About experiments

Running an experiment allows you to compare the performance of different ad type and/or text ad style settings of your ad units. Experiments help you make informed decisions about how to configure your ad units and can help you to increase your earnings.

There are two ways to run experiments. You can either:

Life of a typical experiment

A typical experiment that you create yourself, follows these steps:

1. Preparing for your experiment

Before you create your experiment, it’s a good idea to:

  • Decide exactly what you'd like to test in your experiment. If there are multiple changes that you'd like to try (e.g., change the text ad background color and increase the text font size), you might want to run two different experiments - first an experiment for the background color change and then an experiment for the font size change.

    Note that the following ad units are not eligible for use in experiments:
    • Idle ad units
    • Responsive ad units
    • Ad units that are already being used in another experiment

2. Creating your experiment

When you create an experiment you:

  • Select the original ad unit that you want to compare the experiment variation against.
  • Select which settings you’d like to change for the variation.
  • Decide whether you want Google to optimize the traffic split for your experiment. We highly recommend that you let Google optimize the traffic split for you, as this will maximize the efficiency of your experiment.

    Learn how Google optimizes the traffic split for experiments

    We use a multi-armed bandit approach to optimize the traffic split for experiments. Each day, we take a fresh look at your experiment to see how the original and the variation performed, and we adjust the amount of traffic that each receives going forward. If one appears to be doing well it gets more traffic, and if one is clearly underperforming it gets less. The adjustments we make are based on a statistical formula that considers sample size and performance metrics together, so we can be confident that we’re adjusting for real performance differences and not just random chance. As the experiment progresses, we learn more about the original and the variation, and so do a better job in choosing the best-performing variation. Experiments based on multi-armed bandits are typically much more efficient than "classical" A-B experiments based on statistical-hypothesis testing. For more detailed information, see Multi-armed bandit experiments in the Google Analytics Help Center.

The “create experiment” page guides you through each step, and provides you with more information. When you’re finished setting up your experiment, you can start it.

3. Monitoring your experiment

Once your experiment is up and running, it’s up to you to monitor its progress.

After 24 hours, you should see data in both the:

4. Choosing the winner of your experiment

Over time, you should see either the original ad unit or the variation outperform the other. To help you decide the right time to choose your winner, we calculate a confidence score for the two settings. This score indicates how likely the original or the variation is to be the better performing (i.e., the highest earning) setting in the long term.

We recommend that you wait until one of the scores reaches at least 95% (indicated by five blue dots) before you choose that setting as the winner.
  • If you choose the original as the winner, then the settings of the original ad unit are retained.
  • If you choose the variation as the winner, then we apply the settings of the variation to the original ad unit.

In either case, we stop splitting your traffic and collecting data, and your experiment ends.

If you have further questions about experiments, see the Experiments FAQ.