OBSERVED · CABINET CAPABILITIES
The documentation tells you what is permitted. This page tells you what is used: we parsed the CTA link of every ad in the archive and counted which link types actually occur in the TON and EUR cabinets. That gives a direct answer to the question asked most often about bots — where the start parameter works.
It is not exclusive to one rail: deep links of the t.me/username?start=... shape come out of both the TON and the EUR cabinet, and in both they almost always point at a bot.
The cabinets diverge not on start but on whether the button may leave Telegram at all. In the EUR cabinet external links are common practice; in the TON cabinet we practically never observe them.
Breakdown by CTA link type. Shares are computed against each cabinet’s own creative total, not against the whole archive.
| Link type | TON Ads | EUR Ads Cabinet |
|---|---|---|
| t.me, no parametera channel, bot or chat with no data in the link | 363,70791.7% | 66,46665.9% |
| ?start=deep link: the bot receives the parameter with /start | 29,1037.3% | 11,91911.8% |
| startappstraight into the mini-app, skipping the bot chat | 3,7960.96% |
The start parameter is barely used for channels — it is a bot-traffic instrument. We count a link as bot-bound when the target username ends in "bot": Telegram enforces that suffix on bot usernames, so the signal is reliable in the common case, but a bot whose name does not end in "bot" is missed. Read the share below as a floor.
TON Ads
97.3%
of start links point at a bot
28,313 / 29,103
EUR Ads Cabinet
96.6%
of start links point at a bot
11,518 / 11,919
start opens a chat with the bot and hands the parameter to it: the bot receives it with the /start command and can route the user to the right offer or credit the traffic source. startapp opens the mini-app immediately and passes the parameter into the web app. Both occur in both cabinets, but start is markedly more common — advertisers want the bot step, where they can measure and retain.
This is the largest capability difference, and it has nothing to do with start. In the EUR cabinet, sending traffic to an external site is a working, high-volume scenario. In the TON cabinet, external destinations are one-offs in our observations.
TON Ads
0.002%
leaves Telegram
6 / 396,612
EUR Ads Cabinet
21.0%
leaves Telegram
21,220 / 100,876
External links in TON are a statistical zero: 6 of them, 0.002% of the cabinet. We went through them one by one. Every single one is dated to the first two days of our TON observation window, and some carry markers typical of the EUR cabinet in their URLs (tgclid, utm_medium=cpm). That looks more like a cabinet-attribution error at the edge of the window than a working scenario. In practice it means we have not seen a single ordinary external destination in TON across the whole observation period — but "have not seen" and "forbidden by the platform" are different claims, and we are not making the second one.
These are observations of real ads, not an extract from the documentation. A link type missing from the sample means "we have not seen it", not "the platform forbids it" — we cannot make the stronger claim and we do not.
Observation depth differs by cabinet, and it is the depth of OUR observation, not the launch dates of the cabinets. A shorter window on one cabinet does not make it "newer" — it means we started recording it later. Month-over-month comparisons are only valid inside the overlapping period.
Every number on this page is a live aggregate from the database as of the stated date. We do not round for effect and we do not carry figures over from press releases.
Only ads that have a CTA link and a determined cabinet enter the breakdown. Hidden creatives and the small tail the classifier assigned to neither TON nor EUR are excluded: that tail has no CTA link at all, so there is nothing to parse, and giving it a row in a "what each cabinet supports" table would make it look like a third cabinet that does not exist.
Every creative behind this table sits in the open archive — filterable by cabinet, niche, country and advertiser.
Yes. Deep links of the t.me/username?start=... shape appear in creatives from both cabinets — TON and EUR alike — and in both they almost always point at a bot. The difference between the cabinets is not whether start is supported; it is where the button is allowed to lead at all. Live counts are in the table above.
In the EUR cabinet external links are ordinary practice — a large share of the creatives we observe leave Telegram, most often through click trackers and affiliate redirects. In the TON cabinet we practically never see them: every one of the handful of exceptions falls in the first days of our TON observation window, and some carry EUR-cabinet markers in their URLs, which reads as a cabinet-attribution error rather than a working scenario. "We have not seen it" is not the same as "the platform forbids it", and we only claim the first. Exact shares are in the table above.
start opens a chat with the bot and hands the parameter to it: the bot receives it with the /start command and can route the user to the right offer or credit the traffic source. startapp opens the mini-app straight away, skipping the bot conversation, and the parameter arrives inside the web app. Both appear in creatives from both cabinets, but start is markedly more common.
G.Media Agency runs campaigns in both cabinets and knows which bot-bound or site-bound scenario clears moderation where. A 30-day pilot with a transparent dashboard.
| 1,2711.3% |
| external URLleaves Telegram | 60.002% | 21,22021.0% |
|---|
| invite link t.me/+a cut inside t.me links, not a fifth category | 28,0387.1% | 8,6658.6% |
|---|
The first four rows are disjoint and add up to the whole cabinet: every ad falls into exactly one. start= and startapp are only counted on t.me links — an external redirector carrying start= in its query counts as an external link, otherwise the buckets would stop adding up. The invite row is a slice of the parameterless t.me links and must not be added to the rest.
By the target username: if it ends in "bot", we count the link as bot-bound. That is the same convention Telegram itself enforces for bot usernames, so it is reliable in the common case — but it is a heuristic, not a lookup. A bot whose username does not end in "bot" is undercounted, so treat the bot share as a floor rather than an exact figure.
Because this is not documentation. We do not restate the platform rules — we count what actually ran: every creative that entered our archive is broken down by the type of its CTA link. Documentation tells you what is permitted; these numbers tell you what is used. If a link type is absent from the sample, that means "we have not observed it", not "the platform forbids it".
The as-of date sits in the page header, next to the totals. Observation depth differs by cabinet: we have been recording the EUR cabinet substantially longer than TON. That is the depth of our observation, not the launch dates of the cabinets — do not read the shorter TON window as "TON is newer". It also means month-over-month trends are only comparable inside the overlapping period.