X Community Notes Scraper

Export X (Twitter) Community Notes: the crowd-written fact checks that appear under posts. Filter by post, keyword, classification, media, date, and whether a note is actually showing on X.

Why this one is different

Almost every X scraper depends on an endpoint that X can close, and most of them broke when it did. This Actor reads the Community Notes corpus that X publishes itself, as dated snapshots under ton.twimg.com/birdwatch-public-data. No login, no cookies, no proxy, no rate limit, and nothing that stops working when X changes its API.

What you get

One row per note:

Field	Description
summary	The note text shown to readers
tweetId, tweetUrl	The post the note annotates
classification	MISINFORMED_OR_POTENTIALLY_MISLEADING or NOT_MISLEADING
isMisleading	Boolean shortcut for the above
reasons	Why it was flagged: factual error, missing context, manipulated media, satire, outdated, and so on
trustworthySources	Whether the note cites sources
isMediaNote	Note attached to an image or video rather than post text
isCollaborativeNote	Written collaboratively (the newer note format)
createdAt	When the note was written, ISO 8601
noteId, noteUrl	The note itself
noteAuthorParticipantId	Pseudonymous contributor id, as published by X
status, isShowingOnX, statusLabel	Present when Include helpfulness status is on
snapshotDate	Which daily snapshot the row came from

Written notes vs notes people actually see

Most notes never appear on X. A note only displays once contributors who usually disagree with each other both rate it helpful, so the corpus is far larger than what readers ever see.

If you care about the notes that actually shipped, turn on Include helpfulness status and Only notes showing on X. That joins a larger file, so the run takes longer, which is why it is off by default.

Recent notes vs the full archive

By default the Actor reads the recent slice, a few megabytes covering roughly the last several months. Turn on Search the historical archive to include every note back to 2021, about 3 million of them. That downloads a ~480 MB archive, so expect several minutes.

The Actor streams it to disk rather than into memory and stops as soon as your maxResults is met, so a filtered search stays cheap even against the full corpus.

Example uses

Has this post been fact-checked? Put post URLs or IDs into Post IDs or URLs. You get any notes attached to them, or an empty result if there are none.

Track misinformation on a topic. Set Text contains to a term and Classification to Misleading only.

Study manipulated media. Turn on Media notes only with the historical archive, since media notes are largely a pre-2026 population.

Monitor what is actually being shown. Combine Only notes showing on X with Written on or after to see recently displayed notes.

Input

Field	Description	Default
maxResults	Notes to publish	1000
classification	All, misleading only, or not misleading only	all
searchText	Keep notes whose text contains this phrase	—
tweetIds	Only notes on these posts, URLs or bare IDs	—
sinceDate	Drop notes written before this date	—
mediaNotesOnly	Only notes on images or video	false
includeStatus	Join current helpfulness status	false
onlyShowingOnX	Only notes displayed publicly, implies the join	false
includeHistorical	Search the ~480 MB archive back to 2021	false
snapshotDate	Use a specific daily snapshot	latest

Notes on the data

• X publishes with a lag and skips some days. With no snapshotDate the Actor walks back up to 14 days to find the newest snapshot that exists, and reports which one it used in snapshotDate.
• The dataset gives a pseudonymous contributor id, not a handle. That is X's own privacy design and cannot be resolved back to an account.
• tweetUrl uses X's handle-free /i/status/ permalink, because the dataset does not carry the post author's handle.
• The legacy reason flags are all zero on 2026 collaborative notes. They are populated on older notes, so use includeHistorical when you need them.

Output fields

• One row per result with stable keys.

Use cases

• Marketers monitoring social presence and engagement
• Researchers building social-media datasets
• Teams tracking public figures and trends

Related Actors

• Douyin Scraper — Thirdwatch
• Facebook Ad Library Scraper - Active Ads & Creatives — Thirdwatch
• Facebook Marketplace Scraper - Listings, Prices & Alerts — Thirdwatch
• Facebook Scraper - Page Posts, Group Posts, Comments & Pages — Thirdwatch

Last verified: 2026-09

More scrapers at thirdwatch.dev.