Demand Gen A/B experiment: 50 conversions per arm, then what?

Google asks for at least 50 conversions per arm in Demand Gen conversion experiments. What that means for how many AI video variants you can afford to test.

5 min readSume
All posts

Google's Demand Gen A/B experiments need a minimum of 50 conversions per arm to surface results for conversion-based metrics, and about 100 data points to start seeing them. So every extra AI video variant you add multiplies the conversions you need. Two videos need two arms of 50; four need four.

Figures are from Create A/B experiments for Demand Gen campaigns, read 2026-10-03.

The rules on Google's page

Two experiment types exist: Asset A/B experiments, which test creative as a single variable, and Custom experiments, which test audiences, bidding, formats or creative with more than two arms.

Demand Gen experiment rules, from Google Ads Help (read 2026-10-03).
TopicWhat Google says
Minimum campaignsAt least 2 Demand Gen campaigns, both ready but not running
VariableChoose campaigns that differ in only one variable
Traffic split50% recommended for the best comparison
Conversion metricsA minimum of 50 conversions per arm to surface results
BudgetTesting budget as a variable is not recommended at this time

Turn the minimum into a variant count

Estimate how many conversions your account makes in the time you can wait, divide by 50, and render no more variants than that gives you arms for. If you expect 120 conversions in the window, two arms is the honest ceiling, not four.

On Sume, one extra variant is one extra avatar video job per Generate avatar video. The render cost is small next to ad spend, so the conversion count, not the render bill, is what limits the test.

Keep the pair clean

Because Google wants one differing variable, change only the hook scene between the two videos. Keep the same avatar_handle, aspect_ratio, quality, scene and caption style.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume