Telegram narrative intelligence

How narratives form, mutate and travel across language communities on Telegram

Built on a decade of Telegram infrastructure. NarraPrism is the analytical layer above an indexing system that has continuously processed public Telegram channels since 2015.

18M+
Channels and groups indexed
140+
Languages indexed
154
Countries covered
2015
Data coverage since
[ 01 / CLUSTER ]

Narratives, not keywords

Discourse grouped into evaluative claims — "the talks are a path to peace" and "the talks are a surrender" are two separate objects — not topic clusters or keyword counts. A claim that lives across many phrasings is one object.

[ 02 / TRACK ]

Cross-language linking

The same narrative tracked across language communities — one claim, many expressions.

[ 03 / MAP ]

Channels & audiences as separate layers

What channels publish and how their audiences actually react are tracked separately. The gap between them is often the finding.

[ 04 / RESOLVE ]

Communities from repost graph

Communities are detected from who reposts whom, not assigned manually. Surfaces who is amplifying which narrative — empirically, not by guess.

[ REPORTS ]

Published research

Short public cuts from the research corpus, each showing a different region and language community. Every one of them is a surface reading produced without asking the data a hard question. The depth below each is commissioned.

Southeast Asia · Chinese-language layer
The online economy, read from the layer almost nobody reads
Twelve months, 3,187 channels, 6.5M posts. A community watch turning into a marketplace, settlement language surfacing in the shop window, and recruitment traffic up roughly thirtyfold.
Middle East · six language layers
One escalation, six languages, six different events
Eight weeks, 521,000 posts. The layer closest to the event peaked first and fell silent first, and in one language the number of publishers itself dropped by a quarter.
Central Asia & South Caucasus · eight languages
The region is watched for two things. It publishes about neither.
Twelve months, 477,766 posts. Sanctions and labour migration together account for under two percent of what the region publishes about itself. A fifth of it is the price of living.

All three cuts are surface readings; the depth below each is commissioned.

[ WHAT AN ANSWER LOOKS LIKE ]

A public page states a direction. A commissioned cut states an answer. One subject goes in - an actor, a brand, a claim, a corridor, a location, a name. This is the shape of what comes back, with identifiers withheld here because this is a public page.

SUBJECT      [supplied by client]
FIRST SEEN   2026-02-24 03:10 · 2 channels · language of origin
CROSSED INTO +4h   +9h   +31h   +3d   (4 further languages)
FRAME SHIFT  local incident -> attributed action, between hour 6 and hour 14
ORIGINATORS  [withheld] · 3 accounts · 1,140 subsequent carriers
COUNTERED BY 22 channels in the same language, from hour 19
PEAK         412 channels · 8,900 posts · week 3
RECORD       61 of the earliest posts subsequently edited or removed
Channel, account and person identifiers are delivered under agreement, never on a public page.

First results in days rather than months, across any of the languages in the index, answered backwards across years of history as readily as forwards. Send one subject and we will come back with a short read on what the environment holds on it.

Lineage

A decade of Telegram infrastructure

NarraPrism is the third layer in a continuous research stack. The directory and search products generate the indexing pipeline and operational dataset that the analytical layer is built on.

[ DIRECTORY ]

Telegram Groups Directory

Structured public directory organized by topic, country and language. 132,000+ communities published across 271 categories, 72 languages, 154 countries. The original indexing system, continuously maintained.

[ SEARCH ]

Teleteg search engine

Dedicated search across tens of millions of public Telegram communities, with structured filters for activity, language, country, member count. The retrieval layer for researchers, journalists, OSINT practitioners and NGOs.

[ ANALYTICS ]

NarraPrism

Analytical layer for content intelligence and large-scale narrative analysis. Built on the corpus, repost graph and channel metadata accumulated over the previous decade. Current research corpus: 84 languages, 12 months.

Index scale

Top languages in the channel index

The full indexed corpus by language. The research corpus (see Coverage below) is drawn from this base.

Top languages by channel count
Indexed channels · NarraPrism / Teleteg corpus
0700K1.4M2.1M2.8M
Source: channels index, June 2026 · 18M+ channels and groups total.
Audience reach by language
Channel subscribers · top 18 languages
Channel subscribers sum by language · subscriber counts overlap across channels.
Research coverage

What the current research corpus covers

Twelve months of posts, September 2025 – August 2026, from public channels and discussion groups on news, politics, security and society. Every figure below is computed from the archived corpus and can be reproduced from it.

177M+
Posts, Sep 2025 – Aug 2026
107K
Channels and discussion groups
84
Languages detected
12
Months, fully archived

Languages · posts

  • Russian55.4M
  • Persian23.1M
  • Chinese16.9M
  • Arabic15.9M
  • Ukrainian12.0M
  • English6.9M
  • Spanish3.6M
  • Hebrew1.9M
  • Turkish1.7M
  • Italian1.6M
  • German1.5M
  • Portuguese1.2M
  • French1.1M
  • Burmese1.0M
  • Khmer1.0M
  • + 69 more languages 

Regions

  • Eastern Europe
  • Middle East & North Africa
  • Iran & Persian-speaking communities
  • Southeast Asia
  • Central Asia & South Caucasus
  • Western & Central Europe
  • Latin America
  • South Asia
  • East Africa

Content areas

  • News & current affairs
  • Politics & governance
  • Security & conflict
  • Economy & energy
  • Society, protest & rights
  • Crime & fraud
  • Humanitarian issues & emergencies
  • Local communities

Language not determined for about 10% of posts. Volumes reflect current collection and grow as coverage expands.

Beyond the research corpus

The index behind this research holds 18M+ Telegram channels and groups across 140+ languages and grows every day, with 67M+ further channels discovered and queued. Topics range from politics, security and economy to finance and crypto, technology, education, religion, health, commerce, media and local communities, across Eastern Europe, the Middle East and North Africa, Iran and Central Asia, South and Southeast Asia, Europe, Latin America and Africa.

The research corpus above is a small part of this index. Any topic, language or region can be collected and analysed in depth on request.

Research areas · 2025–2026 report series

Every area below is a first cut of the data we already hold. Any of them — or any other topic, region or language — can be expanded substantially: targeted collection of additional channels and groups, longer history, comments and audience reactions, and custom research built around a client's questions.

Flagship · in preparation

One Year, Many Languages

The global Telegram agenda across 84 languages: shared and diverging events, how fast the same event travels between languages, and what draws audience response.

Surface cut of the current corpus · expandable in depth and scope on request
Report · in preparation

The Scam Economy of Southeast Asia

Chinese-language channels on fraud operations in Cambodia, Myanmar, the Philippines and Thailand: schemes, recruitment, exposure and audience reaction.

Surface cut of the current corpus · expandable in depth and scope on request
Report · in preparation

The Cost of Living Across Languages

Prices, currencies, energy and wages as discussed by audiences in Russian, Persian, Turkish, Arabic, Ukrainian and Spanish.

Surface cut of the current corpus · expandable in depth and scope on request
Report · in preparation

Central Asia and the South Caucasus

Agenda, economy, labour migration and infrastructure across the language communities of the region.

Surface cut of the current corpus · expandable in depth and scope on request
Report · in preparation

Persian-Language Telegram Beyond Politics

Economy, water and climate, culture, education and everyday life in the second-largest language community of the corpus.

Surface cut of the current corpus · expandable in depth and scope on request
Report · in preparation

Latin America in Spanish

Agenda, economy and security in Venezuela, Cuba, Mexico, Colombia and Argentina.

Surface cut of the current corpus · expandable in depth and scope on request
Pilot findings

Three patterns the platform has surfaced

From a pilot study (June 2026) on a 3.4M-post corpus covering Trump, Putin and Zelensky across eight language ecosystems. Each finding is grounded in a stated method and explicit limitations.

01

The same "peace" framing, opposite audience responses

3.5× war vs peace engagement in pro-Trump channels · reversed in anti-Trump channels

Posts about Putin, grouped by channel stance toward Trump. In pro-Trump channels, war-framed posts receive 3.5× more likes per 1K views than peace-framed posts, with anger near zero on both. In anti-Trump channels, peace-framed posts draw strong anger (6.56 per 1K views). The same word performs a different function for each audience — measured in reactions, not in what channels publish.

02

Russian-language audiences react more angrily to Putin posts than to Zelensky posts

2.63 / 1.67 anger reactions per 1K views · Putin / Zelensky posts, Russian-language channels

A 57% gap, measured on audience reactions rather than channel output. Among the most-forwarded sources in Russian-language coverage of Putin, pro-government and Ukrainian outlets appear at near-equal volumes. What channels publish and how their audiences respond can point in different directions.

03

Criticism of Zelensky from three unconnected ecosystems

3 independent infrastructures, one narrative direction

US pro-Trump, Russian pro-war and Ukrainian opposition channels all carry critical framing of Zelensky — through different sources, in different languages, for different reasons. Convergence without coordination is harder to detect than a single coordinated campaign, and easy to misread as one.

Analytical corpus by language — Trump posts
2.43M posts · 8-language deep analysis
0200K400K600K800K
Source: ClickHouse trump_posts table · NarraPrism analytical layer, June 2026.
Peace vs war framing
Engagement per 1K views on Putin posts · by channel stance · peace vs war frame
Peace-frame posts War-frame posts
War-framed posts outperform peace-framed posts 3.5× in pro-Trump channels; neutral channels are near parity; anti-Trump channels show the reverse. n = 22,777 posts across 8 languages.
Detected communities

Top 12 of 100+ communities — surfaced from repost graph

Communities are detected from who-reposts-whom on a 6,761-node, 9,696-edge graph. No manual labels in the detection step. Names assigned post-hoc by reading the channels in each cluster and describe only language, region and stance. Ranking reflects this corpus (posts about three leaders), not the relative size of any movement.

Pro-Russian (Russian-language)1,147
Pro-Trump (US)617
Pro-Ukrainian462
German right-wing398
Iranian state media — A338
Iranian state media — B304
Pro-Russian (English-language)265
English-language news aggregators185
Israeli news media154
Pro-Russian (anonymous channels)142
Hamas-aligned (Arabic-language)107
+ 89 smaller communities

Count = channels and source-nodes assigned to the community. Cross-reference: 355 narrative-to-community links computed with weight = share of narrative posts originating in the community.

Flagship report — in preparation

One Year, Many Languages

The global Telegram agenda, September 2025 – August 2026: which events each language community discussed, where agendas converged and diverged, how fast the same event travelled between languages, and what drew audience response. Built on the full research corpus, with every figure reproducible from the archive.

177M+ posts 84 languages 12 months Reproducible query set
Request early copy Sent on publication

Platform capabilities

Channel stance × audience reaction

What a channel publishes and how its audience reacts are two different opinions. NarraPrism separates them by default — and the gap is frequently where the analytically interesting result lives.

Cross-language narrative alignment

A single claim tracked across languages. Convergence between unrelated infrastructures (no shared sources, similar messaging) is detectable because the analysis sits above language.

Communities from repost graph

Communities are detected from amplification structure, not assigned by hand. Surfaces who is reposting whom — and which narrative is being amplified by which empirical cluster of channels.

Narrative-to-community attribution

Each narrative is linked to the communities amplifying it, with weights. Lets the analyst answer: "this claim — which network is carrying it, and how strongly?"

Methodology

NarraPrism ingests content from public Telegram channels. The current research corpus covers twelve months (September 2025 – August 2026) across 84 languages, drawn from a continuously updated index of channels and groups meeting activity and size criteria. The corpus is archived; every published figure can be reproduced from it.

Narratives are extracted as evaluative claims and deduplicated into canonical objects across languages and channel groups. Channel stance and audience reaction are treated as separate analytical layers — the gap between them is, in practice, where the interesting result tends to sit. Audience signal is derived from platform engagement, normalized to within-language baselines, not from generated emotional labels.

Communities are detected from the repost graph via graph-based community detection — no manual labels in the detection step. Each narrative carries weighted attribution to the communities amplifying it.

Every report states confidence bounds and known failure modes. Detailed methodology, validation procedures and reproducible query sets are provided to institutional partners under NDA.

Who this is for

Policy analysts

How audiences in different language communities receive a policy decision — channel framing vs audience reaction, before mainstream coverage catches up.

Think tanks & research orgs

Cross-language narrative briefs on contested topics, with traceable method and reproducible queries — citable in publications.

NGOs & humanitarian organizations

Detection of narrative environments that affect mission areas (migration, conflict, human rights) — empirical, not anecdotal.

Journalists & investigators

Repost-graph traces from a story back to source communities. Convergence patterns between unrelated networks.

Risk & due diligence

Narrative risk around markets, sectors and counterparties across languages — early signals with traceable method.

Media & reputation monitoring

How a brand, organization or topic is framed across language communities, and which networks carry that framing.

About

An international team of Telegram researchers and data engineers — a decade of continuous work in this space.

The team began operating large-scale messaging communities before Telegram existed. When Telegram introduced supergroups and a proper moderation architecture in 2015, that early operational involvement produced the first structured dataset on the platform — maintained and extended ever since.

Over a decade, the collection infrastructure has processed a significant portion of the Telegram ecosystem. The majority is filtered at the collection layer — inactive accounts, short-lived groups, communities below threshold. What is retained is the operational corpus on which NarraPrism's analytical layer runs.

NarraPrism is an independent commercial company. It is not affiliated with, funded by, or acting on behalf of any government, political party, campaign or advocacy movement, and takes no position on the conflicts and actors it measures. One methodology is applied to all sides. NarraPrism does not provide influence, campaign or content-promotion services.

Verification, team credentials and references are provided to institutional partners under NDA on request.

Request access

Tell us about your use case

NarraPrism is currently in private access. We work with organizations that have a defined use case — research institutions, policy and public-sector teams, NGOs, media, and corporate risk and compliance teams. Custom analysis and sample reports are available on request.

Direct contact: research@narraprism.com

Email us