Meta’s AI can make your ads now. It still can’t decide what they should say.

This all came out of Advertising Week New York (October 5 to 8). Andrew Hutchinson at Social Media Today had the most complete writeup, and a few other outlets covered it too. What shipped, and who can actually use it:

Zoom out and the direction is obvious. Meta spent the last couple years automating where your ads run, and now it wants to automate making and managing them too. Earlier this year it started pulling placement exclusions out of ad sets (I wrote about that in Meta is removing placement exclusions). This week it went further and started generating the ads.

Production stops being the constraint

For most small brands the thing capping testing has always been production. Every new video needs a shoot or an editor, so you test maybe a few concepts a month. That cost mostly goes away with image-to-video and Ads Creative Studio. You could take your best statics, turn them into video this afternoon, and have ten hook variations running by the weekend.

Inputs become the constraint

Meta’s generators pull from your product photos, your catalog, your past ads and, once the import ships, your brand guidelines. They’re really good at making more of what you already have.

What they don’t know: that your best buyer is a nurse on night shifts. That your biggest competitor has been running the same “clean ingredients” claim for six months straight. That gift-givers need a totally different first line than people buying for themselves.

And the photos matter more than people think. Honestly, the product shots on most PDPs aren’t good enough to generate great ads from, and when we swap in better product photography for a brand it’s the single biggest jump in output quality we see. Same idea with Meta’s tools.

Why does any of this matter so much? Delivery. Meta now matches each ad to whoever’s most likely to respond, and if two ads are near-identical it basically treats them as one ad. So ten tweaks of the same concept mostly land in front of the same people, while a few ads built for different buyers each find their own crowd. Variation helps an ad that already works. It doesn’t give you new ads.

This is why when I review batches from our team I cut anything that has almost the same format and copy as another ad in the batch. We ask for 15 to 20 ads per batch, highly varied, roughly half BOF and half fresh takes on the brand’s top ads from the ad library.

Expect more of the same in feeds

If every brand in a category drops similar product shots into the same generator, everything starts looking the same. Black Friday feeds will be wall to wall discounts, and what cuts through is going to be ads built around a specific buyer with a specific problem. Slick motion won’t save a generic ad.

Checkout is moving into the chat

With Messenger checkout on Shopify and the Meta AI checkout test, more purchases may start and finish without ever leaving Meta’s apps. That makes the conversation your ad kicks off a lot more important. It lines up with something Adobe has been tracking on the open web too, where traffic from AI assistants to retail sites keeps climbing fast. People ask an assistant before they buy now, and an assistant can only recommend a product it actually understands.

You’ve got until November 27, which is Black Friday this year. All of this slots into the BFCM 2026 creative calendar, where the goal is having proven ads live by Halloween.

1. Write down who you sell to. Two to four personas. For each one: the problem, the objections, the actual words they use. Meta’s tools can’t make this part for you.
2. Get a one-page brand guide together. Voice, claims you can and can’t make, colors, fonts, logo rules, banned words. We ask every client for this. When a brand doesn’t have one we pull fonts and colors from their own ads in the ad library, which works, but it’s slower. Have it ready for when Meta opens guideline imports to your account, and use it with every other AI tool in the meantime.
3. Clean up your asset library and catalog. Generators reuse whatever’s in there. Delete the off-brand photos, fix product titles, add clear prices, sizes and benefits. Upgrade the product photography if you can.
4. Turn proven statics into video, then check every single output. Pick statics that have held their cost per purchase for weeks and run image-to-video on them. Then look at each one before it goes live, especially claims and whether the product looks right. We’ve caught a quarter-zip generated with no zipper, a product rendered see-through when it’s solid, and a tin that came out warped because the model copied it off a reference ad. Automated QA missed a few of those. A person didn’t.
5. Vary winners. Don’t use variation instead of concepts. New concepts go first, one per persona. Once something wins, use Ads Creative Studio to play with its hooks, length and end scenes.
6. Put the assistant on a leash. It can draft changes and send you reports all day. But through Cyber Week, nothing touches budget or targeting until someone on your team approves it. Use Customer Lifecycle Strategy to split new-customer and existing-customer campaigns too, so each gets its own message.

Copia does the step before Meta’s tools. It reads your site and products, studies the ads your competitors are running live on Meta, builds buyer personas, and generates Meta-ready ad creative for each one. What you get is a handful of concepts that differ from each other and from what competitors are running, and that’s exactly the input Advantage+ creative and Ads Creative Studio want.

Self-serve starts at $99 a month. If you’d rather hand it off, Managed ($2,000/mo) gets you 200 statics and 2 videos delivered a month, finished and ready to ship.

Want to see what your category already looks like before you pile on? Drop a competitor into the free Competitor Ad Teardown. You’ll get every Meta ad that brand is running, with the hooks and angles broken out.