Dynamic creative optimization has quietly stopped being a separate product. It is becoming the default way buying platforms choose which ad to show, but the people running campaigns are not keeping up.
A February 2026 EMARKETER survey for Perion of 111 US agency and marketing professionals found that 89.2% call creative important to optimisation, while only 3.6% say creative performance is well understood and actively optimised. In other words, the gap between what marketers believe and what they run is enormous.
What DCO actually does
Instead of trafficking hundreds of finished banners, you supply one template per format, a feed of headlines, images, prices and calls to action, and rules or algorithms that decide what appears for each impression. The system assembles the ad in milliseconds after a demand-side platform wins the auction, then shifts delivery towards combinations that generate clicks, conversions or sales.
- Template: one HTML5 or video shell per ad size with slots for variable elements.
- Feed: a spreadsheet or product catalogue where each row is a content variation.
- Decisioning: fixed rules such as audience, location or schedule, or algorithmic selection per impression.
- Optimisation: the system learns which combinations perform and gives them more impressions.
The arithmetic is both the appeal and the trap. When Facebook introduced dynamic creative in November 2017, advertisers could upload up to 30 assets across five headlines, ten images or videos, five body texts, five descriptions and five calls to action. That multiplies to 6,250 possible ads, but Facebook said it served only a small subset for testing.
Why the numbers should make you pause
Vendors and platforms report strong lifts. Amazon says DSP advertisers using DCO saw an 8.4% lift in paid units, 7.9% more conversions and 7.2% higher return on ad spend compared with those not using it, based on internal data from August 2025. SEGA, meanwhile, says Bannerflow’s social dynamic ads cut a three-week design cycle to about one week and reduced CPM by 44% and CPC by 14%, figures published in August 2026 without a disclosed baseline or method.
None of those numbers has been independently audited. More importantly, thousands of variants spread a fixed budget thin, so many combinations never get enough impressions for a reliable verdict, and the optimiser may settle on an early winner.
Brand control is a separate risk. Assembled ads can combine an image and copy no human approved together. In June 2026, REI said Meta had auto-enrolled it in an AI personalisation tool that produced an Instagram ad showing a malformed bicycle.
What to do about it
Treat DCO as a media-buying control, not a creative strategy. Start with a small set of strong assets, cap the combinations you actually test, and verify results outside the platform that built the variants and picked the winner.
If you run catalog ads, check your defaults. Meta switched on Dynamic Media for Advantage+ catalog ads from September 1, 2025, with an API opt-out, and DV360 removed manual creative rotation for display and video line items from August 12, 2026.
The old question was whether to adopt DCO. The new question is how much of your creative decisioning you are comfortable handing to an algorithm you cannot fully audit.
Source: PPC Land
