Threads Search Scraper: Keywords, Hashtags, Dates

Threads Search Scraper: Keywords, Hashtags, Dates

Scrape Threads search results (threads.com, Meta's text app) without a login: keyword and tag searches, the top posts or the newest posts, or one page per date window down to one hour for depth. One row per post with the text, the author, the time, likes, replies, reposts, quotes, media, language, the post's topic tag and whether its text contains your search words. Pay per post, from $0.45 per 1,000.

Built for social listening and brand monitoring, campaign and hashtag tracking, research and AI data sets, and scheduled keyword monitors: name a memory and every later run delivers and charges only the posts it has not delivered before.

Quick start

1. Open the Actor and type one or more words into Search words (coffee, running shoes, ai agents), or topic tags into Tags (#sourdough).
2. Pick the Search depth: the top posts page (the default), the recent page, both, or Date windows with a Window length of 24, 12, 6, 3 or 1 hour and a date range (the last 7 days when you give none).
3. Set the caps: posts per search (100 when left out; the Console form starts at 20) and posts per run (default 1,000).
4. Run. Rows land in the dataset as JSON, CSV, Excel or through the API; a free coverage row per search (and per window) says what was read, what passed your filters and why it stopped.
5. For a monitor: give a Memory name and schedule the run; each post comes once.

What you get

One post row per Threads post a search found, charged once per run. Fill rates measured on 1,181 distinct posts delivered by 26 runs of the real actor's test matrix of 2026-10-07 (build 0.1.1; the later audit fixes changed no row field) (1,092 from keyword searches with the search words filter on or off, 89 from tag searches; top and recent pages, hour and day windows; platform-results/fill-rates-r1.json):

Field	What it is	Filled
postId, code, url	the numeric id, the short code and the post's page	100%
text	the full text of the post	99.9% (a media only post has none; with the search words filter on, every keyword row has text)
keywordInText	keyword searches: true when the post's own text contains every search word, false when Threads answered it as a related post without them; null on tag searches	100% of keyword rows
postedAt, postedAtUnix	when it was published (UTC)	100%
authorUsername, authorId, authorVerified, authorProfilePic, authorUrl	the author	100%
authorFullName	the author's display name	92.2%
likeCount, replyCount, repostCount, quoteCount	engagement at the time of reading	99.9%
language, languageTag, languageSource, languageUnknown	the language as an ISO code: Threads' own label when it has one (Chinese, Japanese and Korean posts), else a statistical classification of texts of 80 characters or more (languageSource: "text"); unknown and kept when neither gives one	63.5% (6.2% Threads' own label, 57.3% classified from the text); the other 36.5% are short texts, kept with languageUnknown: true
topicTag	the post's own Threads topic tag (one per post at most)	24.0%
mediaType, hasMedia, imageUrl, images, videos, carousel	text, image, video or carousel, with the media URLs (they expire after some days)	57.8% carry media: an image on 57.6%, a carousel on 31.8%, a video on 9.7%
linkAttachment	the link card (url, title, image)	1.7%
isReply, replyToUsername, replyToUserId, rootPostAuthorUsername, parentPostCode	reply context (search pages carry replies too)	29.6% of rows are replies
isQuote, quotedPost, repostedPost	the quoted or reposted post (code, url, author, text, time)	when present
source, sourceQuery, sourceWindow	search or tag, the word or tag that found the post, and top, recent or the UTC window (start/end)	100%
viewCount	always null here: Threads shows no view count in search answers without a login	0%

Free rows next to the posts: one coverage row per search (pages read, posts read, filtered with the reason, repeated, known to the memory, delivered, charged, the stop reason, allPagesRead, the windows read, failed and not read; a search is never reported complete: complete is always false), one coverage row per date window read (window, status, postsRead, postsMatched, postsDelivered, pageCharged), a status row for an unusable value or for searches a stopped run did not start, and one summary row with every counter and what was charged.

How deep it goes

Threads answers one page per search to a visitor without a login (about 20 to 30 posts on the top page, 10 to 20 on the recent page); in our 252 measured calls it never answered a second page. Depth comes from date windows: the same search limited to a window answers a fresh page for that window, and Threads holds the window's end to the second. Measured on 2026-10-07 for one full day of each word:

Word	One day window	Four 6 hour windows	24 one hour windows
nike	19 posts	77 distinct	501 distinct
sourdough	21 posts	96 distinct	563 distinct
apify (a rare word)	5 posts	4 distinct	3 posts in two of the hours

Windows shorter than one hour are not honoured by Threads, so one hour is the finest window offered. The window door reaches back at least 700 days. A window page is a sample: the same window read twice shared about 15 percent of its posts, so no search here is complete and none is reported as complete (complete is always false; allPagesRead says whether every planned page or window was read); the coverage rows say exactly what was read.

Window pages also carry related posts: on the measured windows 62 to 66 percent of the posts did not contain the search word (shoes and running for "nike", bread for "sourdough"), against 23 to 67 percent on the top pages of eight words. That is why the search words filter is on by default: a post whose own text does not contain every search word (at the start of a word, any case, any order: "nike" matches Nike, #nike and Nikes) is left out and not charged. Switch it off to keep every post Threads answers; each row still says keywordInText true or false. The rule is plain on purpose: lower case only, no accent folding (cafe does not match café), no stemming; an emoji only search word matches nothing, so switch the filter off for one.

How much does it cost to scrape Threads search results?

Pay per event. Prices per 1,000 events (USD):

Event	Free plan	Bronze	Silver	Gold
Post (one row)	0.49	0.48	0.47	0.45
Search page (one page with at least one post that passed your filters)	2.00	2.00	2.00	2.00
Actor start (per run)	0.006 per run, no discount	0.006	0.006	0.006

Platinum and Diamond plans pay the Gold price. The Store pricing card shows these same prices per 1,000 events: "$0.49 / 1,000" on the Post row means one post costs $0.00049, and "$2.00 / 1,000" on the Search page row means one search page costs $0.002, a fifth of a cent.

Free: posts without your search words (with the filter on), posts outside your dates or languages, posts already delivered in the run or known to your memory, an empty page, a page whose posts all failed your filters, a word Threads has no posts for, refused calls, and every coverage, status and summary row.

Worked examples on the Free plan:

• One word, the top page, 20 posts delivered: 0.006 + 0.002 + 20 x 0.00049 = $0.0178.
• One busy word, one day in 1 hour windows, about 24 pages and 200 posts delivered: 0.006 + 0.048 + 0.098 = about $0.15.
• A daily monitor of 5 words under one memory that finds 10 new posts in total: 0.006 + 5 x 0.002 + 10 x 0.00049 = about $0.021.

How much one run reads: Threads lets one exit make about 40 to 70 search calls before it asks for a pause of 30 minutes and more; a run moves to a fresh exit up to 4 times, so one run reads at most about 200 to 300 search pages and then stops, naming every window it did not read. Split larger jobs (100 words over many hour windows) over several runs.

Work that earns nothing is bounded: a run may work 40 seconds plus 6 seconds per post and 8 seconds per search page it charged; past that, searches not yet read are skipped free with a status row. A search with no posts, a list of words Threads hides, or windows whose posts all fail your filters therefore cost you only the start fee.

Input

Field	What it does	Default
searchQueries	words or phrases, one per line, up to 100; a pasted Threads search URL (threads.com/search?q=...) is read as its word, or as a tag for a tag search; a profile or post address, and a mention (@name: Threads' search without a login answers none, so search the name without the @), are refused free	prefill coffee
tags	topic tags, with or without #, up to 100	none
searchMode	top, recent, top_and_recent or windows	top
windowHours	for windows: 24, 12, 6, 3 or 1 hours per window, on a UTC grid from midnight; at most 744 windows per search	24
postedAfter, postedBefore	YYYY-MM-DD, YYYY-MM-DDTHH:MM (UTC), today, yesterday, "7 days", "2 weeks", "1 month" (whole UTC days counted back from the end day, so "7 days" with no end reads today and the seven days before it), "6 hours" (counted back from the run's start)	for windows: today and the six days before it; no limit otherwise
requireKeywordInText	keep only posts whose text contains every search word (keyword searches)	on
languages, dropUnknownLanguage	ISO codes to keep (en, es, ja); unknown kept unless dropped	none, off
minLikes, excludeReplies, onlyWithMedia	engagement and type filters	0, off, off
maxPostsPerQuery, maxPosts	caps per search and per run	100 (the Console form starts at 20), 1,000
memoryName, resetMemory	pay once per post across runs	none
proxyConfiguration	Apify datacenter proxy (one exit at a time; a run moves to a fresh exit when Threads asks it to wait, up to 4 times)	datacenter

Alias keys are read too: queries, keywords, searchTerms, hashtags, maxItems, maxResults, startDate, endDate, strictKeywordMatch.

API examples

curl -X POST "https://api.apify.com/v2/acts/automation_craft~threads-search-scraper/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"searchQueries":["coffee"],"searchMode":"windows","windowHours":"1","postedAfter":"12 hours","maxPostsPerQuery":500}'

import { ApifyClient } from 'apify-client';
const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('automation_craft/threads-search-scraper').call({ tags: ['sourdough'], searchMode: 'windows', postedAfter: '3 days' });
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items.filter((x) => x.type === 'post').length);

from apify_client import ApifyClient
client = ApifyClient("<APIFY_TOKEN>")
run = client.actor("automation_craft/threads-search-scraper").call(run_input={"searchQueries": ["ai agents"], "languages": ["en"]})
posts = [x for x in client.dataset(run["defaultDatasetId"]).iterate_items() if x["type"] == "post"]

What this Actor does NOT do

• It does not log in and uses no account, no cookies of yours, no account pool and no third party data service: it reads the search threads.com shows a visitor, keeping only the anonymous cookies the site sets during a run.
• It does not page a search past the one page Threads answered in our measurements; depth comes from the recent page and from date windows, and no search or window is reported as complete.
• It does not filter by author: Threads answers its author filter for some accounts only (several large accounts gave no post for any word in our tests), so it is not offered.
• It does not deliver view counts: search answers carry none without a login.
• It does not search mentions (@name): Threads' search without a login answers none (0 posts for "@nike" where "nike" read 22 in the same run); a mention is refused free and the name without the @ finds the posts that carry it.
• It does not filter by a minimum number of replies (minimum likes, media and replies on or off are offered).
• It does not search for profiles or accounts by keyword, and does not read profiles or the replies under a post: a username or a post URL in the input makes the run stop at once, free, with a row that says so.
• It does not expand your words into plurals or variants: list each wording you want as its own search word (the search words filter already matches a word at the start of a longer one).
• It does not label short texts: Threads' own label is used when present, texts of 80 characters or more are classified statistically (the source is on every row), and the rest is marked unknown and kept unless you ask to drop it.
• It does not use residential proxies by default.

FAQ

How do I search Threads posts by keyword?

Put the words into searchQueries, one search per line. Each search reads the top posts page Threads shows for it (or the recent page, or one page per date window), keeps the posts whose text contains every word (switch requireKeywordInText off to keep all) and gives one row per post with a coverage row per search.

Can I search Threads posts between two dates?

Yes. Give postedAfter and postedBefore (a day, a day and time in UTC, or "7 days", "6 hours") and choose searchMode: "windows": the Actor reads one page per window of windowHours across the range, newest first, and filters every row to the range again (Threads' window start is loose). Shorter windows find more posts; one hour is the finest Threads honours.

Can I scrape Threads posts by hashtag or topic tag?

Yes. Put the tags into tags (with or without #). Tags are read on Threads' tag surface with the same modes, date windows included. Tag pages are not filtered by the search words; each row carries the post's own topicTag and the tag you searched in sourceQuery.

Why are some posts missing my keyword, and how do I keep them?

Threads answers related posts too: on date window pages most posts in our measurements did not contain the word. With requireKeywordInText on (the default) those posts are left out and not charged; switch it off to keep them, and use the keywordInText column to tell them apart.

Why do some rows have no language?

In our measurements Threads labelled only Chinese, Japanese and Korean posts (on 0 of 104 posts of five English top pages). Other posts are classified from their text when it is at least 80 characters long (languageSource: "text", a statistical classification); shorter texts are languageUnknown: true and kept, unless dropUnknownLanguage is on. With a languages list the run's summary row counts the kept and the dropped unknown posts; every row carries languageUnknown.

Why does this Actor run with limited permissions?

Least privilege: it only needs its own storages and the named memory store it creates itself, so it asks for nothing else in your account.

More data tools by Automation Craft

• Meta Ad Library Scraper - All Placements, Filters
• Facebook Ads Library Scraper - Page Ads, No Login
• Instagram Ads Library Scraper - Creatives, Video
• TikTok Profile Scraper: Exact Stats, No Login
• Google Trends Scraper - Compare and Trending Now
• Google News Scraper: Search, Topics, Decoded URLs

Changelog

• 0.1 (2026-10-07): first release.