On the Telegram map, news stays inside its language. Betting and anime don't
We placed 451,369 public Telegram channels on one map by what they say about themselves. Two news channels in the same spot write in the same language 80% of the time, two betting channels 30%, and 80% of all crypto channels fit into 1.1% of the map
The TGScope map places 451,369 public Telegram channels on one plane by what they say about themselves. Channels whose descriptions and topics read alike end up next to each other, whatever language they write in. That turns the map into a measuring tool: it shows how tightly a subject holds together, and whether channels in different languages that cover the same subject become neighbours.
Some subjects live in one corner of the map
Take 80% of a category's channels and count how much of the map they cover. All channels together need 37.6% of it. Food fits into 0.6%, betting into 0.7% and crypto into 1.1%. At the other end, news needs 6.8%, education 7%, business 7.8% and events 16.3%.
Size alone does not explain it. A random set of 14,767 channels, as many as there are crypto channels, would spread over 35.7% of the map, so real crypto channels are 31.5× more concentrated than chance. Business has a similar number of channels (13,232) and is only 4.5× more concentrated: its channels are scattered over many neighbourhoods of the map.
| Category | Channels | Map area holding 80% | Random set of that size | More concentrated than chance |
|---|---|---|---|---|
| Food | 4,909 | 0.6% | 32.3% | 50.4× |
| Betting | 5,818 | 0.7% | 33.1% | 44.9× |
| Crypto | 14,767 | 1.1% | 35.7% | 31.5× |
| Music | 8,201 | 1.2% | 34.1% | 28.9× |
| Anime | 5,343 | 1.2% | 32.6% | 27.6× |
| Sports | 9,997 | 1.3% | 35% | 27.3× |
| Finance | 9,820 | 1.6% | 34.8% | 22.1× |
| Motivation | 6,975 | 1.8% | 33.8% | 19.1× |
| Religion | 26,507 | 1.9% | 36.4% | 19.5× |
| Beauty | 9,491 | 1.9% | 34.7% | 18.5× |
| Games | 12,198 | 1.9% | 35.2% | 18.3× |
| Marketplace | 15,200 | 2.1% | 35.7% | 17.3× |
| Humor | 10,904 | 2.1% | 35.1% | 17× |
| Movies | 7,176 | 2.1% | 33.8% | 16.3× |
| Health | 9,709 | 2.2% | 34.7% | 16× |
| Art | 7,974 | 2.2% | 34% | 15.4× |
| Adult | 22,860 | 2.6% | 36.4% | 14× |
| Politics | 9,104 | 2.6% | 34.4% | 13.4× |
| Blog | 17,964 | 4.6% | 36% | 7.8× |
| Tech | 15,127 | 4.6% | 35.8% | 7.7× |
| News | 44,595 | 6.8% | 37% | 5.4× |
| Education | 46,638 | 7% | 37% | 5.3× |
| Business | 13,232 | 7.8% | 35.4% | 4.5× |
| Events | 19,705 | 16.3% | 36.1% | 2.2× |
News stays inside its language. Betting and anime do not
Pick two channels that sit in the same spot on the map. 61.9% of the time they write in the same language; for two channels picked anywhere it is 19.6%. Languages form islands, and the regions guide shows which ones.
How strong that pull is depends on the subject. Two news channels in the same spot share a language 80.4% of the time, against 20.9% for two news channels picked at random: news in each language forms its own clusters. Blogs (81.3%) and humour (75.4%) behave the same way. Betting channels in the same spot share a language only 30.1% of the time and anime channels 28.6%, close to the 20.1% expected at random: Russian and English anime channels share the same neighbourhoods. Tech is mixed too (39.3%).
| Category | Neighbours share a language | Any two channels of the category | Main languages |
|---|---|---|---|
| Anime | 28.6% | 20.1% | Russian 34%, English 26%, Persian 7.4% |
| Betting | 30.1% | 11.8% | English 23.5%, Chinese 15.8%, Russian 14% |
| Tech | 39.3% | 21.9% | Russian 31.4%, English 30.4%, Persian 11.4% |
| Sports | 47.9% | 23.5% | Russian 42%, Persian 15.8%, English 15.6% |
| Motivation | 46.4% | 20.1% | Arabic 31.1%, Russian 26.5%, Persian 15.1% |
| Religion | 60.2% | 36.6% | Arabic 58.9%, Persian 8.7%, Russian 6.7% |
| Events | 48.1% | 17.7% | Russian 29.2%, Arabic 22.4%, Uzbek 15.2% |
| Finance | 48.5% | 16.2% | Russian 22.9%, English 22.3%, Persian 22% |
| Crypto | 56% | 27.2% | English 42.3%, Russian 27.6%, Persian 10.9% |
| Games | 60.5% | 32% | Russian 52.9%, English 16.2%, Chinese 7% |
| Health | 59.7% | 27.8% | Russian 48.7%, Persian 13.4%, Arabic 11.1% |
| Movies | 54% | 14.8% | Russian 27.3%, Persian 17.7%, English 14% |
| Art | 72.6% | 48.2% | Russian 67.6%, English 13.1%, Persian 7.7% |
| Food | 73.6% | 49.2% | Russian 69.1%, Persian 7.1%, Uzbek 6% |
| Beauty | 74% | 49.2% | Russian 68.6%, Persian 10.6%, Arabic 8.8% |
| Education | 58.4% | 16.4% | Russian 23.4%, Persian 20.3%, Arabic 19.6% |
| Music | 64.8% | 28.5% | Russian 47.2%, Persian 21.3%, English 10.7% |
| Adult | 64% | 22.9% | Chinese 36.8%, Russian 24.3%, English 16% |
| Marketplace | 61.6% | 17.2% | Russian 30.1%, Persian 21.1%, Arabic 13.9% |
| Politics | 62.2% | 12.3% | Russian 23.3%, Persian 17.4%, English 12.2% |
| Business | 69.5% | 24.1% | Russian 43.3%, Persian 15.4%, English 11.4% |
| Humor | 75.4% | 25.6% | Russian 42.1%, Arabic 22.4%, Persian 15.3% |
| Blog | 81.3% | 39.6% | Russian 60.2%, Arabic 13.9%, Persian 10.6% |
| News | 80.4% | 20.9% | Russian 40.8%, Persian 13.9%, English 8.6% |
What the map can and cannot tell
The map is built from what channels say about themselves: topic tags written by a language model from recent posts, and the channel's own description. Two channels are neighbours when their profiles read alike, not when they share readers or link to each other. A subject that looks international here is one whose channels describe themselves in similar terms across languages. 21.5% of the channels on the map have no clear category and are left out of the tables. How the map is drawn is explained in how the map is built.
Method
Map. The October 10, 2026 version of the map: 451,369 channels, each with a position, a category, a language and a subscriber band. Positions come from multilingual-e5 embeddings of channel tags and descriptions, flattened to two dimensions.
Area. We laid a 64 × 64 grid over the map; 2,031 cells hold at least one channel. A category's area is the smallest number of cells, densest first, that hold 80% of its channels, as a share of occupied cells. The random baseline is the average area of 20 random sets of channels of the same size.
Same language. For each category we take every pair of its channels that share a cell and count how often both write in the same language. The comparison is the chance that two of the category's channels picked anywhere share a language. Language is the one detected in each channel's description. Categories with at least 4,000 channels are listed.
Data and citation
The aggregates behind this article are in one file: telegram-map-topics-languages.json. It holds counts and shares only.
To cite this study: