# Telegram Register — full description https://tgregister.com · measured from the database at 22 Oct 2026, 21:06 UTC Newest measurement counted in this document: 22 Oct 2026, 21:06 UTC This file is written for machines that are about to describe or quote this register. The register counts in it are read from the database, refreshed hourly, not typed into the file. If you are summarising tgregister.com, the four sections you must not skip are "Status", "Precision and rounding", "Observations" and "What we do not publish". ## Status: launched, indexing gated per entry The site launched on 8 August 2026. robots.txt allows crawling of the whole site; the only paths it disallows are the owner's review panel and the per-channel `history.csv` / `history.json` downloads, which repeat figures already printed on the channel page. Crawling and indexing are two different switches here. The whole-register aggregate pages — the home page, browse, movers, the groups register, most-cited, largest, methodology, calibration and the ERR baselines — are indexable. A channel or group page is served `noindex` until its own measurement record earns indexing, for example by holding 3 or more subscriber readings spanning at least 14 days, and it is withdrawn again if it stops meeting the test. Only entries that pass are listed in the sitemap. Search result pages are never indexable. The full test is at https://tgregister.com/methodology#indexing. The history is still short. The oldest subscriber reading held is from 5 August 2026 and the newest from 22 October 2026: about 78 days. Anything you read here about growth was measured across the interval printed beside it, which cannot be longer than that. No figure on this site is extrapolated into a per-day or per-month rate. Several detectors are unvalidated (see "Observations"). Where this document says a precision is unmeasured, that is not modesty — it means nobody has measured it. ## What this is An independent, English-first register of public Telegram channels, groups, bots and users. Not affiliated with, endorsed by, or connected to Telegram. Channels cannot pay to appear, cannot pay to be removed, and cannot pay to change what a page says. ## Surfaces - [How we measure](https://tgregister.com/methodology): the four public surfaces we read, why readings from different surfaces never confirm each other, why view counts above 1,000 are approximate, and the claims we refuse to make. Read this before quoting any number from this site. - [Browse the register](https://tgregister.com/browse): every entry we hold a subscriber measurement for, arranged by size band, by the language of its own posts, and by how recently it last published. A topic is also recorded on some entries, but it is an LLM classification of the channel's own text rather than a measurement, is labelled as such wherever it appears, and covers part of the register, with the share and the model that produced it printed on /browse — there are no topic browse pages yet. - [Risers and fallers](https://tgregister.com/movers): what changed between two readings of the same page, by absolute subscribers and as a share of size. 3,349,168 entries currently qualify. - [The groups register](https://tgregister.com/groups): a second register, covering public Telegram groups and supergroups with an exact member count on each — measured the same way as channels, published separately because groups carry no post archive. - [Engagement baselines](https://tgregister.com/data/err-baselines): published ERR (views-to-subscribers) percentiles by subscriber band and language, so the comparison behind any engagement observation can be reproduced. - [Most-cited entries](https://tgregister.com/cited): ranked by position in the measured citation graph — weighted PageRank over forward and mention edges. Forwarded-by and named-by counts are shown separately and never summed, because a mention is trivially manufacturable and a forward is not. - [The largest entries](https://tgregister.com/largest): top entries by latest measured subscriber count, each with the change since the first reading we hold. The register's oldest reading is from 5 August 2026, so that column covers at most the time since then and is not a growth rate. - [Calibration](https://tgregister.com/calibration): the register's own error bars — how often each estimate and detector here is actually right, including where it is wrong: creation-date estimate coverage, detector precision from hand-checks against live Telegram pages, and how much of what the register holds it refuses to print as exact. - [Search](https://tgregister.com/search?q=): exact-handle and partial-name look-up over the register. - [A channel page](https://tgregister.com/channel/telegram): the unit of the register — subscriber history, engagement, reaction mix, Stars, polls, marked advertising, citations and any recorded observations, each with its own sample size and dates. Also, where held: Telegram's own recommendations (their answer, not ours, dated), near-duplicate content clusters (which channels share text — never who copied whom, a reading measured wrong 18 times in 45), references to handles that are no longer live, posts edited after publishing, links the channel declared about itself, and a Wikidata cross-reference where one exists. The one claim that separates it from a Telegram directory: **everything is measured, and nothing is estimated.** Every number was read off a public Telegram surface by our own crawler on a date we record. Nothing is licensed from an analytics vendor, scraped from another directory, or supplied by a channel owner. Where we have not measured something, the register says so and shows nothing — an empty state is a statement about our coverage, never about the channel. ## The register in numbers All figures below were measured in one pass, at the time printed at the top of this document. | Measure | Count | | --- | --- | | Entries in the register | 4,337,377 | | — of which still answering | 4,194,855 | | — recorded as no longer answering | 142,522 | | Entries carrying at least one measurement | 4,337,377 | | Entries with two or more readings | 4,310,342 | | Entries with two or more profile-page readings | 3,349,168 | | Subscriber/profile readings held | 41,413,864 | | Usernames tracked (including retired aliases) | 4,375,337 | | Handles queued, not yet scanned — NOT coverage | 2,518 | | Posts held | 103,352,780 | | Post view/reaction readings held | 331,435,736 | | — of which arrived rounded by Telegram | 211,608,314 (63.8%) | | Forward edges between registered entries | 2,167,760 | | Mention edges between registered entries | 5,524,517 | | Distinct avatar images recorded against an entry | 6,117,543 | | — archived to object storage | 6,114,012 | | Post deletions recorded | 630,658 | | Post revisions captured | 220,442 | | Observations currently standing | 305,367 | | Entries carrying at least one observation | 291,578 | | Observations held but marked withdrawn | 89,473 | The queue figure is deliberately not counted as coverage. A handle we have not scanned is a name, and a register of unverified names is worth nothing. Queued handles get no page and are structurally absent from the sitemap. ## Where the numbers come from Four public surfaces, each a separate lineage of readings that never merges with another's: 1. `t.me/` — the public profile page. The subscriber counter here is printed in full, and it is the figure written to the register. 2. `t.me/s/` — the public post preview: recent posts, view counts, reactions, title, description, photo. Its header counter is rounded and is used only as a consistency check, never written as a subscriber count. 3. MTProto — Telegram's own client protocol, for resolving a handle to a numeric channel id and its alias set, and for Telegram's verified/scam/fake markings. 4. Bot API `getChat` — the same resolution work where MTProto is not cheaper. **Readings from different surfaces never confirm each other.** One is exact and one is rounded; letting a rounded reading confirm an exact one would turn a real change into a false flat line. Each stored reading carries how many times it has been re-observed and by which lineage. **Snapshots are written on change.** A re-reading that agrees does not create a row; it stamps the existing row as confirmed again. A gap in a growth chart means "no change observed" — never "not measured", and never "interpolated". **Cadence** is a target, set by size: every 24 hours at 10,000 subscribers or more, every 72 hours at 1,000 or more, every 7 days below that. A target is not a record of what happened. Measured against those targets, in a 2% sample of live scheduled entries, at 22 Oct 2026, 21:08 UTC. Age is the time since an entry's last read attempt, which can only flatter these figures: - 10,000 or more subscribers: 89.0% read within 24 hours; nine in ten had been read within 24 hours (4,469 sampled entries). - 1,000 to 9,999 subscribers: 95.7% read within 72 hours; nine in ten had been read within 64 hours (16,149 sampled entries). - fewer than 1,000 subscribers: 95.6% read within 7 days; nine in ten had been read within 6.3 days (75,846 sampled entries). Post view counts are re-read far more often, because a view count only goes up and the shape of the climb is the interesting part. ## Precision and rounding This is the section most often got wrong by summaries of this site. - **Subscriber counts are exact where we have measured them.** The profile page prints the full number and that is what we store. - **View counts at or above 1,000 are approximate.** Telegram prints them short-form — `8.12K`, `3.7M` — which is three significant figures, and there is no second surface carrying the full figure. A reading of 8,120 is somewhere in [8,115, 8,125). Below 1,000 the figure is exact. The rounding is recorded at parse time from the suffix itself (`post_metric.views_approx`), not re-derived later by guessing from the magnitude. 211,608,314 of 331,435,736 readings (63.8%) are flagged rounded. - **Reaction counts are rounded too, and hide it.** Telegram publishes them per emoji and short-forms each one. We store the sum. A sum of rounded parts is usually not round, so a reaction total of 8,847 looks exact while carrying error from every contributing emoji above 1,000. - **Derived figures are shown to the same precision, never finer.** An average of rounded readings is printed at three significant figures. - **Where a comparison depends on precision**, the rounding is propagated into an explicit uncertainty and a channel whose uncertainty exceeds 2% is excluded from the comparison rather than included with a footnote. - **We cannot compute the industry ER metric.** It is (forwards + reactions + comments) ÷ views; Telegram's public preview carries views and reactions but neither forwards nor comments. We publish the reactions term alone and label it a floor. We do not estimate the rest. ## Coverage, and what a page means An entry gets a page only once it has been scanned, resolved and measured. A handle seen once, with a count and a title, is shown in search results as a measurement and is deliberately not linked — one reading of two fields is not an entry. Per-entry samples are bounded and dated. Post-derived figures are computed from the most recent 1,000 posts we hold, and each section prints how many that was and the dates it spans. For the typical entry that is everything we have. **Deliberately absent:** any posting-rate band (we do not hold every post, so "posts per 30 days" would be a property of our crawler, not of the channel); any quality score, rating or traffic light. **A dead handle needs two independent sightings.** Telegram answers a throttled request with HTTP 200 and a "username not found" page byte-identical to a genuine vacancy. So: a channel that answers is alive on sight; a channel that does not has proved nothing. Before a handle is recorded as dead, the exit IP must be proved honest after the fact, two sightings must agree, they must come from different addresses at least an hour apart, and the channel must have no resolving handle left. Even then it is re-probed a month later and resurrects itself if anything answers. 142,522 entries are currently in that state, and the count shown beside them is explicitly the last measurement taken while the handle still resolved. ## Movement, and why so few entries have it `channel_snapshot` is insert-on-change, so an entry with two rows has moved. 4,310,342 entries hold two or more readings, but only 3,349,168 hold two or more of the *same lineage* — the `t.me/` profile page, where the counter is printed in full. A movement is a subtraction, which is stronger than a confirmation, so it inherits the rule that surfaces never mix. The rest are not on the boards, and that is the reason. Two further exclusions, both published as counts rather than applied silently: a movement smaller than the rounding its two endpoints could account for is withheld, and a transition that at least doubles or halves between one reading and two later ones that agree with each other is withheld. Nothing decides which reading was wrong; it declines to subtract them. Every board prints the elapsed interval per row. Nothing anywhere converts a change into a rate. ## Observations (trust signals) **An observation is evidence, not a verdict.** 305,367 observations currently stand against 291,578 of 4,337,377 entries. Every one is recorded at severity zero, which is not "low" — it means the detector that produced it has not had its precision measured against hand-checked cases, and until it has it is not entitled to a grade. That is why nothing is coloured as a warning and why there is no score, rating or traffic light anywhere on this site. **No observation renders without its evidence.** A sentence like "content duplicates 19 channels" with nothing under it is an accusation; the same sentence with the post pairs linked is a measurement a reader can check. Only the second is published, and an observation whose evidence cannot be narrowed renders nothing. **Observations are withdrawn, not archived.** Each detector recomputes its whole result set rather than adding to it, and an observation the current pass no longer supports is cleared and disappears. 89,473 rows are currently held in that withdrawn state. We do not go on publishing a claim we have retracted. Detector keys currently standing, and whether the site publishes them: | Key | Entries | Published on channel pages | | --- | --- | --- | | `avatar_shared` | 226,937 | no — recorded only | | `avatar_impersonation_candidate` | 35,678 | no — recorded only | | `clone_copy` | 14,046 | yes | | `clone_source` | 13,684 | yes | | `clone_mutual` | 11,271 | yes | | `avatar_clone` | 2,312 | no — recorded only | | `err_low` | 576 | yes | | `err_high` | 528 | yes | | `view_reaction_decoupling` | 190 | yes | | `ext_domain_newly_registered` | 88 | no — recorded only | | `ext_domain_blocklisted` | 55 | no — recorded only | | `ext_address_flagged` | 2 | no — recorded only | 40,295 observations across 6 keys, written by three detectors, are published. 265,072 across 6 keys are recorded in the database and render nowhere: no wording has been written for them, and a detector appears on the site when someone words what it means, not when it starts writing rows. Do not describe those keys as findings this site publishes. The published count is rows, not distinct findings: `clone_copy`, `clone_source` and `clone_mutual` are one finding written up to three times, once per direction the corpus supports, and a channel page renders them as one. The three published detectors: - **Duplicate content** (`clone_copy`, `clone_source`, `clone_mutual` — one finding, written once per direction the corpus supports). Post bodies appearing word for word on more than one registered channel, found by comparing text fingerprints and re-scoring the actual bodies. The page lists every channel involved and links the matching post pairs. The *direction* — who published first, and whether a credit was shown — is much weaker than the match, and is presented separately with the measured error rate of that reading beside it: of 45 pairs re-read by hand on the live t.me pages, the text match was wrong 0 times and the "no attribution" reading was wrong 18. Duplicate content has at least three ordinary explanations: a channel mirroring itself, an unattributed copy, and two channels independently reposting the same wire story. The method separates the first two imperfectly and cannot separate the third at all. - **Engagement rate against cohort** (`err_low`, `err_high`). Recorded only when a channel is in the extreme 1% of its cohort *and* at least three times away from that cohort's median, on two independently computed versions of the figure. A low ERR is most often an audience reading inside the Telegram app without opening the channel. - **Views moving without reactions** (`view_reaction_decoupling`). Intervals in which a post's view counter climbed while its reaction counter did not, by more than the channel's own reactions-per-view history and more than Telegram's rounding can account for. 190 exist across the whole register, the detector's precision is unmeasured, and it asserts no cause. **Two words we do not use.** "Scam" and "fake" appear on this site in exactly one place: where Telegram has applied its own marking, in which case the label names Telegram as the source. They are never our characterisation of a channel, and no measurement we take is reported using them. 3 entries carry Telegram's scam marking, 0 its fake marking, and 10,679 its verified badge. ## The deletion and revision ledgers Both only ever grow forward. Neither is a claim about intent, and neither is complete — they hold what we were looking at when it changed. - **Deletions.** 630,658 posts are recorded as deleted, promoted from 641,939 candidates seen missing at least once. A single absence is not a deletion: the same liveness asymmetry applies, because a throttled fetch and a removed post look alike. 1,016,056 reads were refused outright, each one a read in which so many posts we hold went missing at once that the read itself was the thing in doubt — a throttled or truncated page, or a channel gone private, is likelier than a mass deletion. The whole read is refused rather than trimmed to a cap, and the refusals are counted rather than silently dropped. - **Revisions.** 220,442 superseded post bodies across 48,556 entries, each keeping the previous text, its hash, its links and its length beside the new one. It grows only as posts we already hold are edited. ## ERR cohort baselines (open data) Edition 2026-10-22, 30-day rolling window, 40 cohorts. ERR = **mean views per post over the window ÷ subscribers × 100**. That is TGStat's definition, used unchanged so the figures are comparable rather than a house metric, and it is the same figure computed on every channel page. Cohorts are fixed half-decade subscriber bands (each spanning a factor of 3.16, from 1,000 to 1,000,000+), the same bands split by the modal language of the channel's own posts, and population deciles published as description only. Eligibility gates — at least 1,000 subscribers, at least 8 posts in the window, at least 8 carrying a view reading taken 24h or more after posting, at least half the window's posts carrying a reading, and a relative uncertainty on ERR of at most 2% — are why a cohort's population is far smaller than the register. **ERR declines with size, structurally.** In the current edition the median falls from 14.3% in the 1,000–3,162 band to 2.95% in the 1,000,000 and above band — a factor of 4.8, across channels doing nothing differently except being larger. A larger audience is a more passive one, and the denominator grows faster than the numerator. Comparing a small channel's ERR against a large one's therefore measures audience size, not quality. This is why the dataset is cohorted and why a single global "good ERR" figure would be worthless. A low ERR has many innocent causes, including audiences that read in the Telegram app without opening the channel, and channels whose subscriber growth long predates their current output. These baselines describe a distribution. They do not label any channel. Files: - https://tgregister.com/data/err-baselines/err-cohort-baseline.csv - https://tgregister.com/data/err-baselines/err-cohort-baseline.json — carries the definitions, gates and caveats in its `meta` block, so it stays interpretable away from the page. ## Third-party data is quarantined The database holds a small amount of third-party material — other Telegram datasets, domain and address intelligence — in a separate `ext` schema, kept for cross-checking and for discovery. It is never merged into a measurement, never shown as one, and never republished here as ours. Where a third-party source contributes an observation, the observation is one of the keys marked "recorded only" in the table above, which means it does not render on a channel page at all. If you are quoting a figure from tgregister.com, it is one we measured. That is the whole point of the separation. ## What we do not publish - Estimates, models, projections, or any number we did not observe. - Growth rates. The history is 78 days deep; a "+4% per day" extrapolated from a handful of observed hours is a number nobody measured. Every movement is published with the interval it was observed across. - Currency values for Telegram Stars. We publish the count of Stars sent and no money figure: what a Star costs a reader and what it pays a channel are different numbers, Telegram takes a share we cannot observe, and converting one into the other would be an estimate wearing a measurement's clothes. - Per-option poll vote counts. Telegram publishes percentages and a single voter total; multiplying them would manufacture a tally that looks measured. - Sentiment for reactions. Emoji do not carry stable meaning across languages, and sorting them into positive and negative would publish our reading of a culture as a measurement of a channel. - A complete advertising count. We count a post as an ad on an `erid` token (required on paid placements under Russian law since 2022) or a self-declared `#реклама`/`#ad` hashtag. Nothing classifies a post as an ad by reading it, so ad load is a floor and can only ever be one. - Anyone's intent. ## URL structure - `/` — register status and recently measured entries - `/channel/` — an entry; the canonical unit. `/channel/@` redirects here - `/browse`, `/browse/size/`, `/browse/language/`, `/browse/activity/` - `/movers`, `/movers/size/` - `/groups`, `/group/`, `/groups/largest` — a second register, covering public Telegram groups and supergroups, measured the same way and published separately because groups are tracked on a different table with no post archive behind them. Never a Telegram user account: those are retained internally as a by-product of discovery and are deliberately not published anywhere on this site. - `/data/err-baselines` (+ `/err-cohort-baseline.csv`, `/err-cohort-baseline.json`) - `/methodology`, `/methodology#indexing`, `/methodology#disputes` - `/calibration` — the measured accuracy of the register's own methods - `/cited`, `/largest` - `/search?q=` — noindex by design - `/sitemap.xml` → `/sitemap/channels-.xml` (n = 0 to 127; files 0 to 15 hold entries with 5,000 subscribers or more; file sizes vary) ## Licence, attribution and citation The ERR cohort baseline dataset is free to use with attribution to https://tgregister.com. Cite the current edition as: "Telegram channel ERR (views-to-subscribers) cohort baselines, edition 2026-10-22, tgregister.com". Editions are dated and files are regenerated in place, so cite the edition date — the numbers move as the register grows. For any other figure, cite the page and the date you read it, and prefer the measurement date the page prints to the date you fetched it. Every count on this site is a measurement with a timestamp, and quoting one without a date turns it into a claim about the present that it was never making. Please do not present figures from this site as Telegram's own, and do not describe an observation as a finding of fraud. Neither is true, and both are the misreadings this file exists to prevent. ## Corrections and disputes Message @Mytgregister_bot on Telegram with the handle and the specific claim. If the underlying measurement is wrong it is corrected at source and every page derived from it follows within one refresh. If a detector's reading does not hold up, the observation is cleared and stops rendering. An observation cannot be removed by request alone, and there is no price for removing one. Corrections are free and permanent. Full statement: https://tgregister.com/methodology#disputes