Stop hunting for post ideas for good: get dozens of ready-to-go options every morning

Section 01Why engagement on other people's channels is the cheapest source of ideas

Experts struggle with one question for years: what to write about today. They sit and dream up topics out of thin air, guessing at what will land. Some posts go into the void. Some happen to hit, and there's no way to tell why that particular topic worked.

Meanwhile, right next door, in ten of your competitors' channels, the audience has already voted. With reactions, comments, shares. And every day you can see which topic people responded to more strongly than usual. It's a free focus test on exactly the audience you care about – it sits with your niche neighbors too.

How to properly measure "did it land" in Telegram

Telegram has no recommendation feed. A subscriber sees every post from a channel they follow, regardless of how strong the post is. So the view count itself tells you very little. The quality of a post only affects views indirectly: through shares, and through someone showing the post to a friend.

The key metric in Telegram is engagement relative to reach. In plain terms: reactions plus comments plus shares, divided by views. On a channel with 10,000 subscribers, an average post typically pulls 30 reactions and 4 comments on 3,000 views – that's about 1%. If some post pulls 200 reactions, 80 comments and 50 shares on those same 3,000 views, that's 11%, ten times the average. That's the signal: the topic landed.

What monitoring other people's engagement gives you

• Ready-made viral topics. You're not guessing what will land. You see what already landed with your niche neighbors
• A benchmark against similar channels. If a competitor your size got three times more reactions on a post, your delivery is lagging. You see it instantly
• Fresh trends. If one topic surfaced with three different authors in a week, a new demand has entered the niche. You catch it before anyone else
• The audience's voice in the comments. Under a viral post there's always a long thread of real questions and objections. That's material for your next post and for your ads
• Publishing rhythm. You can see what hours your competitors' posts take off. You can tune your own rhythm to match

Section 02What the AI client hunter is built from

Before we dive into the steps – the big picture. The AI client hunter is made of six parts that work together on one small server.

That's ~$25–27 a month total. One post from the digest that lands harder than usual will pay for the subscription six months in advance. And there are 30 digests a month.

How it works in one paragraph

Every morning at 7:30 the hunter wakes up on the server. Through Telethon, under your second account, it goes down the list of competitor channels, pulls every post from the last 24 hours, calculates each one's engagement and compares it to that channel's average. From all the channels it picks the top 5 by engagement. For each it gathers three ideas for adapting it to your blog. The bot sends the digest to your DMs at 8:00, with buttons. Tap a button – Claude writes a post in your style, and the bot sends it to the chat.

Section 03Sign up and pay for Claude

If you already have a working Claude Pro or Max and you've used Claude Code – skip to the next section.

If not – here's a short rundown. The full guide to signing up and paying is in the Threads article – I won't repeat it here.

Section 04A second Telegram account for reading

The hunter needs a separate account. Not the one you message people from, and not the one your channel runs on. It's the same move as a separate card for subscriptions – not critical, but it saves you from sudden losses. A full breakdown of what an expert should generally protect from automated AI scanners is in the guide on protecting yourself from AI hacking.

What to know before signing up

One number, one account. Telegram ties an account firmly to a phone number. Want a second one – you need a second number.

A virtual number works too. Telegram accepts numbers from any normal carrier. Virtual ones from services like 5sim, SMS-activate or OnlineSim are fine. The price is about $0.50–1 for a one-time number, $3–7 for a month's rental.

Don't use disposable numbers. Telegram periodically asks you to revalidate the account. If the number is no longer yours, you lose the account. Rent for 30 days or use a second SIM card.

Section 05Telegram api_id and api_hash keys

For Claude to log into Telegram through Telethon under your second account, you need two keys: api_id and api_hash. Free, on Telegram's official site, 5 minutes.

Section 06A bot to wake you up with ideas

The bot does two things: in the morning it sends the digest with buttons, and on a button tap it generates a post in your style. You make it for free through @BotFather in Telegram, in about a minute.

Section 07Get a VPS

A VPS is a rented computer that runs in a data center 24/7 without you. The heart of the system: while you sleep it reads channels, while you eat breakfast it assembles the digest.

Why a VPS and not a laptop

On a laptop the script only runs while the laptop is open and connected. Close the lid and the hunter stops. If you're doing this for serious work, you need a server.

Plus a VPS sits in a stable data center. Telethon runs cleanly on a proper server, and the connection stays solid – no flaky home Wi-Fi to worry about.

Where to get a server

Any inexpensive VPS works. The easiest to work with is Hetzner Cloud – cheap, reliable, with data centers in the US and EU.

Which server to order

After it's created, save the server's IP address and the root password. You'll need them on the next step, when we move Claude in.

Section 08Install Claude Code on the VPS

This is the best part. Instead of typing commands in a terminal, you'll open Claude Code on your laptop and just tell it in words what needs to be done. It'll connect to the server and install everything itself.

If you don't have Claude Code yet, download it from claude.com/claude-code, install it, and sign in with your Anthropic account.

I have a VPS on Hetzner Cloud:

IP: 185.12.45.67
User: root
Password: MyPassword123

Connect over SSH and prep the server for the AI client hunter.

What to do:
1) Update the system, install python3, pip, screen.
2) Install Claude Code on the server and authorize
   it under my account (I'll enter the code at the link).
3) Install the libraries telethon, python-telegram-bot,
   anthropic.
4) Create a folder /root/hunter and prepare an empty
   config.env file there – I'll put the keys in it.
5) Verify everything works: run a short
   Telethon ping with my api_id and api_hash.

Write these variables into /root/hunter/config.env
(in KEY=VALUE format, one per line, no spaces):

TG_API_ID=12345678
TG_API_HASH=string-from-my.telegram-keys
TG_PHONE=+1XXXXXXXXXX  (the second account's number)
BOT_TOKEN=bot-token-from-BotFather
MY_CHAT_ID=my-id-from-userinfobot

Section 09Where to find competitor channels

The hunter is only as good as the list of channels it watches. Weak channels, weak digest. So build the list with care: a one-time, 30–40 minute effort determines the value of every future morning digest.

Who's a "good competitor"

It's not the one who sells the same thing. It's the one whose audience overlaps with yours but whose product doesn't. That way the digest doesn't turn into spying on an enemy – it becomes a source of complementary ideas.

If you're an expert in systematic client acquisition, your "competitors" in this sense are marketers, content creators, producers, copywriters, sales experts, personal-brand authors. They have your audience too, but they talk to it from different angles – and that's exactly what you need.

Six working ways to build the list

1. Similar channels in a competitor's card

Open any large channel in your niche, tap the name at the top – the description opens. Below the description, Telegram itself shows a "Similar channels" block, a list the algorithm assembles based on audience overlap. It's the highest-quality source: people who read channel A probably read B, C and D from that block. Run through 3–4 large channels in the niche and you've got 15–20 channels to start.

2. TGstat and Telemetr directories

Filter by topic and size. A good range is 10,000 to 100,000 subscribers with an ERR (views per post / subscribers) above 15%. Channels like that have a live audience. Don't grab everyone indiscriminately – check that the channel is authored, not a news aggregator.

3. Folders and curated lists from thought leaders

Many well-known experts publish folders of their favorite channels or post recommendations. It's a ready-made base someone already put taste into. Find 2–3 respected authors in the niche and see who they recommend.

4. Ad marketplaces like Telega.in and similar

Ad exchanges show channels in a niche that have an audience willing to pay for advertising. If a channel is on the platform, it's alive and it converts. Just open the catalog and run through your topic.

5. Mentions in the comments

Under viral posts in large channels, other experts often get mentioned – "so-and-so covers this better," "read such-and-such channel." A natural recommendation network that surfaces strong authors the directories don't show.

6. Your own subscribers

Ask 5–10 people from your audience what other channels in an adjacent topic they read. In half an hour you'll collect 20–30 fresh links no directory will ever show you.

How many channels is optimal

30–60 live channels is the sweet spot. Fewer, and the digest gets thin – there won't be viral posts every day. More, and the hunter's overnight run gets pricier and longer, while the top 5 still boils down to the same leaders.

It's better to have 40 concentrated channels with active authors than 100 full of memes and reposts. Once a month, review the list: which channels give signal and which have died – cut the dead ones, add new ones.

Section 10How Claude captures your channel's voice

This is the key step that separates the hunter from generic AI text generation. Without an understanding of your voice, Claude writes an "average good post" – without you in it. With an understanding of your voice, it writes a post a reader can't tell apart from one you wrote yourself.

The good news: Claude does all the work. No hours spent hand-building style rules – you just hand it a link to your channel.

Here's my Telegram channel: https://t.me/my_channel

Read the last 100 posts through Telethon
(use the second account and the api_id and api_hash
from config.env). Build a sample of my voice from them:

– Do I write with "you" formal or casual, what's the tone,
  is there humor or not
– Which techniques I like: contrasts, personal stories,
  quotes, short punchy sentences
– What my post structure looks like – the hook, paragraph
  length, how I usually end
– Which words and turns of phrase I repeat often
– Which words and turns of phrase never appear with me
  (the slots where other people usually have AI markers)
– Which topics I write about most often

Save the result on the server as voice.md in the
hunter's folder – posts will be generated from it later.
And send the file here in the chat, I'll read it
and tell you where to fix.

Section 11Building the hunter with Claude's hands

At this stage Claude already has: the keys in config.env, the list of competitor channels, and a sample of your voice. All that's left is to ask Claude to build the script itself. It does everything itself – writes the code, installs the dependencies, sets up the schedule. From you, just one thing: give it a detailed brief.

In the /root/hunter folder, build the AI client hunter.

What it should do:

1) Every night at 7:30 ET:
   – through Telethon, under our second TG account,
     visit the channels from my list (you saved it)
   – for each channel, pull the last 30 posts,
     compute average engagement (reactions +
     comments + shares ÷ views)
   – from posts in the last 24 hours, pick the ones
     whose engagement is at least 1.7x higher than
     that channel's average – meaning they took off
   – collect the top 5 across all channels

2) At 8:00 hand the top 5 to you and ask for:
   – for each: 1–2 sentences on what the post is about
   – the engagement numbers and why it took off
   – 3 ideas for adapting it to my blog,
     using voice.md (why this topic is for me,
     what angle to take, how my post differs)

3) The morning bot sends to my DMs:
   – a header "Morning digest for [date]"
   – a card for each post with a description and ideas
   – under each card, 3 inline buttons
     [Idea 1] [Idea 2] [Idea 3]

4) When I tap a button:
   – the bot takes the source post and the chosen idea
   – passes it to you along with voice.md
   – you write a finished post in my voice
   – the bot sends it as clean text

Install everything needed (telethon, python-telegram-bot,
anthropic). Authorize Telethon under the second TG account.
If it asks for an SMS code for the first login, message
me in the chat and I'll forward the code that comes
to the second Telegram. After that Telethon runs on its own.

Set up cron and a background service for the bot so it
runs around the clock. Logs go to /root/hunter/log/.

When it's ready, run a full cycle right now on one
test channel and show me the result.

Section 12How to work with the morning digest

Getting the digest is a third of the job. The second third is filtering it right. The third is not turning the hunter into a copy machine.

15 minutes in the morning

Open the digest over coffee, before the workday starts. Run through the 5 posts with one filter: where's a topic I can say something of my own on, not just repeat someone else's?

• If I have personal experience with the topic – take it. Someone else's post gave the angle, your experience gives the uniqueness. Tap the button, get a draft, polish it to fit you
• If the topic's hot but I have no experience – skip it. Someone else's posts under your name are a fast track to losing the audience's trust
• If the same topic took off with two different authors in one day – that's a trend signal. Worth going in even with basic experience, the demand is real
• If a post is written in a manner opposite to yours – a rare gift. You can write your counter-take: "this topic has another side too," and build a whole separate post on it

Out of 7 morning digests a week, you get 3–5 real posts, written and published. Over a month, 12–20 finished publications, none of which you had to drag out of thin air. If, on top of the idea, you want a full set right away – "post + reel + caption + adaptations for Meta and Threads" – plug in the 10-prompt pipeline for experts.

The main rule – you're the editor, not the observer

The hunter wrote a draft – that's not the final, it's a half-finished part. Read it out loud. Find the spot where you stumble – rewrite it. Add one real example from your own practice. Cut the paragraph that feels unnecessary. After ten cycles like that, the hunter starts writing in your style more and more precisely – you're effectively fine-tuning it with your edits.

Section 13Account safety and risks

Telegram doesn't forbid monitoring public channels – you read exactly what any subscriber sees on their phone. But Telegram does have an anti-spam system that can flag an account that's "too automated." To avoid trouble – three simple rules.

Rule 1: don't scan too fast

If you set it to "pull the entire history of every channel for all time in one pass," Telegram will temporarily limit the account. Telethon paces its own requests. Don't remove those pauses to speed things up – you save nothing and lose the account.

Rule 2: one account, one server

Once you log Telethon in on the VPS, Telegram remembers that session. If you open the same account on three other devices in parallel, Telegram gets suspicious. Logged in on the server – don't use that account from other devices.

Rule 3: for private channels, subscribe from this same account

Telethon reads public channels by link without any subscriptions – that's part of Telegram's open API. Private channels are different: to get in, the second account has to be among the subscribers. So if your list has private invite-only channels, join them ahead of time from the second account specifically.

Section 14What's next: extensions

Once the basic hunter is running and you're used to the morning digests, the extensions suggest themselves. A few directions the people who've used the system for 2–3 months move toward.

Comparison with your own channel

You can bolt on a second module to the hunter that gathers your own posts' engagement in parallel and compares it to the competitors' top posts on the same day. In the morning you see not just "the neighbors had a hit," but "your engagement over the same period was X times lower." It's immediately clear which topics work for you, which don't, and which deliveries you should borrow.

Real-time reaction

The basic hunter makes one digest a day. You can set up a second cycle – an instant one: as soon as a post in any channel hits abnormally high engagement in its first hour, Claude sends a notification right away. The viral idea reaches you not in a day, but in an hour – while the topic's still hot.

A content series from one idea

On top of a finished post, you can plug in a second agent that unfolds a series from one idea: a channel post, a Threads thread, a reel script, an article thesis. In parallel, it's worth periodically running your own archive through a channel breakdown and blind-spot search, so the hunter works not "in general," but against your topic gaps. One morning top – five content units across different platforms.

Several directions in one hunter

If you have several projects, keep several channel lists and several digests. One for the main business, a second for a new niche, a third for experiments. Claude builds each digest toward its own goal and its own voice sample.