Presell
Measuring a Presell Page
A new page with its own analytics tells you what the new page did. It cannot tell you whether it beat the page you were already running, because the two are counted by different instruments with different attribution. This guide covers the arrangement that fixes that, and what it will and will not tell you.
Quick Answer
The tag comes in two snippets: one for the page we built and one for the page it replaces. Both sides carry the same instrument, measurement is never metered, and no winner is declared until the sample supports one.
- Sessions, events and experiments are free and unlimited, on both pages.
- No third party cookie, and Do Not Track and Global Privacy Control are honoured.
- A rate is reported as not enough data yet rather than as a number.
- A split that has gone lopsided is reported as broken rather than as a result.
On this page
Why two snippets
Tagging only the new page gives you its numbers and nothing about whether it was an improvement, because the comparison would be against a different analytics product with a different definition of a session and a different attribution window. Two instruments cannot settle an argument about one change.
So you get both snippets every time, and the second goes on the page you are comparing against. The tag knows which is which from the snippet itself rather than from anything the browser sends, which matters because the collector is public and the arm is the one field somebody would most want to move.
Why measurement is free
Sessions, tag events, experiments and control page traffic carry no charge and no volume ceiling, whatever you spend on ads. This is not generosity. If tagging the page you already had cost anything, customers would stop tagging it, and without the control there is no comparison to have and no defensible number to publish afterwards.
It also removes a whole system nobody wanted: no session tier to bill, to explain, or to argue about mid month.
Running a test on the page itself
An in page test varies sections inside one page rather than splitting traffic across two URLs, so there is no redirect, no second address to keep in sync and nothing for an ad platform to treat as a different destination. You choose which sections vary, what counts as a conversion, and how large a sample has to be before anyone looks.
Both arms are identical until you edit the second one, which the panel says out loud rather than letting you discover it. The losing arm is never in a painted frame: the page reveals the assigned arm during parsing, and if anything at all goes wrong, the first arm renders complete. Only the experiment can be lost, never the page.
- Pick the sections that vary and the event that counts.
- Set the sample the test needs before it is read.
- Read the interval rather than the point estimate.
- Promote the winner into the document, which removes the loser.
What it refuses to tell you
No winner is declared for you. A rate comes back as not enough data yet until the sample means something, every rate carries an interval, and a skewed split is reported as broken rather than as a walkover, which is how a blocked script surfaces instead of masquerading as a result.
Promoting an arm the numbers do not favour is allowed, and it is never silent. It asks you to confirm, and the refusal names the verdict and the sample it is based on. Promoting is your call. Calling it a proven winner is not.
FAQ
Does the tag set a cookie?
No third party cookie. It honours Do Not Track and Global Privacy Control, and it collects no personal information. What it records is enough to count views, scroll depth, clicks and conversions per arm per day.
Can I tag a page you did not build?
That is exactly what the second snippet is for. The page you are comparing against is usually one you built years ago, and it has to carry the same instrument or the comparison is not one.
What happens if the script is blocked?
The first arm renders, complete and unflickered, and the experiment is lost rather than the page. A content security policy that blocks it shows up as a broken split rather than as one arm winning.
Can I start a test without running it?
Not today. Creating a test starts it, and both arms are identical until you edit the second, which the panel states plainly. If that proves confusing the fix will be a proper draft state rather than a silent one.
Related help pages
Presell
Exporting and Publishing a Presell Page
Get a page onto your own domain: export one self-contained file, or connect a WordPress site or Shopify store and publish to it. Both are free and unlimited.
Presell
Presell Credits, And What Is Never Metered
What a Presell credit is, what spends one, the long list of things that never will, and why a credit here is not the same unit as an Answer Radar credit.