How-to seriesSeptember 28, 2026
How to research Chinese social platforms
Research Chinese social platforms with source-preserving queries, platform-specific endpoints and translations that keep uncertainty visible.
Researching a brand across Chinese social platforms starts with naming it correctly. An English brand name, its official Chinese name and a local nickname can return different discussions. Translation after collection cannot recover posts that the query never found.
Vaaya's catalog includes platform-specific social-data routes. Use them to produce a source-preserving research table, with the original text and a translated summary kept together.
Prepare the terms and scope
Choose the business question, date window and platforms. Start with a small set suited to the content you need. Product notes on Xiaohongshu and a Weibo discussion can answer different parts of the same question.
Collect the official brand spelling, Chinese spelling and relevant product names from first-party sources. Ask someone familiar with the language or market to review ambiguous terms. Write exclusions for names shared with unrelated brands or people.
Connect your agent from Vaaya's installation guide, then use:
Research [brand/product question] on [Chinese platforms] between [dates].
Use these verified names and aliases: [terms]. Exclude [unrelated meanings].
Consult Vaaya for current routes, required inputs and quotes. Budget: [amount].
Keep original URLs, platform IDs, original text, dates and native metrics.
Add an English summary beside each source; flag uncertain translations.
Separate brand posts, independent discussion and unknown sponsorship.
Report coverage gaps. Do not post, message users or access private accounts.
1. Discover a route for each operation
Use vaaya/discover with the platform and operation in query. Inspect the action, endpoint and required fields returned. The social-data reference explains how the endpoint becomes a parameter in tikhub/fetch or tikhub/submit.
The current catalog lists these read routes:
| Platform | Listed endpoint | Required endpoint inputs |
|---|---|---|
/api/v1/weibo/web/fetch_search |
keyword |
|
| Xiaohongshu | /api/v1/xiaohongshu/app_v2/search_notes |
keyword |
| Bilibili | /api/v1/bilibili/web/fetch_general_search |
keyword, order, page, page_size |
This is a routing reference, not a report of successful research calls. Confirm accepted values for optional or enumerated fields before execution. If you add Douyin, discover its specific search route instead of substituting its name into another platform's path.
2. Collect a bounded first sample
Run the approved aliases with a small page limit. Keep the query that produced each row. It helps explain why a nickname surfaced a discussion the official name missed.
Inspect the response structure before processing the entire batch. Posts, notes, videos and users may appear in different sections. Preserve platform IDs, canonical URLs and timestamps when provided. Filter the actual dates against your window if the endpoint does not enforce it.
For a follow-up read, use the identifier required by that route. A display name, account handle and platform user ID are not interchangeable. Carry forward IDs returned by the first call instead of guessing them from a URL.
3. Translate without losing the evidence
Keep a short original-text excerpt or paraphrase in the source language beside the English summary, along with a link to the full post. Label machine translations so a reviewer knows where wording may need attention.
Ask the agent to flag uncertain slang, irony, abbreviations and product variants. Do not force every post into positive, neutral or negative when the meaning is unclear. A complaint quoted inside a rebuttal should not automatically count as the author's complaint.
Separate observations from interpretations. A post may show a particular use case; it rarely establishes how common that use case is across a country or age group.
4. Compare themes without combining incompatible metrics
Use this illustrative output schema for the research file:
| Field | What to retain |
|---|---|
| Platform and source | Original URL, content ID and query |
| Original text | Relevant source wording with context |
| Translated summary | Meaning and uncertain terms |
| Timing | Published time and collection time |
| Native metrics | The platform's own labels and values |
| Review finding | Theme, counterexample or question |
Keep likes, saves, views and comments in their native columns. Missing values stay unknown. An unavailable view count does not mean nobody viewed the content, and totals from different platforms should not be presented as a unique audience.
Handle incomplete access
If an endpoint errors or requires inputs you cannot establish, stop that route and record the gap. Repeated paid retries do not make an unsupported input valid. If a search returns no matching rows, describe that request's result without declaring the platform silent.
Save the query set and actual call receipts with the brief. Give the reviewer the strongest source-linked findings and uncertain translations first, so the next research step can resolve a specific question.
Questions
Can I use the same endpoint and parameters for every Chinese platform?
No. Discover the route for each platform and operation. Even similar searches can require different identifiers, pagination fields and input names.
Should I discard the original Chinese text after translation?
Keep the original source link and relevant text alongside a translated summary. Preserve ambiguous slang, names and product terms for review.
Does a few-platform review represent the whole Chinese market?
No. State the platforms, queries, dates and returned sample. A bounded social review can identify useful discussions without measuring the entire market.