- Draft: what visitors would get after you publish. It uses your unpublished Brand changes (personality, opening line, things to do and avoid) and the Knowledge articles you saved as drafts.
- Live now: what visitors get today.
Question sets
A set holds up to 50 questions; a workspace keeps up to 20 sets. Each question has:- the question, in English, Spanish or Portuguese, as visitors would ask it;
- optionally, the page it is asked on (an
https://address), so the guide can use the targets and actions declared for that page; - its own rules.
Rules
A rule on the set applies to every question in it; a rule on a question applies to that question. A question has up to 6 rules, and so does a set.Reading the results
Each question shows the draft answer and the live answer side by side: the text, the sources it cited, what it pointed at, and whether it answered. Changes between the two are named (answer, sources, what it points to, whether it answers), and every rule shows whether it passed on each side. A failing rule links to where you fix it: Personality, Knowledge, Guide behavior, or targets and actions. A test turn has no visitor and no page in front of it. Puntego reports the target or action the guide chose; whether it is visible is checked on your site.Publish waits while a rule fails
While the latest completed run of any set has a question whose draft answer fails a rule, or could not be produced, publishing is paused:- Settings → Brand → Live status lists the failing questions and keeps Publish draft disabled;
- publishing a Knowledge article, or saving a change to a live one, is refused until the tests pass. That includes adding a note (it goes live at once) and restoring a published version. Saving it as a draft still works.
- approving a Knowledge article that Copilot drafted (Inbox → Copilot → Draft review) waits too, because approving it publishes it at once. The draft review lists the failing questions and keeps Approve exact change disabled for that article; the draft stays open, and you can still reject it. Other drafts (an AI walkthrough, which is created as a draft) are not affected.
A failure is checked again
The same question can get a slightly different answer each time. So when a rule fails on the draft, Puntego asks the draft that question again, up to 2 more times, and the question only pauses Publish when it fails on most of its tries (2 of 3). The result says how it went, for example “Failed 1 of 3 tries. Passes on the majority, so it doesn’t pause Publish.” The draft answer shown is the first one. Each extra try uses one test answer; if the month has none left, the tries that ran decide: a single failed try pauses Publish, and a tie (one failed, one passed) shows as not checked, which also pauses Publish until you run the set again.Tests older than your changes
If you change anything the guide uses after your latest run (Brand, Knowledge articles, guides and their walkthroughs, corrections, guide behavior rules, targets and actions, Verified Actions), Live status (and Copilot’s draft review, when an article draft is waiting) says your tests ran before your latest changes and links to Tests. It is a reminder: Publish stays available. Run your sets again to check the current draft.Runs and limits
A run starts in the background and keeps going if you leave the page; you can cancel it, and the answers it already produced stay. One run per set at a time. Each question is answered twice, so a run of 50 questions uses 100 test answers, plus one for each extra try of a failing question. Test answers have their own monthly allowance (calendar month, UTC) and never count as visitor conversations, AI replies, or analytics (Analytics → Runtime tokens and cost leave them out):
A run that would go over the month’s allowance does not start. A canceled run gives back every answer you were not shown, including the question it was answering when you canceled. A set keeps its 10 most recent runs, and always its latest completed run, the one Publish checks.