How the AI tests your VSL

What it tests, how it splits traffic, when it decides and what stays in your hands, without the jargon.

A person facing a curved monitor full of video versions

Everyone knows they should be testing their VSL. Almost nobody tests often, because every test means preparing variants, splitting traffic, waiting and deciding. VSL Max's AI maximization runs that whole cycle on its own. Here's how.

What the AI tests

The AI works on what surrounds the video and, if you want, on segments of it:

  • Headline above the video, written from the transcript.
  • CTA button: when it appears and what it says.
  • Smart autoplay: the format and wording of the invite to turn on sound.
  • Mini trigger: the small button in the corner of the video that leads to the offer. Whether it shows, when and what it says.
  • Progress bar: its pacing and turning point.
  • Colors: read from your page and applied to the player.
  • Segments: if you upload versions of the opening, problem or offer, the AI splits traffic between them.

The video file is never changed.

How traffic is split

In each test, the current version competes with one or more challengers. Each visitor is randomly assigned a version and keeps seeing the same one if they come back to the page. That way one version's results don't contaminate another's.

The AI changes one thing at a time. If the headline and the button changed together, there'd be no way to tell which one made the difference.

The split doesn't stay fixed at 50/50 until the end. As data comes in, the version that's doing better gets more traffic, and one that's clearly worse loses ground fast. It's a statistical method known as a bandit: it learns while it tests, instead of spending half your traffic on a bad version until the test ends.

When the AI decides

The AI only decides after a minimum volume of visits and results. Before that, any difference could be luck. That's why, with little traffic, it spends a few days in learning mode before the first change, and the dashboard shows when that's happening.

When a version wins with confidence, it becomes the new default, and the AI opens the next test. The current version only goes if it's beaten; a tie changes nothing. And if a challenger drops more than 20% against the current one, it's pulled right away, without waiting for the test to end. The check runs every 15 minutes.

Why sales, not clicks

A sensationalist headline can drive more clicks and fewer sales. That's why the AI judges each test by what it can influence closest to the sale: how many people reach the pitch, how many click the button and how many buy.

To use sales, the AI needs your checkout connected. See how to connect your checkout.

The three levels

You choose the pace for each video:

  • Conservative: starts a test every 7 days, with one challenger, and only applies changes with high confidence. Good for a VSL that already sells well.
  • Balanced: a test every 4 days, with two challengers. It's the recommended starting point.
  • Aggressive: a test every day, with three challengers. Good for a new VSL or one with lots of traffic.

What stays with you

  • One robot per optimization. Don't want the AI touching the headline? Turn off just that one. The rest keep going.
  • Nothing gets overwritten. The starter kit only fills in what's empty, and your headlines go into tests like any other.
  • History with results. Every change is logged with the effect it had, and any of them can be rolled back.

If you turn off every robot, VSL Max becomes just a player. It works, but the AI is what makes the difference.

Upload your VSL. The AI sets up the first test.

14 days or 100 free plays, $0 today. From the transcript come 4 headlines, 3 button texts and the pitch timestamp.

Start with 100 free plays