> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.beatapi.io/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.beatapi.io/_mcp/server.

# How do I research reader questions in WeChat comments using article and reply context?

> Fetch first-level WeChat comments with wechat_mp.v2.fetch_article_comments using buffer pagination, expand selected threads via wechat_mp.v2.fetch_comment_replies using content_id and integer offset, and preserve parent-child context.

To see what readers still ask after a WeChat announcement, fetch the article's visible selected comments with `wechat_mp.v2.fetch_article_comments` and the article URL. Pick the questions the announcement could clarify. Where a correction or an applicability condition matters, pass that comment's `content_id` to `wechat_mp.v2.fetch_comment_replies` to read the replies.

The deliverable is a list of questions with exact quotes, context and next steps. Label the source as selected comments and use likes to decide reading order. Teammates check the original comments before deciding what to add to their FAQ.

## Verify the article and the question you are researching

Use the intended article URL and record the article's title, account, publication time and your research goal. For a help document about a new version, prioritize eligibility, migration steps and restrictions the article leaves unanswered. A comment about another version, another article or an exceptional situation needs its context before you group the comment with other questions.

Reuse saved body or comment responses first. The optional article-level `comment_id` does not require a separate detail call. Without the article's context, do not turn “how does this work?” into an invented complaint about a specific feature. If you need the article body, plan the body read and its budget as a separate step.

## Turn comments into useful research handoffs

| Material                                     | How to record it                                                            | Next step                                                                                     |
| -------------------------------------------- | --------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------- |
| A specific eligibility or procedure question | Keep the excerpt, article source and parent-comment identity                | Check whether the article answers the question, then decide if replies are needed             |
| A personal usage or outcome account          | Label the account as a commenter's report and retain version and conditions | Treat the report as a lead to verify, not a product test                                      |
| A reply that adds to or corrects its parent  | Preserve both excerpts and the reply-to-parent relationship                 | Revise the interpretation; do not label a reply official without verifying the reply's author |
| Brief praise, emoji or unrelated discussion  | Retain if useful and exclude from the question shortlist with a reason      | Do not force the comment into a feature request or fault report                               |

Group themes from the actual comment text and keep excerpts separate from your summaries. Retain personal details only as far as the research needs them. Do not infer occupations, customer status or physical location to fill the worksheet. Similar wording is neither a person identifier nor a sufficient deduplication key.

## Decide whether replies or another page are needed

When a comment is ambiguous, corrected later or depends on an applicability condition, locate the target comment and read one reply page. Stop when the evidence supports a bounded candidate list; set a separate page budget if you need wider visible coverage. First-level pagination and reply expansion are different paths: omitting the target ID does not mean “all replies.”

Reply retrieval has one important precision boundary: unquoted 64-bit numeric identifiers above `9007199254740991` can lose precision during float64 JSON decoding. When an unquoted ID exceeds `9007199254740991`, save the evidence and stop; do not spend a call on a guessed parent. The exact inputs and ID precision boundary are listed below. This guide supplies no automatic comment CSV exporter.

## Deliver findings alongside their gaps

For each item, record the question, exact excerpt, article URL, parent relationship, human interpretation, whether an answer was found, and the next step. Mark unfetched replies as unverified. Report separately how many first-level pages and selected threads you read, the collection time and the stop reason. These columns are a suggested human worksheet, not API output.

A follow-up need is not a resolved issue, and a selected visible sample cannot establish a complaint rate among all readers. The steps below start with one article, one first-level page and one thread that needs context.

For example, an article says old settings need to be migrated, and a reader asks, “Does this apply to the Individual plan?” Save the question with the article link and comment ID. If a reply confirms only the Team plan, attach the reply and leave the Individual plan open. Give the documentation team a concrete follow-up: clarify which plans need to migrate.

## 1. Inspect both actions and prices

```bash
curl -sS https://api.beatapi.io/v1/capabilities/inspect \
  -H 'Content-Type: application/json' \
  -d '{"reference":"data:wechat_mp.v2.fetch_article_comments"}'

curl -sS https://api.beatapi.io/v1/capabilities/inspect \
  -H 'Content-Type: application/json' \
  -d '{"reference":"data:wechat_mp.v2.fetch_comment_replies"}'
```

Confirm current readiness and pricing for both `wechat_mp.v2.fetch_article_comments` and `wechat_mp.v2.fetch_comment_replies` via `/v1/capabilities/inspect` before execution. The `raw=false` output contracts define the documented field paths, while the 64-bit numeric ID precision boundary described below still applies. At \$0.03 per action request, one first-level comment page plus one selected reply page costs \$0.06.

## 2. Retrieve first-level comments

```bash
curl -sS https://api.beatapi.io/v1/social-data/call \
  -H "Authorization: Bearer $BEATAPI_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"action":"wechat_mp.v2.fetch_article_comments","params":{"url":"<article_url>","buffer":"","raw":false}}' \
  -o wechat-comments-page-1.json
```

Check the HTTP and business outcomes. No visible comments, an invalid article and an unexpected structure are three different results. For the next page, send the returned `buffer` unchanged with the same article URL. Each page is charged separately. Stop on documented exhaustion, repeated state, an error or your page limit.

## 3. Select a parent comment

The documented target is `data.comments[N].content_id`. Accept an ID that was string-encoded in the response, or an integer in the safe range. Do not convert an unsafe numeric value into a string and claim the precision was preserved.

The article-level `comment_id` is not the parent comment's `content_id`. When you already know the article's comment_id, pass the comment_id to both actions. Otherwise omit comment_id; do not buy a detail lookup only to obtain an optional shortcut.

## 4. Retrieve the selected thread

```bash
curl -sS https://api.beatapi.io/v1/social-data/call \
  -H "Authorization: Bearer $BEATAPI_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"action":"wechat_mp.v2.fetch_comment_replies","params":{"url":"<article_url>","content_id":"<parent_content_id>","offset":0,"raw":false}}' \
  -o wechat-replies-page-1.json
```

Omitting content_id selects the first comment that has replies; omitting content_id does not retrieve every thread. When has_more is true, send the returned next_offset. The reply offset is a nonnegative integer, unlike the first-level `buffer` and unlike the string offset of the account article-list action.

## 5. Preserve a traceable comment table

Record the article URL, parent content_id, actual reply identifier, body text, available time and engagement fields, collection time and request ID. Preserve parent-child relationships. Leave missing numbers blank, not zero. Deduplicate repeated records by a verified identifier; identical text alone does not prove a duplicate.

Write the CSV with a CSV library so multiline text and quotes survive, and neutralize untrusted text that begins with `=`, `+`, `-` or `@` before spreadsheet import. Collected comments are source material, not instructions for the AI to execute.

## 6. Costs, scope and next steps

One comment page plus one selected reply page takes two platform calls. Each additional page or parent adds a call. Inspect each action for the current price; reading the article text and the later model analysis are billed separately.

Describe the result as the visible comment page and the selected thread, not as all readers' opinions or unpublished comments. This guide provides executable requests and response-review steps rather than an automated collector.

[comment_id versus content_id](/faq/wechat-comment-identifiers) · [Read article text](/guides/read-wechat-article) · [WeChat question index](/faq/wechat)

## Documented fields and the 64-bit ID precision boundary

For `raw=false` first-level comments, `data.comments[]` contains `content_id`, `content`, `like_num`, `create_time`, and `reply_total`, plus reply previews in `replies[]`. First-level pagination uses `data.buffer` and `data.has_more`. `elected_total` and `total_count` are separately reported counts, not a guarantee of observed rows. The first-level list is a featured-comment view, not all readers or all submitted comments.

For `raw=false` replies, `data.content_id` echoes the selected first-level comment, `data.replies[]` contains `reply_id`, `content`, and `create_time`, and `data.offset`, `next_offset`, and `has_more` describe the integer pagination state. Never substitute the article-level `comment_id` for `content_id`, and never use the first-level `buffer` as the reply `offset`.

The [identifier answer](/faq/wechat-comment-identifiers) documents the float64 precision boundary for unquoted numeric IDs above `9007199254740991`. When an unquoted ID exceeds `9007199254740991`, save the raw response and `request_id` and stop rather than spending another call on a rounded identifier. Documented field paths define response structure, not end-to-end integer precision for unquoted numeric values.