Twitter Keywords Scraper Pro

Collect public X (Twitter) posts without logging in, without cookies and without an API key. Search by keyword, walk an author's recent timeline, fetch exact post URLs, or open a conversation thread — then export clean rows with the author, engagement counts, media, quoted post and reply context already filled in.

Coverage note: X puts its own full search behind a login. This actor uses the public Bing News x.com index plus public post pages, so it returns a real subset of public X posts, not X's complete search index. That hard limit is explained below.

What you get

Every dataset row is one post (type: "post") with:

• Post: id, url, text, createdAt, language, conversationId
• Engagement: likeCount, replyCount, retweetCount, quoteCount, bookmarkCount, viewCount, engagementRate
• Flags: isReply, isRetweet, isQuote, isNoteTweet, possiblySensitive
• Reply context: inReplyToTweetId, inReplyToUserId, inReplyToUsername, parentTweet
• Quote context: quotedTweetId, nested quotedTweet
• Entities: hashtags, mentions, urls, cashtags
• Media: photo, video or GIF with url, previewImageUrl, width, height, variants, durationSeconds, altText
• Author: id, username, name, bio, followers, following, location, website, profileImageUrl, bannerUrl, createdAt, postCount, mediaCount, verified, isBlueVerified
• Attribution: searchTerm (keyword, handle, URL or conversation ID) and collectedAt

Empty values are omitted at every depth. Missing counters are omitted, never invented as zero. An empty run writes a RUN_SUMMARY to the key-value store and leaves the dataset empty — there are no status or sentinel rows.

Four modes

Mode	Input	What it collects
search	keywords	Public posts found for each keyword through the Bing News x.com index, optionally supplemented by X guest search when you supply a proxy.
byAuthor	authorHandles	The profile page's metadata and recent posts, plus the public syndication timeline (up to ~21 posts per author).
byUrls	postUrls	Exact posts from URLs or IDs, using the post page, syndication, FxTwitter and oEmbed fallbacks.
conversation	conversationId	The focal post plus the reply previews X renders publicly, with the focal post kept first.

Quick start

Keyword search

{
  "mode": "search",
  "keywords": ["apify", "#AI", "OpenAI"],
  "maxItems": 50,
  "sort": "Latest",
  "language": "en"
}

Posts by author

{
  "mode": "byAuthor",
  "authorHandles": ["@apify", "https://x.com/OpenAI"],
  "maxItems": 20
}

Specific post URLs

{
  "mode": "byUrls",
  "postUrls": [
    "https://x.com/apify/status/2118015298603274372",
    "2117931222320259247"
  ]
}

Conversation thread

{
  "mode": "conversation",
  "conversationId": "https://x.com/apify/status/2118015298603274372",
  "maxItems": 20
}

Inputs

Input	Type	Default	Notes
mode	select	search	search, byAuthor, byUrls or conversation. Required.
keywords	string list	–	Used by search. A single string is also accepted.
authorHandles	string list	–	Used by byAuthor. Accepts @handle, handle or a profile URL.
postUrls	string list	–	Used by byUrls. Accepts post URLs and bare IDs.
conversationId	text	–	Used by conversation. Accepts an ID or a post URL.
maxItems	integer 1–1000	20	Per keyword/author in search and author modes; total cap in URL and conversation modes.
sort	select	Top	Top (engagement), Latest (newest), Top + Latest (top half by engagement, then the newest remaining posts).
language	select	Any language	Any language or an ISO-639-1 code. Applied after collection.
dateFrom, dateTo	date pickers	–	Inclusive UTC date bounds.
excludeRetweets, excludeReplies	boolean	false	Drop reposts or replies.
onlyVerifiedUsers	boolean	false	Keep verified or blue-check accounts.
onlyImages, onlyVideos, onlyQuotes	boolean	false	Keep photo posts, video/GIF posts, or quote posts.
minimumLikes, minimumRetweets, minimumReplies	integer	–	Drop rows below the threshold; rows whose count is unavailable are dropped and reported.
mentioning, author, inReplyTo	text	–	Client-side operator filters: mention handle, author handle, reply target handle.
proxyConfiguration	proxy	–	Optional. Nothing needs a proxy; when supplied it is also used for the best-effort guest search supplement.

No cookie, token, API key or proxy field is required. Legacy saved inputs that use maxTweets, searchType, startDate, endDate or tweetLanguage keep working: the actor maps those keys in code even though they no longer appear in the form.

How collection works

1. Discovery (search) — queries Bing News RSS for site:x.com <keyword> with plain variants (the keyword, quoted phrase, x.com, twitter, news, today, 2026, update, announcement) and collects unique post IDs. A small job stops after one request; large jobs page through first=13,26,39,52 with a roughly 460-ID ceiling.
2. Hydration — fetches each ID from the public syndication CDN in parallel. Deleted, protected and tombstoned posts are skipped and counted in the summary. Local probes hydrated 197 of 202 IDs (97.5%) and 193 of 199 (97.0%).
3. Profile and post pages — byAuthor reads the logged-out x.com/<handle> page (exact profile metadata and recent posts) and falls back to the syndication profile timeline. byUrls and conversation read the logged-out post page, which embeds the focal post, quoted post and visible reply previews.
4. Fallback enrichment — FxTwitter fills missing view/bookmark/retweet/quote counters and author profile fields. oEmbed is the last resort for a known post ID. A mirror outage never fails a run; missing fields stay omitted.
5. Client-side filters — X-style operators (mentioning, author, inReplyTo) and every date/language/engagement/media/verification/flag filter are applied locally after hydration. The index never receives opaque operator text.
6. Deterministic output — dedupe by post ID, sort, cap, recompute engagementRate and emit omit-empty rows.

Coverage and honest limits

• Full X search requires a login. X's search page redirects logged-out visitors to a login wall, and the guest SearchTimeline endpoint answers only a minority of requests and virtually never from datacenter IPs. This actor therefore uses the public index as its primary discovery path. Building full search coverage would require a logged-in account pool or a paid provider, which this actor deliberately does not use.
• The index is smaller than X search. A keyword usually yields tens to a few hundred public posts, weighted toward newsworthy and recently linked posts. Broad one-word keywords can hit the roughly 460-candidate discovery ceiling; narrow keywords may yield only a handful.
• Top is an engagement sort. It orders by likes and reposts; it is not X's private ranking model. Top + Latest combines an engagement half with the newest remaining posts.
• Conversation depth is a preview. Logged-out X renders the focal post plus a few visible replies, so a thread returns those posts, not every reply.
• Not covered: geo/place filters, People/Lists tabs, follower/following lists, sentiment/AI scoring and full-text search beyond the index. These need endpoints that are not public without a login.

FAQ

Do I need cookies, a login or an API key? No. The actor is cookieless and keyless. A proxy is optional; the default run works without one and never creates a paid proxy group on its own.

Why do I get fewer posts than the X search box shows? X's full search is login-gated. The actor reads the public Bing News x.com index and public post pages, which cover a real but smaller subset of public posts. That is the documented hard limit, not a bug.

How many posts can one keyword return? The discovery buffer is capped at roughly 460 unique IDs, but the practical yield is usually tens to a few hundred per keyword, depending on how many public posts the index knows. maxItems still caps what is emitted.

How fresh are the results? Discovery favours recently indexed and newsworthy posts, and each post's exact timestamp comes from X's public payload. Use dateFrom/dateTo to keep only the window you need.

Does Top match X's Top tab? No. X does not expose its ranking model publicly, so Top sorts the collected posts by likes and reposts. Latest sorts by the real post timestamp.

Why did a run return nothing? The dataset stays empty and the key-value store RUN_SUMMARY says why: no indexed posts, all candidates deleted/tombstoned, every post removed by filters, or invalid input. The status message shows the same reason. There are no placeholder rows.

Why are view or bookmark counts missing on some rows? Syndication does not expose every counter. The actor asks FxTwitter to fill the gaps, but when the mirror also lacks a value the field is omitted rather than fabricated.

Can I get replies and quotes? Yes. Reply IDs/handles and the parent post are emitted when X exposes them, and quoted posts are nested under quotedTweet with quotedTweetId. conversation mode adds the visible reply previews of a thread.

Is the actor affiliated with X? No. It reads only public, logged-out pages and public mirrors. It does not bypass a login, sell account access, or require your credentials.

Output summary

The default dataset view is called Posts. For every run, the RUN_SUMMARY key in the default key-value store records status (complete, partial or empty), mode, candidate and hydrated counts, emitted rows, dropped-filter counts, index coverage and elapsed time.