We publish a directory of public Telegram channels and re-measure it every week. This week's export covers 5,205 channels with 642,717,136 subscriptions between them, measured on 14 Sept 2026.
Here is what the distribution looks like, because the shape is not what most people assume and nobody seems to publish it.
The top is very small and very heavy
The largest one per cent of the index is 52 channels. They hold 38.5% of all the subscriptions we measure.
Widen it to the top ten per cent, 520 channels, and the share is 79.4%. Most of the audience sits in a tenth of the list.
The median channel, meanwhile, has 15,988 subscribers. The mean is 123,481, 7.7 times the median. When a mean runs that far ahead of a median, every average you read about "Telegram channels" is describing a handful of giants and nothing else.
The tail is most of the list
37.9% of the channels we index have fewer than 10,000 subscribers. That is 1,974 channels, and they are the ones a media buyer never sees on a leaderboard.
Only 128 channels, 2.5%, have passed a million. Verification is rarer than the badge count suggests too: 567 channels, 10.9%, carry Telegram's verified flag.
The posting frequency nobody budgets for
The median channel posts 1.89 times a day, across 5,024 channels with a measured history. That is the number to size a reader or a digest around.
But 148 channels post at least 20 times a day, 2.8% of the index, and that is the cohort that quietly breaks a naive pipeline: a poller sized for two posts a day falls a day behind in its first week and stays there. Those rows carry posts_per_day_is_lower_bound=1 in the export, because they filled the whole history window inside a single day and their real rate is higher than the number.
Recommendations are sparser than the network diagrams imply
Telegram will tell you which channels it considers similar to a given one, and we store that graph. The median channel is recommended by exactly 1 other channel in our index. The most recommended one is pointed at by 47.
Only 306 channels, 5.9%, are recommended by nobody at all. So the graph is connected almost everywhere and thin almost everywhere, which is awkward if you were planning to walk it for discovery: you run out of edges before you run out of interest.
What the topics look like
20 topics cover the index. The five largest: regional-ru with 609 channels, education 548, science 386, news 355, crypto 310.
That ranking says more about what gets indexed and recommended than about what exists on Telegram, and we would rather say so than let you read it as a map of the platform.
How this was measured
One export, /data/channels.csv, week 2026-W38, measured 14 Sept 2026: 5,205 rows, one per channel, keyed by username. The file is public and every number on this page recomputes from it.
The cohorts round down. Sort by subscribers descending and take the first floor(0.01 × 5,205) = 52 rows for the top one per cent. That cohort is 1.00% of the index, never wider than the share it claims. The same rule gives 520 rows for the top ten per cent. Round up instead and you get one extra channel in each, which is where a recomputed figure usually parts company with a published one.
The set is the same one the rest of this site is built from: public channels of 1,000 subscribers and up, minus anything Telegram flags as scam or fake. One filter feeds the pages, the export and this page, so the three never disagree.
Growth is published with the interval it was measured over. The two readings behind it are 31 Aug 2026 and 14 Sept 2026, with no reading in between, so the interval is 14 days rather than seven. The column keeps the name weekly_growth_pct, because a column name is a contract with code that already reads it, and growth_days sits beside it in the export so the interval travels with the number. Read the interval, not the column name.
The data behind this page is on the downloads page, and how it is measured in full. Reading the same fields live, per channel, is one call: the free plan is 2,500 calls a month with no card, and the endpoint reference is the whole contract.
First published on tgatlas.org.
