Telegram Channel Messages Scraper

Export posts from public Telegram channels — text, date, views, reactions, media links, forwards, replies and link previews — to CSV, Excel or JSON. No Telegram account, no api_id, no phone number, no bot token.

Two things worth knowing before you read further. Telegram runs the keyword search itself, so looking for one brand in a channel with 50 000 posts costs a few requests instead of two and a half thousand — and you are billed for the posts that match, not for the archive. And a repeat run collects only what is new, walking forward from the last post it saw, so a channel you already have is never downloaded or billed twice.

From $0.99 per 1,000 messages, billed once per stored message. A post a filter removed, a post reached twice in one run, and a scheduled run that finds nothing new all cost nothing. Public channels only — groups and private links need an account this actor does not use.

{ "channels": ["durov", "t.me/bloomberg", "@breakingmash"], "maxPosts": 100 }

What one row looks like

Real output, collected 2026-08-18 from a post published the day before. Two edits for width and nothing else: the media URLs are truncated, and the columns that came back empty on this particular post are left out.

{
  "channel": "cointelegraph",
  "messageId": 71658,
  "url": "https://t.me/cointelegraph/71658",
  "date": "2026-08-17T12:08:28+00:00",
  "text": "🚨 JUST IN: Strategy raised $333.7M through stock sales last week but bought no Bitcoin.\n\nNews | Markets | YouTube",
  "textLength": 113,
  "views": 7030,
  "viewsText": "7.03K",
  "viewsAreExact": false,
  "reactionsTotal": 49,
  "reactionTypes": 3,
  "reactions": [
    { "emoji": "❤", "count": 24, "exact": true, "paid": false },
    { "emoji": "🤡", "count": 14, "exact": true, "paid": false },
    { "emoji": "🔥", "count": 11, "exact": true, "paid": false }
  ],
  "isEdited": false,
  "isForwarded": false,
  "isReply": false,
  "hasMedia": true,
  "isAlbum": true,
  "photoCount": 2,
  "photos": [
    "https://cdn5.telesco.pe/file/VYx-WLzNEWcAJ9IVQ…",
    "https://cdn5.telesco.pe/file/HP7Su0DnsROPe5cvP…"
  ],
  "videoCount": 0,
  "documentCount": 0,
  "links": [
    "https://cointelegraph.com/",
    "https://cointelegraph.com/category/markets",
    "https://www.youtube.com/@cointelegraph"
  ],
  "hashtags": [],
  "mentions": [],
  "channelTitle": "Cointelegraph",
  "channelSubscribers": 340875,
  "channelSubscribersText": "340 875 subscribers",
  "channelSubscribersAreExact": true,
  "channelPhotoCount": 19900,
  "channelVideoCount": 1270,
  "channelLinkCount": 57700,
  "scrapedAt": "2026-08-18T03:00:17+00:00"
}

Three columns in that row are worth pointing at, because they are the ones a buyer usually has to find out the hard way.

viewsAreExact. Telegram publishes large counters rounded — 7.03K views, a 55.1K reaction — and there is no unrounded copy anywhere on the page it serves. Counts under a thousand are exact, and this column tells you which of the two you are holding, so an average over a column means something.

isAlbum. Two photos under one post, and Telegram counts them as two messages internally. Rows are still one per post, and you are billed once.

channelSubscribersAreExact. The channel page rounds the subscriber count to 341K. Switch on Add the exact subscriber count and the actor also reads the channel's profile page for the real number — one extra request per channel, never billed.

Fields

Up to 52 columns per row. Empty ones are dropped, so a text-only post has no media columns and a CSV of one channel has no columns of blanks.

Column	What it is
channel, channelUrl	the channel this post is from
messageId, url	the post's number and its permanent link
date	full ISO 8601 timestamp with time zone
text, textLength	the post as plain text, line breaks intact
textHtml	the same with bold, italics and links — optional
views, viewsText, viewsAreExact	the number, the string Telegram printed, and whether it is exact
reactionsTotal, reactionTypes, reactions	totals, plus every emoji with its own count
authorSignature	the author's name, on channels that sign posts
isEdited	the post was edited after publishing
isForwarded, forwardedFrom, forwardedFromUrl	where a forwarded post came from
isReply, replyToUrl, replyToText	the post being answered
hasMedia, isAlbum	what is attached, and whether it is one album
photoCount, photos	direct image URLs
videoCount, videos, videosWithFileUrl	thumbnail for every video, plus a playable file URL for most — 71 of 82 measured carried one, and the count says how many you can actually fetch
documentCount, documents	file title and kind
linkPreviewUrl, linkPreviewSite, linkPreviewTitle, linkPreviewDescription, linkPreviewImage	the article card under a shared link
links, hashtags, mentions	taken from the post's own markup, not guessed from the prose
channelTitle, channelDescription, channelIsVerified, channelAvatar	on every row, at no extra cost
channelSubscribers, channelSubscribersText, channelSubscribersAreExact	subscriber count
channelPhotoCount, channelVideoCount, channelLinkCount, channelFileCount	the channel's own totals
matchedQuery, scrapedAt	which keyword found this post, and when

Five ready-made views — Posts, Engagement, Media links, Links and mentions, Channel details — are on the Output tab, so a CSV of nine columns is one click rather than a spreadsheet edit.

Search inside a channel, and pay only for what matches

searchQuery is handed to Telegram, which filters its own archive server-side and returns only the posts that match — paged, so a common word returns all of its matches rather than the first twenty. Nothing that does not match is downloaded, stored or billed.

{ "channels": ["durov"], "searchQuery": "Apple", "maxPosts": 50 }

Collect only what is new

Switch on onlyNewSinceLastRun and each channel gets its own resume point, stored in your account and carried between runs. The next run reads forwards from the last post it saw, so nothing is skipped even when a channel publishes hundreds of posts between runs — and nothing you already have is downloaded or billed again.

A scheduled run that finds nothing new finishes green and charges you nothing — zero billable rows, because there were no messages to bill for. Measured on two channels: two requests, under three seconds.

For one channel you can name the point yourself: afterPostId: 3215 reads everything published after t.me/<channel>/3215, oldest first.

Filters

Every filter runs before a row is stored or billed, so a filtered-out post costs nothing.

Input	What it does
searchQuery	Telegram filters its own archive by keyword
oldestPostDate	stop as soon as posts are older than this
newestPostDate	skip posts newer than this
onlyWithMedia	photos, video or files only
onlyWithLinks	posts that link somewhere, including shared-article cards
minViews	a floor on the view count
maxPosts, maxPostsPerChannel	the total, and how deep to read one channel

A post that carries no view count at all is dropped by minViews and counted separately in the run summary, because it cannot be shown to clear the bar.

Recipes

• Media monitoring. channels = the outlets you track, searchQuery = your brand, onlyNewSinceLastRun = on, run hourly. You pay for mentions.
• Crypto and signal channels. channels = the rooms you follow, oldestPostDate = this morning, minViews = 1000.
• Archiving a channel before it disappears. maxPostsPerChannel: 0 and let it walk to the beginning; resume a long backfill with afterPostId.
• Content research. One competitor channel, Engagement view, sort by reactionsTotal — which posts actually landed.
• Dataset building. includeTextHtml keeps the formatting; links and hashtags come out as their own columns for labelling.

What it costs

From $0.99 per 1,000 messages, billed once per stored message — and the rate never rises above $2.00 whatever plan you are on. One price covers the whole row: the text, the reactions, the media links, the link preview, the channel details and the keyword search are all included — there is no second event on top.

• Nothing is charged for a post a filter removed.
• One post reached twice in a run — by an ordinary read and by a keyword search — is stored and billed once.
• A run that finds nothing new charges nothing at all.

You are charged for results, not for machine time. Compute, traffic and storage for your run are not added to your bill — under pay-per-event the rate on the card is the whole price.

Apify's free plan includes $5 of credit a month, and that credit is what pays for results here — about 2 500 messages a month at the free-plan rate. There is no separate free trial: you are billed per post from the first post, so start with maxPosts: 20, look at the output, then scale.

Speed: 300 posts in 65 seconds on 256 MB — the smallest size this actor runs in — using 15 requests and 2.1 MB of traffic.

Limits

Stated up front rather than left to be discovered.

• Public channels only. Groups, private channels and invite links (t.me/+…) need a real Telegram account, which this actor does not use.
• A channel owner can switch the web preview off, and a channel that has done so cannot be read. Telegram serves the same page for a switched-off channel, a group and a username that does not exist, so the error says exactly that instead of guessing. Of twenty channel names tried on 2026-08-18, nineteen returned posts.
• Large counters are rounded by Telegram — views and reactions above a thousand. viewsAreExact and the exact flag on each reaction say so per row.
• Custom (premium) emoji reactions carry an ID and no character, because the character is not on the page at all. Standard emoji come through as emoji.
• Comments are not collected. They live in a channel's linked discussion group, which is a separate fetch per post.
• Related-channel discovery is not offered — the page Telegram serves does not contain it. Checked across 356 posts from 19 channels.
• Media are links, not files. Every photo, video and document comes back as a URL you can fetch yourself; nothing is downloaded into storage you pay for.
• Not seen in a 356-post sample: polls, voice notes, audio files, round video messages, stickers, locations and contacts. They are not claimed as supported.

Output

Dataset rows, exportable as CSV, Excel, JSON, XML or RSS, plus a RUN_SUMMARY in the key-value store with what each channel did: posts stored, pages fetched, which names were not readable channels, which filter dropped what, duplicates skipped without being billed, the resume point saved per channel, traffic used, and whether Telegram slowed the run down at any point.

FAQ

Do I need a Telegram account, API key or phone number?

No. This reads the public web preview Telegram itself publishes for channels that have it switched on — the page behind the "Preview channel" button. There is no login, no api_id, no api_hash, no session string and no bot token, and none of those will ever appear in the input.

Will my run wait behind somebody else's?

No. Nothing is shared between users of this actor — no pooled account, no session, no queue. Your run starts when you start it.

Can I scrape a private channel or a group?

No. Those need a real Telegram account. If you paste an invite link the actor says so plainly rather than reporting the channel as missing.

How far back can it go?

To the beginning of what the channel publishes. Set maxPostsPerChannel: 0 and it walks the archive until Telegram stops offering an earlier page.

How much does it cost to export 10 000 posts?

$9.90 at the card rate, or $20 on the free and starter rate — and nothing else, because the compute and traffic the run uses are not added to your bill. It takes about 36 minutes and 70 MB, on the evidence of a 300-post run that took 65 seconds. Use maxPosts to cap a run before you start it.

Why did a channel return fewer posts than I asked for?

Telegram serves a window of about twenty message IDs per request, and an album of nine photos occupies nine IDs while showing as one post. So a channel that posts albums yields fewer posts per request than one that posts text. The actor counts real posts and keeps paging; it never assumes twenty per page.

Why is a view count zero or missing?

Telegram does not publish one for every post. A missing count is left empty rather than written as zero, because zero is a number people sort and filter on.

Can it watch channels on a schedule?

Yes. Switch on onlyNewSinceLastRun and put the actor on an Apify schedule. Each run collects and bills only what appeared since the last one, and a run that finds nothing new finishes successfully.

Other scrapers we maintain

Same engine, same pricing model, same promise: flat rows, stated limits, and no charge for a row that came back empty.

Google Maps Contact Scraper	Local businesses by category and city — phone, website, split address and opening hours.
Local Business Phone Scraper	The same job from directory sources, with Better Business Bureau accreditation on the row.
YouTube Influencer Scraper	Creators by niche, with the website and socials they publish and a subscriber filter.
Threads Hashtag Scraper	Threads posts by hashtag or keyword, with author, engagement and links.
RedNote Xiaohongshu Profile Scraper	Public RedNote creator profiles in bulk — followers, bio, IP location and their notes.
ATS Jobs Scraper	Open roles straight from company career pages on Greenhouse, Lever, Ashby and Workday.