Split respondents between paths

Use Split when respondents should see different versions of a survey. For example, send half to a step-by-step workshop introduction and half to an example-led introduction.

This assigns each response to one path. It does not shuffle the order of questions or answer options.

Build the paths

Add the questions, content blocks or sections for each version, and decide where they should rejoin or end.

Add the split

Select the node before the paths, open Advanced → Conditional Logic, add a condition rule and change its action to Split.

Choose who enters

An empty Split condition applies to everyone who reaches it. Add conditions if only certain respondents should enter the split.

Choose destinations and weights

Select a different destination for each path. Use Add path for more than two, or Split evenly to give every path equal weight.

Choose the mode

Select Balanced or Random, then preview each route and check how it ends.

Choose Split as the rule action.
Choose the randomization mode; Balanced is shown here.
Set destination weights and check their displayed percentages.
Add another path or use Split evenly for equal weights.

Local documentation example.

What the weights mean

Weights are ratios, not a required total of 100. Each must be a whole number of at least 1. The percentage beside a path shows its share of the total weight, rounded for display.

Weights	Intended shares
1 and 1	50% and 50%
7 and 3	70% and 30%
1, 2 and 1	25%, 50% and 25%

Balanced or Random?

Balanced assigns the next response to a path that is furthest behind its weighted share. It uses assignments across the survey, with random tie-breaking. Small totals still require whole people: three responses cannot be divided into two equal groups.

Random chooses independently for each response according to the weights. Small samples can be uneven. Revisiting the split within the same response keeps the same path while its setup is unchanged.

Assignments are not completed responses

Balanced mode counts responses assigned at this split, including people who later stop answering. Equal assignments do not guarantee equal numbers of completed responses.

Keep the paths separate

A split changes where someone goes next; it does not automatically skip every other version later. At the end of Version A, add a Go To rule to the shared follow-up or End so that Version B is not shown as well. Keep the normal path meaningful for people who do not match a conditional split.

Randomization

Essential

Split respondents across weighted paths for A/B tests and experiments.

Compare plans

See testing splits and building a complete branching flow.