How Do You Filter Telegram Messages by Describing What You Want in Plain Language?

Keyword lists work when the thing you want to catch has words. They fail when it does not: "anything that is really an advert", "posts that are not about this market", "messages that are just noise". Those are judgements, and a criterion written in ordinary language handles them.

How it works from where you sit

You write the criterion as a sentence โ€” what should pass, or what should be dropped โ€” and each message is judged against it before it reaches the destination. There is no syntax to learn and no pattern to debug.

It runs alongside the ordinary filters, not instead of them. The cheap checks come first, so a message already dropped by a keyword or a length rule never reaches the AI, and never costs anything.

Writing a criterion that works

  • Be concrete about the subject: "about EURUSD or gold" beats "about markets".
  • Say what to keep rather than everything to drop, when the thing to keep is narrower. It is easier to describe.
  • One criterion per rule. Two unrelated conditions in one sentence make the judgement worse, and two rules cost the same.
  • Test it against real messages from the source before leaving it running. It is a judgement, and judgements can be wrong.

When it is unsure, the message passes

This is a decision worth stating plainly. If the criterion cannot be evaluated โ€” an error, the credit ran out, an answer that cannot be interpreted โ€” the message is relayed. It fails open, never closed.

The reasoning: a message that slips through can be seen and dealt with, while a message silently withheld is invisible, and nobody notices what never arrived. Filtering more aggressively when uncertain would trade a visible problem for an invisible one.

What it costs

Charged per message judged, against your AI credit. The practical way to keep that low is to put the ordinary filters in front of it: let a keyword list remove the obvious noise, and let the AI judge only what survives.

This or a pattern?

If the thing you want can be described mechanically โ€” a format, a fixed structure, a word with boundaries โ€” a pattern is cheaper, instant and predictable, and there is a separate page on writing one. Use a criterion in plain language for what a pattern genuinely cannot express: subject, tone, intent.

Which plans include it

The AI content filter belongs to the Platinum plan and is not part of the 7-day trial. The plan comparison is on the pricing page.

Common questions

Can I filter Telegram messages by describing what I want instead of listing keywords?

Yes. You write the criterion as a sentence and each message is judged against it before reaching the destination. It handles things a word list cannot express, such as whether a post is really an advert.

What happens if the criterion cannot be evaluated?

The message is relayed. It fails open on purpose: a message that slips through can be dealt with, while one silently withheld is invisible.

Does it replace keyword filters?

No, it runs alongside them, and it is better placed behind them: the cheap checks remove the obvious noise first, so the AI judges only what survives and costs less.

Is it available during the trial?

No. The AI features belong to the Platinum plan and are outside the 7-day trial.

Try Phantom Forwarder free

7-day trial, unlimited rules. Cloud-based, anonymous, nothing to install.

Related

More guides