AI content rewriter for Telegram channels
Taking posts from a source channel and publishing them in your own channels in another language is work that has to happen seven days a week. That is exactly the kind of task worth handing to a system.
Straight up: this is a showcase build I made myself, not a client job. I show it so you can see what a system like this looks like from the inside and what stays with a person in it. My client jobs with a verifiable review are here: Luko Tools and a desktop personal finance application.
The problem: for every post, copy, translate, rephrase, find an image and publish. Roughly two hours a day, seven days a week.
The solution: the system picks up a new post in the source channel, rewrites it into the target language in its own voice (not machine translation), generates an illustrative image and publishes to two target channels.
The detail that matters: this is rewriting, not translation. Facts and figures are preserved, but the text reads naturally in the target language — a difference the reader spots immediately.
The situation it addresses
- Watch the source channel and catch a new post.
- Copy the text and translate it.
- Rephrase it so it does not read like machine translation.
- Find or create an image.
- Publish to each target channel separately.
For a single post that is a few minutes. At dozens of posts a day it is a full-time job — and above all a job that cannot be paused, because the channel has to publish continuously, not in batches.
How it works
- Watching the source. The system catches every new post in the source channel.
- AI rewriting. The text is rewritten into the target language keeping the meaning and the facts, but in its own voice.
- Image generation. An illustrative image is produced for the post.
- Publishing. The finished post appears automatically in two target channels.
The whole thing runs continuously, so posts go out at night and at weekends too — which matters for news content, because a story loses value with time.
The details that mattered
Rewriting, not translating
Machine translation is immediately visible in the content. Rewriting that keeps the meaning reads naturally, and it is the main difference from cheaper solutions.
Facts must not change
Numbers, names and dates are carried over from the original post. The model is explicitly forbidden from adding anything that was not in the source.
Automatic images
The channel looks consistent without needing a designer attached to it.
Round-the-clock operation
Publishing happens continuously, not in batches — for news content that is the difference between current and stale.
What stayed with a person
Choosing the source. The system publishes what it is given. The quality of the source channel sets the quality of the output, and that is the operator's decision.
Checking a sample. The model occasionally drifts from the intended voice or phrasing. Without a regular check nobody would notice.
Sensitive and contentious topics. Here a person should decide whether to publish at all.
Responsibility for the content. The channel's operator is responsible for what is published, not the model. Including copyright in any content taken over.
Taking over and processing someone else's content has copyright implications that have to be settled before deployment. Nexflow does not provide legal advice — I handle the technical side, assess the legal side with your lawyer.
The outcome
The stretch from catching a post to publishing in two channels runs unattended. Instead of continuous work through the day, what remains is checking a sample and deciding on sensitive topics.
The “roughly two hours a day” figure is an estimate of how long that work would take by hand — not a measured result at a client. This is a showcase build, so the number gives a sense of scale, it is not a reference.
Frequently asked questions
Is it translation or rewriting?
Rewriting. The text is carried into the target language keeping the meaning and the facts, but phrased afresh so it reads naturally. Machine translation is immediately visible in the content, and that is exactly what this avoids.
Does the AI change facts and figures?
The model is explicitly forbidden from adding information that was not in the source, and both facts and figures are carried over from the original post. Even so I recommend regularly checking a sample — it is part of running the system, not an optional step.
Can it be used for channels other than Telegram?
Yes. The principle is the same for any source and target with an API — social networks, a website or a newsletter. Telegram was chosen here because that is where the audience was.
What about copyright?
Taking over someone else's content has copyright implications that must be settled before deployment and independently of the technology. Nexflow does not provide legal advice — I handle the technical side, assess the legal side with your lawyer.
What does a solution like this cost?
A build of this scope runs in the €700 – 1,200 range depending on the number of sources, targets and the complexity of the processing. On top of that come running costs — the language model and image generation are billed by volume.
Related services
Related articles
Related work
Have a content routine that runs every day?
Tell me what you publish and where the source material comes from. I will tell you what can be handed to a system and what should stay with you.
Get a free consultation →