How is a simulation different from an A/B test?

An A/B test measures what real visitors do. A simulation predicts which candidate gets more of them to act, before you spend that traffic. Use the simulation to choose what is worth testing. Use the experiment to decide what ships.

Is this a chatbot pretending to be a customer?

No. Personas judge each candidate against the goal you set and say which one they acted on, and why. The result is a prediction, marked as such. It is not a paragraph from a chatbot, and it is not a measured lift.

Do I need a lot of traffic to use Simulations?

No. Simulations do not use your real traffic, which makes them especially useful on lower-traffic sites where every experiment slot counts.

RunPivot key art for Simulations. Introducing Simulations. Pressure-test your ideas before you spend real traffic.

Product launch · September 30, 20266 min read · Published September 30, 2026

Introducing Simulations: pressure-test ideas before you spend real traffic

Today we're launching Simulations. Compare a few versions of a page, set what visitors should do, and see which one personas predict will get more people to act. Then confirm it with a real A/B test.

Real traffic is expensive. Most sites can only run a handful of tests at once, and a lot of that budget still goes to finding something obvious: a buried button, a headline that doesn't match the ad, a page that asks for the sale too early. Simulations is the step before that spend. It does not replace the experiment. It helps you decide which idea is worth one.

See Simulations

What comes back

Take a pricing page. Name the run Pricing page CTA. The goal is to start a free trial. Candidate A is the page you have. Candidate B is a mockup with the trial button in the first screen.

The paid-ad visitor acts on B, because the button is where they look first. The researcher still wants A, because B never shows the proof. That split is the brief for the real test.

Every result is marked predicted. It has not been checked against a live test on your site. It is not a measured lift, and it is not a reason to skip the experiment. When you are ready, the same idea can go into a RunPivot A/B test without starting the brief again.

To access it

In the left sidebar, click Simulations, then start a new persona simulation. Give it a name, such as Pricing page CTA. An existing experiment is optional, so you can start from a page you have not tested yet.
What are you comparing? Add up to four candidates. Each one is a variant, a URL, or a mockup, and one is the control. Mockups can be up to 5 MB. Tall images are split into screens from top to bottom, so a full-page screenshot is more useful than a crop.
What should visitors do? Set a goal, such as starting a free trial, requesting a demo, or clicking the main button. A metric is optional. Personas built from your traffic judge each candidate against that goal and predict which one gets more visitors to act.
Before you run it, check the estimate: model calls, the token maximum, and timing. A run is batched and can take minutes to hours. If it needs more tokens than your plan has left this month, use fewer personas, candidates, or samples. Then click Run simulation.
While it runs, each visitor reads, scrolls, and decides. You can watch who has finished and what they just did, including which candidate they acted on. You can also leave the page. The run keeps going.
When it finishes, you see which candidate each visitor preferred, why, and where they stopped. The result is marked predicted, and it has not been checked against a real test on your site. Use it to choose what is worth testing, then create the experiment and confirm it with real visitors.