Can people find my new Xiaohongshu note under its target keywords?
To check whether people can find a new note under its target keywords, pair each keyword with known note IDs, search each keyword with xiaohongshu.app_v2.search_notes, and match the note IDs in the returned results. Record each match’s position, then follow whether the note keeps appearing for the keyword and how the observed position changes over time.
For example, a brand publishes three coffee-machine tutorials intended for “coffee machine cleaning”, “steam wand cleaning” and “coffee machine descaling”. Pair each note with the relevant terms: one note can target several terms, and several notes can be observed for one term. Reuse note IDs from publishing records, links or saved search results; locating the IDs needs no extra detail query.
What will the observation help you decide?
Position means the order returned by this API observation. Different times and different App users can see different results. Record absence as an explicit state so later observations stay comparable.
How do you collect comparable records?
- Fix keyword, note IDs, sort and filters. Use
sort_type: generalforxiaohongshu.app_v2.search_notes;comment_descendingaddresses a different ordering. - Choose a time window and depth, such as a similar time each day and one initial page per keyword. Save the actual time, inputs and responses as described in the search guide. Configure scheduled runs in your own task environment.
- Match targets at
data.data.items[].note.id. Some items are not notes, so keep raw page positions and note-item positions separately instead of treating every page as twenty notes. - For greater depth, use returned
data.next_pagewith the first response’sdata.search_idanddata.search_session_id. Start each independent observation at page one.
For example, comparable note-slot positions of 12, 9 and 8 mark a term worth watching. A position of 8 followed by absence from one scanned page gives two states, not a fall to rank zero. To investigate an explanation, read the target note and the nearby content; position records alone do not show that an edited title caused a change.
To hand the task to an assistant with the tools configured, use a prompt such as:
Observe these keyword–note pairs with consistent general ordering and filters, initially one page per term. Retain actual time, query conditions, found positions and absence states. Compare matching historical records and identify terms worth following. Explain why more pages or content reading would help before expanding.
The script track_positions.py below reuses saved responses to build an observation table and adjacent-run comparisons. Collection and local analysis stay separate, so you can analyze saved results repeatedly.
Prepare the saved runs and target list
Create rank-input.json next to the saved responses, and replace the illustrative keyword, IDs, filenames and times. Each keyword can have several target notes, and one input file can hold several keyword keys. Supply at least one saved run for every keyword. Add a later run only after you have actually collected the run; track_positions.py does not schedule, search or retry.
For a multi-page run, append page entries to the same pages array, using the observed next page and the original session inputs. track_positions.py rejects page gaps, changed sessions, mismatched response pages, duplicate request IDs and changed query or filter parameters. An omitted default and an explicit default count as different inputs, so record the same complete parameter set for every run.
Save as track_positions.py, then run this offline command:
Read the deliverables correctly
observations.csv/.json: one row per target note and saved observation.item_position_on_pagecounts every returned item;note_slot_in_saved_runcounts note-bearing slots across the consecutive saved pages. Non-note items do not count as note slots; repeated notes still occupy slots, and the target’s first appearance is kept. The two coordinates are documented observation positions, not a normalized site-wide rank.state: not_found_in_scanned_pages: the position fields stay empty, and the row still reports the pages, items and unique notes scanned. The state does not set a lower bound for every user and does not prove that the note is gone.changes.json: compares adjacent saved observations for the same keyword and note. A positiveobserved_note_slots_upmeans an earlier position in these returned note slots. If either observation did not find the note, the change is null, not a jump from zero. Different page depths are explicitly marked. Identical settings do not eliminate personalization or changing platform results.coverage.json,input-manifest.jsonandraw-*.json: original response bytes, request IDs, recorded page times and hashes. A collection spans the first to the last page time, not one atomic instant. The hashes check local integrity; the hashes do not authenticate query parameters on the server or prove an App-visible ranking. Keep full responses private.
A weekly report can read: “Camping coffee, general order, one page per check. The note appeared in slot 8 on Monday and slot 5 on Wednesday.” If the note is missing from Friday’s first page, record “Not found on page one” and leave the change blank. To explain a difference, review the neighboring notes and what changed between checks.
What does the observation cost?
track_positions.py makes zero calls. The cost of new collection is the number of keyword observations times the pages actually requested, multiplied by the current search price on the pricing page. Thirty keywords once daily, one page each, is 210 calls per week; a second page every time doubles the call budget. A page does not hold a fixed number of notes, and checking several target IDs in one saved response needs no separate calls. Do not add detail or comment calls just to determine position.
Sources: one previously paid search page, replayed locally with a returned target and a deliberately absent synthetic ID, produced found and not-found rows without a new request; that page used comment sorting, not general sorting. Multi-page and two-time-point comparisons were tested with synthetic fixtures. For ongoing monitoring, set up your own collection schedule with a budget; this page creates no automation.

