Telegram Channel Scraper — Posts, Reach & Keyword Search

What it does

Reads the public preview pages Telegram serves to any anonymous visitor, so there is no account to create, no phone number to burn and no session to expire. Full post text, view counts, reactions, links, inline buttons and exact subscriber counts, plus keyword search that runs on Telegram's side rather than on yours.

{
  "channel_username": "durov",
  "telegram_channel_id": "1006503122",
  "message_id": "421",
  "url": "https://t.me/durov/421",
  "datetime": "2026-08-14T09:31:04+00:00",
  "text": "Telegram now supports…",
  "views": 3140000,
  "views_raw": "3.1M",
  "reactions_total": 48213,
  "links": ["https://telegram.org/blog/"],
  "link_domains": ["telegram.org"],
  "mentions": [],
  "tme_handles": [],
  "hashtags": [],
  "emails": [],
  "has_photo": true,
  "has_video": false,
  "forwarded_from": null,
  "is_edited": false,
  "from_cache": false,
  "data_age_hours": 0
}

Why this one

Use cases

• Track what a set of public channels published, with view counts and reactions as the reach signal.
• Find posts mentioning a technology, a company or a product across many channels, without knowing the channel names, using discover mode over the curated catalogue.
• Build a reach report with exact subscriber counts rather than the rounded figures the preview displays.
• Monitor job channels for postings mentioning a stack, and read the contact handles the post carries.
• Measure how a story propagates, using `forwarded_from` and the post datetime.

Input

Every field has a default, and the defaults are deliberately small so a first run is cheap enough to inspect before you commit to a sweep. This table mirrors the Actor's own input schema field for field.

Output and fill rates

A field being in the schema is not the same as it having a value. The percentages below were counted on real runs; the sample sizes are in Measurements. Anything not listed here is not promised.

Every key is always present. A field that exists but is empty comes back as explicit null, so a parser never has to guess.

Datasets

Different record types go to different datasets, so the main table never carries columns that are blank on most rows.

Pricing

Pay per delivered result. Charges are applied as each row is produced rather than in a lump at the end, so an aborted run bills only for what it actually gave you.

Measurements

Each figure is shown with the method that produced it. A benchmark without a method is a marketing claim wearing a number's clothes.

What it will not do

Stated plainly so you can judge fit before spending anything.

• Only public channels. Never groups, never private channels, never invite links — those return a free diagnostic row instead.
• Attachments are not extracted at all: no documents, voice notes, audio, stickers, locations or round videos. After 893 posts across 29 channels there was not one positive case, so the public preview almost certainly does not render them. Fields that would always be `false` were removed rather than shipped.
• Discover mode searches a curated catalogue of verified channels that ships with the Actor. It does not reach channels outside that list, and the run log states how many were searched.
• The catalogue is global and mostly not in English: Telegram's largest job channels are in Arabic, Persian, Chinese and Russian. Searching for a technology name works across languages; searching for an English phrase like `remote` returns less than you would expect.
• Media URLs carry a token and expire. They are a reference, not a permanent link.
• `is_edited` at 5.3% is being re-measured and is currently an underestimate.
• Concurrency is capped at 5 on purpose, which is well inside polite crawling limits rather than as fast as the pages would allow.

Privacy

• The Actor returns the full text of public channel posts, which may include email addresses, phone numbers, @mentions and t.me handles the author wrote. Output may contain personal data and whoever runs it is the data controller for what they do with it.
• Set `includeContactFields` to false and those four columns come back empty. The post text is returned in full either way, because a post with its contacts stripped is still the post.
• It does not cross-reference authors between channels, does not enrich from outside sources, does not build profiles of people, and does not discover channels by following mentions.
• Cached pages are held for 7 days and archives for 90, both expiring automatically. Every row declares `from_cache`, `fetched_at` and `data_age_hours`.
• Data policy and removal requests: telegram.actorstack.dev, or privacy@actorstack.dev

See also the data removal process.

Guides for this Actor

• Scrape Telegram channelsA walkthrough of reading public Telegram channels anonymously: the four modes, when search beats reading a history, and what the anonymous preview does and does not render.
• The Telegram API questionThe Bot API cannot read a channel it is not in. MTProto needs a phone number and an api_id. The public web preview needs neither, and it is what a channel publishes to the open web.
• Search versus crawlReading a channel's history to find 20 posts about one term took 25 requests and 712 KB. Asking Telegram to filter took 1 request and 24 KB. Same 20 results.
• Exact subscriber countsEvery abbreviated figure Telegram displays is a rounded string. One extra 4 KB request returns the exact integer, and both numbers ship so the rounding stays auditable.
• Stemming false positivesTelegram's search stems words rather than matching substrings, so it returns morphological relatives of your term. Filtering them out afterwards is free, and it has to happen before billing.
• The attachment findingDocuments, voice notes, audio, stickers, locations and round videos never appeared once in the anonymous preview. Fields that would always be false were cut rather than shipped.
• Field referenceEvery field the Actor returns, with the rate it was actually filled on across 893 posts in 9 channels — including the two that vary enough by channel type that an average would mislead.
• Channel discoveryDiscover mode searches a curated catalogue of verified public channels by niche. What that catalogue is, what it is not, and why searching for a technology beats searching for an English phrase.
• Channel id versus handleA channel that renames itself breaks every dataset keyed on its handle. The internal id survives the rename, and it is on every row.
• Data protectionA public post can contain an email, a phone number and a handle, and returning it in full is what makes it useful and what makes it personal data. Where that leaves the person running the run.