Ask about this experiment

Most questions about a test are the same three: is this significant, can I trust it yet, and what do I do now. Every experiment has a field in its header, on every tab, where you type that question. The answer appears in a panel beside the page as it is written, in the language you asked in, leading with the answer and then the reasoning.

It reads the figures the Statistics tab shows, computed the same way, for the objective, currency, method and correction chosen there: conversions, rates and revenue per variant, significance, the probability of being best and the expected loss, the power analysis and how long the test still needs, the traffic split check and the secondary objectives. It never invents a number and never calls a variant a winner when the page does not. It also reads the experiment's targeting, variants, objectives and schedule, its brief, and its debrief, which it treats as the team's reading of the result rather than as a measurement.

On a draft it helps before the test starts: whether the settings do what the brief intends, whether the hypothesis names a change, an effect and a measure that can show it, and what the run length planner's inputs imply. It only reads. It does not start, stop or edit anything, and it cannot see a variant's code or individual changes, screenshots or the activity log. Tokens, keys and email addresses in targeting rules and redirect URLs are hidden before anything is sent.

Each experiment has one conversation, shared by everyone who can open it, Members included, and each question shows who asked it. When the result has changed since an answer was given, that answer is marked as out of date. Conversations are kept for 12 months, and each answer uses AI credits from the account that owns the website, with the cost shown under it.