Skip to main content

FAQ

Does this work only for logged-in users?​

No. It works for both anonymous and logged-in users.

  • For anonymous users, recommendations are inferred in real time from on-site behavior and request context.
  • For logged-in users, user_id adds a longer-term view of taste and preference.

What should we use as the visitor identifier in recommendation requests?​

Use user.device_id as the required visitor identifier.

  • Set it to Snowplow domain_userid (long-lived browser identifier).

Can recommendations stay consistent across devices?​

Yes. Send the same business user_id across devices to keep recommendations consistent across sessions and surfaces.

Browse and Search use cursor-based pagination.

  • First page: omit pagination.
  • Next pages: send both pagination.page and pagination.cursor.
  • Set pagination.cursor to the previous response recommendation_id.
  • Keep the same limit for the full pagination chain. Changing limit during pagination is not supported.
  • If the cursor is invalid or expired, the API returns 404. Start again from page 1 without pagination.

Pagination behavior depends on mode:

  • Personalized mode: page numbers are UI labels. Jumps like 1 -> 5 -> 10 are allowed, but the cursor still returns the next unseen result set in the same stream.
  • Ordered mode: deterministic sorts (for example price/name/date) support true non-sequential page jumps.

In personalized mode, a higher page number does not mean better-ranked results.

Where can I find filter syntax and examples?​

Use the Filters section as the source of truth for:

  • filters vs facet_filters
  • supported operators (including $elemMatch)
  • operator behavior (including AND logic for multiple operators in one field object)

Do we need infinite scroll, or can we use traditional pagination?​

Both are supported.

  • Use infinite scroll if you want smaller loads and faster interaction feedback.
  • Use traditional pagination if it fits your UX better.

Recommendation quality and sequence are preserved in both approaches.

Is request limit the same as the number of returned items?​

Use these fields as page-based terms:

  • limit (request): page size. One request returns one page.
  • count (response): number of items returned on this page.
  • total (response): total number of results that match the full query/filters.

In most cases, count equals request limit.

On the last page, count can be smaller than limit.

Can a recommendation_id expire? Can we store it in browser session?​

Yes. Recommendation context is short-lived.

  • Each response includes expires_in (seconds).
  • expires_in is dynamic and commonly around 300 seconds.
  • You can store recommendation_id in browser session for active pagination.
  • When it expires, pagination returns 404 and you should start a fresh first-page request.

What sort and facet fields can clients use?​

Sorting and faceting are configured in the dashboard.

  • You can use any field that is configured in the Dashboard.
  • featured is supported for personalized ranking (including merchandising and experiment logic when enabled).
  • Configured numeric or string fields can be used in ascending/descending order.

Do User Tracking API events update recommendations?​

Yes. Tracking events (for example page views, list views, clicks, and add-to-cart) inform recommendation updates.

Events are tied to device_id and, when available, user_id. Sending user_id enables cross-device continuity.