brands
2
of them in the top tier
33
are narrated stories
$229K
median estimated spend, close to the board's $223K
Top tier means the top 10% of the board by Ad Radar's winner score (live 21+ days). AI-generated ads reach the board and spend at a normal level: median estimated spend is close to the board median. But they rarely reach the top tier, and they get replaced sooner, a median 97 days live against 140 for other production formats.
The likely reason sits in that third number. Almost half are narrated stories, many built around a dramatic personal setup, and 40 of the 77 are health ads. The format is cheap to produce, so many brands produce the same kind of ad, and it wears out fast. Our AI UGC ads analysis looks at the format in more detail.
Where AI production shows up
AI-made ads on the board are not spread evenly. They lean heavily toward the top of the funnel:
| Reader awareness | AI-generated ads | Share of AI ads | Share of whole board |
|---|---|---|---|
| Unaware | 37 | 48% | 34% |
| Solution-aware | 22 | 29% | 20% |
| Problem-aware | 15 | 19% | 20% |
| Product-aware or most aware | 3 | 4% | 14% |
Almost half are written for unaware viewers, and the format of choice is the narrated story (33 ads) or the voiceover listicle (19). Very few AI-made ads speak to people who already know the product. That fits how teams use the tools: AI is cheap for volume at the top of the funnel, where a story has to stop a stranger. It is also where originality wears out fastest, which helps explain the short median life.
Three AI-made ads that worked
Teddy Roosevelt drank a gallon of coffee every single day.
- Est. spend
- $2.9M
- Days live
- 210
- Format
- Video, 184s
A historical hook with AI visuals, live 210 days. The idea (a surprising fact about a famous coffee drinker) is what carries it, not the production.
Never ignore these three hidden distress dog behaviors: paw licking, itching, and head shaking.
- Est. spend
- $520K
- Days live
- 110
- Format
- Video, 50s
One of only two AI-generated ads in the top tier. A specific, checkable list for owners who already see the problem.
Stop giving your cat water in a bowl.
- Est. spend
- $765K
- Days live
- 80
- Format
- Video, 154s
- Variants
- 2
The brand also runs founder-led videos with the same argument. AI production here is a second execution of a proven concept, which is the safest way to use it.
How to use Meta's AI tools well
- Use them for adaptation, not ideation. Image expansion and video expansion save real production time on placements. Let them do that job.
- Start from a proven concept. Run AI variations on an ad that already works, the way KittySpout pairs AI and founder versions of one argument. Don't ask the tools to find a concept.
- Test enhancements against the original. Run the enhanced and plain versions side by side for a few weeks. Meta's averages don't tell you what happens in your account.
- Check every generated element. Generated text can drift from claims you can support, and generated backgrounds can misrepresent a product. Compliance stays your responsibility.
- Count concepts, not outputs. Twenty AI variations add little reach if they are one idea. Our creative diversification playbook shows how to plan new concepts.
For the concept side, a good starting point is what already wins in your niche. Ad Radar can turn a winning ad into a creative brief, and the AI creative brief guide shows how to use Claude for that step, where the thinking happens before any tool generates anything.
Figures marked as estimated spend come from Ad Radar's model of engagement on public Meta Ad Library ads. They are estimates, labeled as such, and are best used to rank ads against each other.