> 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.

# Which WeChat ID selects a comment's replies: comment_id or content_id?

> Pass the article URL and parent content_id to wechat_mp.v2.fetch_comment_replies, treat article comment_id as optional, and separate buffer pagination from integer offset pagination.

To fetch replies to a specific WeChat comment, call `wechat_mp.v2.fetch_comment_replies` with **the article URL and the parent comment's `content_id`**. The article-level `comment_id` identifies the article's comment container and is optional. To get the parent's `content_id`, find the comment's exact text in the article's first-level comments from `wechat_mp.v2.fetch_article_comments`, then copy `content_id` from the same record.

## Verify that the replies belong to your target comment

| Evidence                                | What to establish                                               | If evidence is insufficient                                                                      |
| --------------------------------------- | --------------------------------------------------------------- | ------------------------------------------------------------------------------------------------ |
| Requested article URL                   | Matches the article being researched                            | Revisit saved requests; do not copy IDs from another article                                     |
| Root text and the root's `content_id`   | The excerpt and the identity come from the same saved record    | Check context for similar wording; do not guess from row position                                |
| Reply envelope's `data.content_id`      | Matches the selected parent, with reliable identifier precision | Stop on a mismatch or uncertain precision; do not attribute the replies to the original question |
| Reply text and the reply's relationship | Only observed additions or corrections enter the notes          | Mark unfetched replies and preview-only material as unverified                                   |

A matching `data.content_id` echo is one part of the identity evidence. The echo does not show that the author is official or that the question was resolved. Review the speaker and context, then separate the question, added conditions, corrections and remaining gaps. A reply's `reply_id` identifies that one reply; `reply_id` is not the parent's `content_id`.

## Investigate wrong replies using saved evidence first

Check the saved request for four mistakes: an omitted target `content_id`, the wrong article, a row position used as identity, or an article `comment_id` passed as `content_id`. Reinspect saved requests and responses locally before another paid call. If an ID first arrived as an unquoted number above the safe integer limit, float64 decoding may already have rounded the ID, and converting the rounded number to a string afterward cannot restore the original digits; the last section of this page explains the 64-bit integer precision boundary.

Request another reply page only after the parent identity is confirmed. The article-level `comment_id` is an optional performance shortcut, not a reason to add a detail query. Continue the target thread with the thread's integer `next_offset`; the string `buffer` of the first-level comment list belongs to a different pagination process.

## Hand off context, not an isolated explanation

Keep the article source, the target excerpt, the parent identity, the actual reply excerpt, the collection time, and the human interpretation together. State when the discussion identity is uncertain or when later pages were not retrieved. A plausible explanation without the question answered is not sufficient evidence.

The table below lists the exact actions and fields to check in saved requests.

## What does each action need?

| Action                                | Required      | Optional identifiers                      | Pagination                                       |
| ------------------------------------- | ------------- | ----------------------------------------- | ------------------------------------------------ |
| `wechat_mp.v2.fetch_article_comments` | Article `url` | Existing article `comment_id`             | Empty `buffer`, then the returned `buffer`       |
| `wechat_mp.v2.fetch_comment_replies`  | Article `url` | Article `comment_id`, target `content_id` | `offset: 0`, then `next_offset` when more exists |

`comment_id` is a performance shortcut. Reuse `comment_id` when you already have the value; without `comment_id`, the action resolves the comment container from the article. Do not add a paid detail request only to fetch one comment page. The documentation points to `data.content.comment_id` in article details, but `data.content.comment_id` is not in the current fixed detail-field list, so do not assume `data.content.comment_id` is always present.

## Does omitting content_id fetch every reply?

No. Without `content_id`, `wechat_mp.v2.fetch_comment_replies` selects the first comment that has replies, as documented. `wechat_mp.v2.fetch_comment_replies` neither fetches all threads nor guarantees the thread you intended. With `raw=false`, the action output documents `data.comments[N].content_id` for first-level comments and `data.replies[N].reply_id` for replies, and the reply envelope echoes `data.content_id`. Keep `comment_id`, `content_id`, and `reply_id` separate.

Input IDs are numeric strings, while output IDs can appear as unquoted 64-bit numbers. Keep IDs that arrived as strings as strings, and do not convert IDs through JavaScript `Number`. Keep the article, parent comment, and reply relationships intact, and never reuse IDs from a different article.

## Why can't I use buffer for a reply's next page?

First-level comments page with a string `buffer`; replies page with a nonnegative integer `offset`. The two pagination states are separate. When `has_more` is true, send the actual `next_offset`, not a guessed increment of 20 and not the string `offset` of the account article-list action.

[Collect article comments and selected replies](/guides/wechat-article-comments) · [WeChat question index](/faq/wechat)

Contract basis: free Inspect of `data:wechat_mp.v2.fetch_article_comments` and `data:wechat_mp.v2.fetch_comment_replies`. BeatAPI does not promise access to unpublished, hidden, or deleted comments.

## Why can adding quotes to a large numeric ID still target the wrong comment?

Unquoted JSON numbers decoded through float64 lose precision above `9007199254740991`. For example, `12109128638545265979` rounds to `12109128638545267000` during float64 decoding, whereas the same ID returned as a JSON string stays exact. Quotes added by a client afterward preserve the rounded digits rather than the original ID.

If a saved response contains an unquoted ID above `9007199254740991`, stop any reply workflow that depends on that unquoted ID. Keep the response and `request_id` for review instead of passing the rounded digits to `wechat_mp.v2.fetch_comment_replies`. Neither `raw=false` nor Python big integers can restore digits already rounded. IDs returned as strings in the response keep full precision.