A/B Testing Ad Copy and Creative in Agency AI

Quick answer

The fastest test: ask your AI Strategist, "Create three ad variations for my best seller." Each variation comes back as a new ad card for you to approve, so they run side by side.

In Create Ad, Meta Image and Video ads give you three ways to test: Enable A/B testing for copy, the Creative testing format for images and video, and separate ads in one ad set for a clean read on a single change. Give any test about 7 days before you judge it.

Ad copy: Enable A/B testing

On a Meta Image or Video ad, fill in Headline, Description and Caption (Primary Text), or generate them.
Click Enable A/B testing under Caption (Primary Text). It's marked For higher ad spend.
Click + Add headline or + Add primary text to add options. Up to 5 headlines and 5 primary texts in total.
Create the ad. Meta mixes your headlines and primary texts and shifts delivery toward what performs.

Rules:

An ad has one description. There are no description variations.
Once A/B testing is on, it can't be turned off for that ad.
It isn't available for Dynamic ads or USE EXISTING POST.
With separate Feed and Stories/Reels assets, Meta applies the extra headlines and primary texts to feed placements only.

When to use it: when you have the budget. Agency AI's own launch campaigns skip copy testing under $50/day on Meta and give each graphic ad its own copy instead.

Images and video: Creative testing

Choose the Creative testing format (For higher ad spend) when you create a Meta Image or Video ad. Add up to 10 images or 5 videos. Every one you upload is used. Meta combines your assets and copy and decides what to show each person.

A practical set: your best product photo in Square and Story, plus two Genie concepts in both sizes. Six images, three ideas, feeds and Stories covered. See Creating ad banners with Genie.

The format can't be changed after the ad is created. Placement customization, the default, isn't a test: it runs one asset for Feed and one for Stories and Reels.

Clean A/B: separate ads

To learn whether one specific change matters (angle, offer, hook, image style), isolate it:

Create ad A with CREATE NEW AD in an existing ad set.
Create ad B with DUPLICATE EXISTING AD into the same ad set. It copies ad A's copy and creative. Change one thing.
Leave both alone for about 7 days. Edits restart Meta's learning phase.
Compare them in Manage ads: click the campaign, then the ad set, to see its ads with Cost, Conv., Revenue and ROAS. Cost divided by Conv. is your cost per purchase.
Pause the loser with its Status dropdown. Build the next test on the winner.

Or let your AI Strategist do the legwork. Ask "Which of my ads is winning right now?", then "Stop spending on the losing ad" and approve the card.

Copywriting angles are built for this: make the same ad with three angles and let the data pick. See Copywriting angles explained.

What to test first

Image. Lifestyle vs studio, with or without a person, Genie concept vs photo.
Hook. The first line of the caption and the headline.
Offer. Percentage off vs free shipping vs bundle.
Angle. Social proof vs pain point vs value.

One variable per test. Change two things at once and you can't tell which one worked.

Reading results

Wait about 7 days. Meta's learning phase needs about 50 optimization events in 7 days. Earlier numbers are mostly noise.
Judge on ROAS and cost per purchase, not clicks. Clicks show the creative stops the scroll, not that it sells.
Know what you're reading. Manage ads reports each ad as a whole. A copy test inside one ad tells you how the ad did, not which line won. Use separate ads when you need that answer.

Google

Performance Max mixes your assets for you. Give each asset group up to 15 headlines (we recommend at least 10), 5 long headlines, 5 descriptions and 10 or more images (up to 20). To compare two creative directions, add a NEW ASSET GROUP in the same campaign. See Creating an Agency AI Google campaign.

Questions? hello@agencyai.app.