Block Twitter Accounts by Keyword or Regex (and Why Mute Words Aren't Enough)

What muted words actually do

X's muted words live in Settings → Privacy and safety → Mute and block → Muted words. You add a word or phrase, choose whether it applies to your timeline, your notifications or both, whether it applies only to people you do not follow, and for how long.

Within those limits it works well, and it is the right tool for a topic you are tired of. But note what it does not touch:

• It hides posts. The account is untouched — it can still follow you, reply to you, quote you and message you.
• It matches the text of posts. It does not match display names, handles or bios, which is where bot accounts advertise themselves most reliably.
• There is no way to turn a match into any action other than hiding.
• Muting is per-word and permanent maintenance: a new spam script needs a new word, forever.

Keyword mute vs keyword block

They are complements, not competitors. Mute a topic; block an account. If what bothers you is a subject, muted words are the cleaner answer. If what bothers you is a stream of accounts that will be back tomorrow under a new post, a keyword block is the one that ends it.

Which field to match

In Settings the four fields are independent switches, and the choice matters more than the wording of your rules. Ranked by how often a match means what you hoped it meant:

1. Handle. A handle is chosen deliberately and almost never by accident. airdrop in a handle is a near-certain hit.
2. Display name. Nearly as good, and it is where impersonators and giveaway bots put their pitch.
3. Bio. Good for off-platform contact details and promo language. Only available for accounts whose profile data the extension has already seen, so treat it as a bonus rather than a foundation.
4. Post text. The obvious choice and the worst one. Everything ambiguous lives here: criticism, quoting, jokes, and people asking whether a thing is a scam.

A concrete version of that: with post text on, the keyword giveaway catches the bot and also catches the person writing "another fake giveaway in my replies". With post text off and handle on, it catches @Giveaway_King2026 and leaves the complaint alone.

Writing keywords that hold up

Keywords are plain, case-insensitive substrings — no syntax, one per line. Three habits make them reliable:

• Prefer phrases to words. claim your reward is safe; claim is not.
• Remember substrings do not respect word boundaries. art matches "start", "particle" and "@artist". If you want the word, use the regex box and \bart\b.
• One rule per idea. A rule you can explain in a sentence is one you can fix later. Long clever rules get deleted wholesale the first time they misfire.

The extension also ships with a small built-in list aimed at a specific genre of reply spam, merged into your filters the first time it runs. It is additive only: delete any of those lines and they stay deleted.

Regex, in the amount you actually need

Patterns go in the Regex box one per line, in JavaScript syntax, with no surrounding slashes. Five pieces cover nearly everything:

Two properties of this particular engine are worth committing to memory. Patterns are case-insensitive, so never write [Aa]irdrop. And they are compiled without the Unicode flag, so \p{L} and friends do nothing useful — reach non-ASCII characters with explicit ranges instead.

The two-condition pattern

This is the one construction worth learning, because it converts a word that is too broad into a rule that is safe. (?=…) is a lookahead: it asserts that something is present without consuming it, so several of them in a row mean "all of these must appear somewhere".

(?=.*airdrop)(?=.*\b(dm|telegram)\b)

That reads: mentions an airdrop and pushes you into a private channel. The scam does both. The person complaining about the scam does the first and not the second, so they are untouched. Any ambiguous word can be rescued this way — give it a chaperone that the spam always has and ordinary speech rarely does: a contact channel, an imperative like claim, or a link shortener.

What happens to the text before your rule sees it

Your rule is not tested against the raw page text alone. The extension builds three versions of it and tries each:

1. The raw text, exactly as the page has it — so a rule written to match punctuation or a fullwidth character still works.
2. A cleaned version: invisible characters stripped out, then Unicode-normalized so fullwidth, circled and sub/superscript letters fold back to plain ones.
3. A tight version: the cleaned text with every space, punctuation mark and emoji removed, leaving only letters and digits.

That third one is why a keyword still matches when a spammer spaces a word out with emoji between the characters. Your keywords get the same treatment, so a phrase you copied straight out of a spam reply — invisible characters and all — still works as a filter. The reason any of this is necessary is covered in the invisible-character trick.

The allowlist is part of the filter, not a safety net

The third box in the Filters tab takes handles, one per line, without the @. Those accounts are never flagged and never appear as candidates, whatever they post.

Treat filling it in as step one, not damage control. The accounts most likely to be caught by a broad rule are the ones you read most: the people in your niche discussing the scams, quoting the bots, and using the same vocabulary. Thirty seconds spent putting them in the allowlist buys you permission to write a much more aggressive rule.

Testing a rule before you trust it

1. Add the rule with scanning on and fully automatic off — which is the default, and the reason it is the default.
2. Scroll a busy reply section and watch the Candidates tab. Every row names the rule that flagged it.
3. Read the list without blocking anything. A rule that produced three obvious bots and one person you recognize needs a chaperone before you use it.
4. Only when a rule has produced a clean list twice should you consider letting it run without review — and even then, the accounts you care about belong in the allowlist first.

The full workflow around this, including pacing and rate limits, is in how to mass block accounts on Twitter.