Read A/B test results, the weekly trend, and testing with low traffic
You'll learn how to read a test's numbers and weekly trend, what the time-to-verdict estimate tells you, and how to test honestly when your site gets little traffic.
Know what your test can tell you before you start, read what it's telling you while it runs, and make a sensible call when traffic is thin. RightMessage never stops you from testing because of traffic, but it's honest about how long a verdict takes. Verdicts can take months, or may not come.
Before you start: the estimate
The setup card and the Start dialog show two numbers, estimated from the last 28 days of the campaign's visitors and goal conversions:
| Number | What it means |
|---|---|
| Time to a verdict | "About N weeks", if one version is about 20% better than Control. |
| Smallest difference it can spot | "N% or more", reliably spotted within 26 weeks. Smaller real differences will probably end without a verdict. |
With little traffic, Time to a verdict reads "Probably over 26 weeks", and the note says "You can still run the test and choose a version yourself." Without enough recent history, both read "—": "There isn't enough recent visitor and goal history to estimate how long this test might take. You can still run the test." Starting a test needs Pro.
To get a sense of scale, two real (anonymized) campaigns measured in October 2026:
- A paid-ads campaign with about 356 visitors a day, converting about 11%, needs about 9 weeks to detect a 20% lift. The smallest lift it reliably detects is about 8%.
- A typical small account sees about 16 visitors a day, against about 609 a day needed for a timely verdict.
While it runs: the versions table
Each card's versions table shows Visitors, Conversions, Rate, Difference, and Likely range vs Control for each version. The likely range is the set of differences that still fit the results. Early on it's wide; each weekly check narrows it. A version wins when its whole range is better than Control and loses when it's all worse. See the Testing tab reference for every column and tag.
Results are checked every Friday, and each week's visitors get 7 days to convert before they count. So the numbers on the card change weekly, not live. The earliest verdict comes at the week 2 check, about three weeks after the test starts, and only once every version has at least 20 visitors who converted and 20 who didn't.
Goal value, if your goal has one, appears next to results but never decides a winner.
The weekly trend
Weekly trend opens "Difference from Control by week": the Observed difference for each challenger, with its Likely range band, from the first weekly check onward.
The trend is descriptive. "At low traffic the difference swings from week to week. Only the weekly checks can call a verdict." Use it to see whether a difference is steady or bouncing around, not to pick a winner early.
The 26-week limit
The last check is at week 26. A version with no verdict by then is No clear winner, and the test is Ready to end. That doesn't mean the versions are the same: it means the difference, if any, was too small to show with your traffic. Tests never end on their own, so you decide what to keep.
If more than four weekly checks are missed, the test reads Can't reach a verdict. End it and choose a version yourself, or restart it to measure again.
Testing with low traffic
Low traffic never blocks a test. When the estimate says a verdict is unlikely within 26 weeks, you can still learn something:
- Test bigger changes. A different offer or headline angle is more likely to show a large difference than a reworded button.
- Use fewer versions. Each challenger splits the traffic and the allowed error, so each one needs a little more evidence.
- Pick a goal that happens more often, such as a click toward your demo page rather than a closed deal, if it's a fair stand-in for what you care about.
- Test where the traffic is. An Original page test covers everyone who doesn't match an audience, which is often most visitors.
- Decide anyway, honestly. If the trend is steady and the range leans one way, you can end the test and keep that version. The card will say it wasn't proven, and that's fine as long as you know it.
For more on planning around traffic, see A/B testing on a low-traffic site. The rules behind every verdict are in how verdicts work.