Every configuration built on your storefront is recorded — the answers given, and how far it got: saved, shared, added to cart, ordered. The What shoppers choose page turns that into a ranking of the answers to each of your questions.
The column that matters
Which fabric gets picked most is mildly interesting; it is largely a report on what you put at the top of the list. The number worth reading is of those carted, ordered.
The gap between the two is where the surprises are. A choice everybody configures and nobody buys is usually priced wrong, or is promising something the price then contradicts. You could not see that before, and it is the reason this page exists.
A choice nobody has carted yet shows no carts yet rather than 0%. Those are opposite findings and they must not share a number.
Filtering
Pick a product to see one option set on its own, and set a date range to compare a period against another.
How far it got narrows to configurations that reached a particular point — saved, shared, added to cart, ordered, or abandoned. Ticking Ordered answers "what do the people who actually bought choose", which is usually the question worth asking. Tick several to combine them; tick none for everything.
While a status filter is on, the of carts ordered figure is left out. Filter to "added to cart" and orders are excluded from the question by construction, so any rate would read 0% — meaning "none of these were bought" when what happened is that nobody asked about orders.
All the filters live in the address bar, so a view worth showing somebody can be sent as a link.
What is counted
Only questions that offer a fixed set of answers — dropdowns, swatches, cards, chips, checkboxes. Measurements and free-text boxes are left out on purpose: ranking "most common width" gives you one row per shopper, which is a list rather than a finding.
A question you have since deleted is left out too. An answer whose choice was deleted is still counted, under the value it was saved with — the shopper still gave that answer, and dropping it would quietly shrink the totals.
Multi-answer questions count each pick separately. Choosing "Handles" and "Feet" is two answers to one question, not one answer called "Handles, Feet".
Getting it out
Download CSV gives one row per choice — flat, so it sorts and pivots. It respects whatever filters are on screen.
The same report is on the API for anything that needs it on a schedule:
GET /api/v1/option-analytics?from=2026-01-01&to=2026-03-31&status=ordered&format=csv
product_config_id narrows it to one option set, status is repeatable, and
format=json (the default) returns the full structure including per-choice
counts. See
the API for tokens.
Above 20,000 configurations the page counts the most recent 20,000 and says so. Narrow the dates for an exact answer.