Skip to main content
Every list endpoint uses cursor pagination. You ask for a page size, the response tells you where the next page starts, and you stop when there isn’t one.

Request

Response

Every list response has the same shape:
The last page can be full: a page with exactly limit items may still be the last one. Always use next_cursor, not the item count, to decide whether to continue. You never get an empty trailing page.

Sort order

Each list has a fixed order, so pages never overlap: GET /projects/{project_id}/visibility isn’t paginated. It returns one point per tracked day for a range of at most 366 days. On the reporting endpoints, limit counts answers, and a page always holds whole days. A page can run past limit by the rest of its last day.

Cursors are stable

Cursors point at the last item you received, not at a numeric offset. Adding items while you page doesn’t shift the pages, so you never see an item twice. Items that sort after your position (a new brand or prompt, for example) show up on a later page. Items that sort before it (a claim seen again in fact checks, since they’re newest first) show up on your next pass from the first page. Citation counts can also change mid-pass: a source whose times_cited rises past your position moves to an earlier page.

Rules for cursors

  • Treat cursors as opaque. Their format can change at any time. Don’t decode, build or modify them.
  • Reuse a cursor only with the endpoint and filters that produced it. Mixing them gives unpredictable pages or 400 invalid_cursor.
  • An invalid cursor returns 400 invalid_cursor. Start again from the first page, without cursor.

Fetch every page

Filters

Filters combine with pagination. Unknown query parameters are ignored, and an empty value (?status=) counts as absent.

Next steps

Export to a spreadsheet

Page through every list and write it to CSV.

Rate limits

Pace bulk exports so they never hit a 429.
Last modified on November 18, 2026