In this article
- webhook The two shapes, in plain terms
- stop_circle The emergency stop, and where it lives
- bug_report The three failures, all of which will happen
- touch_app Try it now — five minutes
- record_voice_over What automation cannot supply
- forward Mirroring another platform, and why it reads wrong
- help Common questions at this stage
- At a glance
Once a bot has posting rights in your channel — the arrangement from Step 081 — anything can publish into it: a blog's feed, a shop's orders, a repository's releases, a spreadsheet, a schedule. This is the least difficult thing in the module to set up and the easiest to regret.
The regret is never technical. It is that an automated channel stops being a channel, and the audience works this out well before the person running it does.
What you will be able to do after this step
- check_circle Tell which of your content is safe to automate and which is not.
- check_circle Set up a feed without waking your subscribers forty times.
- check_circle Know the emergency stop and be able to reach it from your phone.
- check_circle Anticipate the three failures every automated channel eventually has.
What a machine should be allowed to say
The dividing line is not importance. It is whether anything about the message depends on knowing what day it is.
sync Safe to automate
front_hand Never automate
The test that settles it: would you be comfortable if this post went out at three in the morning on the worst possible day? Everything on the left survives that question. Nothing on the right does.
webhook The two shapes, in plain terms
Every automation is one of two arrangements, and the choice matters less than the marketing suggests.
Something asks, repeatedly. A tool checks a feed every few minutes, notices entries it has not seen before, and posts them. It requires nothing at all from the source, which is why it works with any blog on earth, and it is late by however long the gap between checks is. For a blog, a shop or a newsletter, late by ten minutes is not a problem anybody has.
The source tells you. The thing that publishes sends a message to an address of yours the moment something happens. It is immediate, and it requires the source to support it and you to have an address that stays up. This is worth the extra machinery for alerts where lateness is the whole failure — a system going down, a payment arriving — and is over-engineering for a weekly article.
Between the two sits the option most people should take first: a scheduled post written by you (Step 073). It has none of the failure modes below, it is not automation in any meaningful sense, and it solves the actual problem of "I will not be at my desk on Thursday" completely.
stop_circle The emergency stop, and where it lives
Before connecting anything, know how to stop it, because you will need this at an hour when you are not near a computer.
The stop is in the channel, not in the tool. Open the channel's administrator list, find the bot, and remove its rights. That takes seconds on a phone, works regardless of what the automation thinks it is doing, and does not depend on logging into a service whose password is on your laptop. Whatever else you set up, this is the control that always works.
Two things are worth doing on the day you connect it, rather than on the day it misbehaves. Give the bot the narrowest rights that let it work — posting, and nothing else. A bot that can also delete, ban or change the channel's details is one bug away from doing so. And cap the rate at the source if the tool allows it: a limit of a few posts an hour turns a runaway loop into an annoyance rather than an event.
bug_report The three failures, all of which will happen
Not might. Every long-running automation eventually produces all three, and each is survivable if it was expected.
Formatting breaks. Titles and excerpts from real sources contain characters that the message formatter treats as markup — brackets, underscores, asterisks, stray angle brackets from badly cleaned HTML. The result is either a post with visible markup in it or a post that silently fails to send. The robust choice is to send plain text and accept that links will be plain too; the fragile choice is to send formatted text and debug it live. Test with a title containing an apostrophe, brackets and a dash before you trust anything.
Something gets posted twice. Polling tools remember what they have already seen, and that memory is lost when the tool restarts, when the source changes its identifiers, or when somebody edits an old entry and it looks new. Readers forgive one duplicate and leave over six, so the thing to check is not whether it can happen but whether it can happen repeatedly without anybody noticing.
A backlog arrives all at once. The classic first-day disaster: the tool is connected, decides the entire archive is new, and publishes two years of posts in a few minutes. Telegram limits how fast a bot may post, so what your subscribers actually experience is a long, slow avalanche arriving out of order. Every automation should be started against a source that has been trimmed, or started with the history already marked as seen — and started at a time when you are watching.
touch_app Try it now — five minutes
Four steps, in this order, before anything is connected to anything real.
- Practise the emergency stop Open your channel's administrator list and find where a bot's rights are removed. Do it once now so your hands know where it is.
- Set the automation to post silently Before the first real post, not after the first complaint.
- Send one test with an awkward title Apostrophes, brackets, a dash and an ampersand. If it arrives intact, your formatting choice is sound. If it does not, fix it now rather than at publication time.
- Watch the first hour Most automation disasters happen in the first hour and are trivially stoppable if somebody is looking.
record_voice_over What automation cannot supply
There is a reason to be careful here that has nothing to do with rate limits.
People follow a channel rather than the source it draws from because of the person in between — the choosing, the framing, the sentence that says why this one matters. Strip that out and you have built a worse version of the feed reader your subscribers already have, delivered to a place they use for talking to friends. The channel's advantage over the source was never speed. It was judgement.
The arrangement that works is a mixture, and it is not subtle: let the machine carry the routine traffic, and write the things that need a person. One human sentence attached to an automated item — why this release matters, what changed, what to read first — converts a feed into a publication, and it is the difference readers actually notice.
It also solves the problem in Step 075 that automation quietly creates. Automated posts generate comments, and comments expect somebody to be there. A channel that publishes without anybody reading it has a room full of unanswered questions attached to it, which is more damaging to a reputation than posting nothing at all.
forward Mirroring another platform, and why it reads wrong
The most commonly attempted automation is a relay from somewhere else — a social account, a forum, a video channel — and it is also the one that most reliably produces a channel nobody reads. The problem is not the plumbing. It is that the posts were written for a different room.
The furniture does not travel. Handles from another platform point at nothing here. Hashtags that organised something over there are decoration over here. "Link in bio" refers to a bio that does not exist, and a numbered thread arrives as a run of disconnected fragments, sometimes out of order, each one a separate notification.
The register does not travel either, and this is the larger half. A post written for a feed is written to interrupt somebody who is scrolling past: it opens loudly, because it has to earn a stop. A channel post arrives as a notification inside a messaging application, where it has already interrupted — it did not need to earn anything. Relayed content therefore reads as shouting, and the reader registers that long before they can say why.
Two arrangements avoid it. Relay the fact and write the sentence: let the automation carry the link and the title, and add a line of your own about why it is here. Or put the mirror in its own channel, labelled plainly as a mirror, so that nobody arrives expecting a publication and finds a pipe. What does not work is the middle version everybody tries first, where the main channel is half relay and half writing and the audience cannot tell which is which.
help Common questions at this stage
Can a bot post into my channel without being an administrator?
No. A bot cannot join a channel the way a person does; it has to be added as an administrator by somebody who already is one, and it needs the right to post messages. That is a genuine security property rather than an inconvenience: nothing can publish into your channel unless a human with authority put it there, so the administrator list is a complete inventory of what can speak as you.
Why do my automated posts sometimes fail with a formatting error?
Because the text arriving from your source contains characters the formatter treats as markup and the pairs do not match — a single underscore in a filename, an unclosed tag left by a badly cleaned excerpt. The formatter refuses the whole message rather than guessing. Sending plain text removes the entire category of problem, at the cost of plain links, and that is usually the right trade for automated traffic.
Can I edit or delete something an automation posted?
You can delete it as an administrator. Editing is different: only the bot that sent a message can edit it, as Step 081 explains, so correcting an automated post by hand is not possible from the app. In practice this means the correction for a bad automated post is deletion followed by a post you write yourself, which is also the version readers prefer.
My feed dumped fifty items into the channel. What now?
Remove the bot's rights first so it stops, then delete, then say something. The deletion is tedious and worth doing; the sentence afterwards matters more than people think, because subscribers who were woken fifty times will decide within the day whether to mute you permanently. A short, plain acknowledgement almost always retains them. Silence almost always does not.
Does automated posting affect how many people see my posts?
Not directly — nothing treats a bot's post differently from yours. The effect is indirect and larger: more posts means more chances to be muted, and muted subscribers keep counting while reading nothing. The measurement to watch is the one from Step 072, the proportion of your audience that has notifications on. It is the only number that tells you whether an automation is being tolerated or merely not complained about.
Should I use a hosted tool or write my own?
Start with a hosted one. The thing you are building is not interesting, and the cost of writing it yourself is not the writing but the years of keeping it running. What matters more than the choice is the credential: whatever you use holds a token that can publish as your channel, so treat that the way Step 086 treats any bot token, and remember that removing the bot's rights in the channel revokes access from every tool at once regardless of who holds what.
At a glance
- 1What to automate
Posts that read fine at 3 a.m. on a bad day; keep judgement calls manual.
- 2Post quietly
Automated feeds should send silently and in batches.
- 3The emergency stop
Remove the bot's posting right from the channel's admin list.
Updated
