Read the results of a Purchasely paywall A/B test, from ChatGPT or Claude
Which paywall variant wins, on which platform, and does the result hold through to paid conversion? Your AI reads the A/B test in Purchasely and answers with the numbers.
Published on November 1, 2026
In short
- With Persyst’s Purchasely connector, your assistant compares the variants of an A/B test: views, “buy” taps, trials started, paid conversions and revenue.
- It checks that the gain holds by platform, by country and by plan, not only on average.
- Product or marketing ask with no Purchasely account, and without waiting for the weekly review.
On the “Annual onboarding” A/B test, which variant wins, and does it hold on Android?
Since October 6, variant B converts 4.8% of views into a trial against 3.9% for A, over 21,400 views in total. Paid conversion follows: 2.1% against 1.7%, and revenue per view is 19% higher. On Android alone the gap narrows: 4.1% against 3.8% on trials, and paid conversion is level. The gain mostly comes from iOS, where variant B puts the annual plan forward.
Sources consulted
- Purchasely · A/B test statistics
- Purchasely · paywall conversion by variant and platform
The problem: a variant wins on average, but does it really win?
A paywall A/B test has been running for ten days. The console shows a conversion rate per variant. Before choosing, you need to check that the gap holds by platform, by country and by plan, that it shows up at paid conversion and not only at trials, and that revenue follows. Those checks take several screens and a few exports.
An AI assistant chains them in one conversation, as long as it can read Purchasely.
What the assistant does, step by step
With Persyst’s Purchasely connector, the question “Which variant wins?” runs this reasoning:
- Find the test (
list_purchasely_catalog): its identifier, variants, placements and start date. - Read its results (
get_purchasely_ab_test_stats): conversion and revenue of each variant since launch. - Check each step (
get_purchasely_paywall_conversion, split by variant): views, “buy” taps, trials started, paid conversions. - Test the robustness by reading again by platform, country or plan, to spot a segment where the variant loses.
The answer gives the gap at each step, names the segments where it does not hold and recalls the volume it rests on.
Asking the right question
- Name the test as it appears in Purchasely, or describe it: the assistant finds it in the catalog.
- Ask for the gap at each step, through to paid conversion and revenue, not only trials.
- Name the splits that matter to you: platform, country, plan.
Three questions to try
- “On which placement does variant B make the biggest difference?”
- “Do trials started with variant B convert to paid as often as those from A?”
- “Compare this test with the previous one on the same screen: is the gain of the same order?”
Setup
- An administrator of the Purchasely account creates a Management API key with viewing scopes only, and the Persyst administrator plugs it into Data sources: the Purchasely guide covers the steps.
- They pick the open apps, then open them to product or marketing.
- Each member adds Persyst to their assistant (getting started) and asks their question.
Limits to know
- Read-only: the assistant creates, stops or publishes no test, screen or variant.
- The paywalls’ content is not readable: only their metadata is.
- Purchasely limits the number of calls per key: over a long period or many splits, the assistant may have to wait or narrow the range.
- No significance computation: the assistant shows the volumes, the decision stays yours.
Questions
Frequently asked questions
Does each member need a Purchasely account?
No. An administrator plugs a read-only Management API key into Persyst and opens the apps to the relevant teams. Members read the results with no account in the Purchasely console.
Can the assistant stop the test or publish the winning variant?
No. The connector is read-only: the assistant reads the results and can suggest a decision, but stopping the test and publishing happen in the Purchasely console.
Does the assistant see the paywalls’ content?
No. Purchasely only returns the screens’ metadata: name, versions, placements. To describe what differs between two variants, say it in the question.
Are the figures reliable when the test has just started?
The figures are Purchasely’s, current to within a few minutes. The assistant flags a low volume, but computes no statistical significance: you set the threshold of views or trials before deciding.