Key takeaways
Nobody outside Twitter can promise a post is safe. The rules that apply labels are public. The detectors behind them mostly aren't.
Five patterns are worth checking: near-copies, the same text on several accounts, low-quality links, posts that read as AI-made, and asking for engagement.
A spam label on a post removes it from recommended feeds, so people who don't follow you stop being shown it.
A link with a mention is only a risk in the published rule when the site is rated low quality.
Twitter's Under the Hood page is the closest thing to a real answer, if your account has access.
You can't know for certain whether a post will get a spam label on Twitter, because the systems that decide aren't public. But Twitter's published code names the patterns that lead to one: near-copies of earlier posts, low-quality links, posts flagged as AI slop, and posts that bait engagement. Checking for those before you post is the best protection available.
Most advice on this topic is guesswork stated as fact. Here we separate what is in Twitter's published code from what nobody outside the company knows. We make VoiceMoat, and it comes up near the end, including what it can't do.
What a spam label does
Twitter publishes much of its ranking and safety code at github.com/xai-org/x-algorithm. In that code:
A spam label on a post drops it from recommended feeds. You can still see your own post.
A spam label on an account hides its posts from non-followers in those feeds. Depending on the rule, it lasts 7 or 30 days.
Large, high-credibility accounts skip many of these rules. The real thresholds aren't published. The repo uses placeholder values.
Our breakdown of the October 2026 update covers what changed that month. Published code can also differ from what runs live.
What the published rules flag, and what they don't say
Pattern | What the published code shows | What isn't public |
|---|---|---|
Near-copy of a post | A duplicate-text job labels clustered posts as copy-paste spam, and replies too | How similar counts, how far back it looks, what the label does to reach |
Same text on several accounts | Twitter's developer guidelines don't allow identical or similar content across accounts in apps | Whether the duplicate check links different authors |
Link with a mention | A spam label when the site is rated low quality and someone you mention doesn't follow you | How sites get rated |
Low-quality link alone | A second rule needs no mention at all | Same |
Reads as AI slop | An account flag brings a 7-day spam label. A post flag brings a separate 30-day label | What the detector looks for, and what the post label does in For You |
Engagement bait | A named policy applies a spam label, with no exemption for big accounts | Which phrases count |
On bait, Twitter's head of product said in March 2026 it was "very close to having Grok spot posts that directly solicit engagement," PiunikaWeb reported. There's no published phrase list, so any list you see, including ours, is someone's best reading.
Five checks before you post
1. Search your own posts. Paste a distinctive phrase from your draft into from:yourhandle "your phrase", or use our advanced search builder. If you find a close match, rewrite the draft around a new example or angle.
2. Don't send the same text from two accounts. If you run a personal and a company account, draft each post separately.
3. Separate links from mentions when you're unsure. You can't see how Twitter rates a site, or easily confirm who follows you. A normal link to a reputable site isn't what this rule targets, but splitting them costs little.
Invented example: instead of "Great piece by @sam: example.com/post", post the link with your own take, then thank Sam in a reply.
4. Cut the engagement ask. "Like if you agree", "Tag a friend" and "Follow for more" are the obvious ones. Beyond any label, as of late October a "not interested" carried a weight of -52.272 against 0.66 for a like. Those weights multiply predicted chances, so they aren't points, but negative reactions matter. Ask a question someone would want to answer.
5. Read it as a stranger would. Watch for stock phrases, a "this, not that" turn in every other line, and a closing question bolted on. Our piece on spotting AI slop has an audit. An AI detector can help, but none of them is Twitter's detector, so a pass proves nothing.
How to find out after you post
Method | What it tells you | Limits |
|---|---|---|
Under the Hood, Twitter's own page | Labels on your account and posts last month, and their effect | A pilot. Reported to need an account at least a year old and 10+ posts last month. One tester found it gives counts and doesn't name the posts |
Search test: our free checker or the steps in our shadowban guide | Whether your posts appear in search | Says nothing about For You |
Your own views | A sudden drop while followers hold steady | Could be many other causes |
OpenTweet's bait checker, free, from a vendor | Flags engagement asks in a draft | Phrase matching, before you post only |
How VoiceMoat handles this
VoiceMoat is an AI social media tool for LinkedIn and Twitter. Studio runs a risk check when you Publish, Schedule or Post on Twitter. It warns and still lets you post. If it finds nothing, you see nothing. If it finds something, you choose Edit post or go ahead anyway.
It looks for:
A near-copy: 80% or more of the wording matches a post on this profile from the last 30 days, posted or scheduled.
The same text on another of your Twitter profiles within 24 hours.
Any link plus any mention in the same post.
AI tells from the Style card.
The Chrome extension adds a bait check inside Twitter's own post box. If a draft says "Like if you agree", one line appears with Rewrite without it (1 credit) or Keep it. A phrase your own posts already use is left alone, and it never blocks posting.
Twitter Analytics can also show a "possible reach label" card when views per post fall hard while you keep posting and followers hold steady. It's an estimate, and the card says so.
What it can't do:
It can't see Twitter's labels or detectors. The 80%, 30 days and 24 hours are our own choices.
It flags more than the published rule does on links. It can't see a site's rating or who follows you, so treat that warning as a prompt to look.
It only runs in Studio. Posts sent through the API or MCP, or typed on Twitter, skip it. The bait check is only in the extension.
A clean check is no guarantee.
audenAI, the brain inside VoiceMoat, is told to avoid these patterns in drafts. It does the work, and you decide what goes out.
What each plan includes
Free | $1 trial | $5 trial | Creator | Pro | Enterprise | |
|---|---|---|---|---|---|---|
Price | $0 | $1 for 3 days | $5 for 3 days | $35 a month | $69 a month | $150 a month |
Risk check in Studio | Yes | Yes | Yes | Yes | Yes | Yes |
Bait check in extension | Yes | Yes | Yes | Yes | Yes | Yes |
Schedule how far ahead | No scheduling | 15 days | 15 days | 7 days | 15 days | 30 days |
Connected profiles | 1 | 1 | 1 | Up to 3 | Up to 4 | Up to 6 |
Analytics refresh | None | Every 7 days | Every 7 days | Every 30 days | Every 15 days | Every 7 days |
The other-profiles check needs two or more Twitter profiles. Extra profiles are paid add-ons. The reach card reads views already synced, so plans that refresh less often give it less to work with.
If you'd like this check built into your posting, try VoiceMoat: Creator is $1 for 3 days.
Frequently asked questions
Can I tell if a tweet will get a spam label before I post it?
Not for certain. You can check for the patterns in Twitter's published code, but the detectors that decide aren't public.
What happens if my post gets a spam label?
In the published code, it's dropped from recommended feeds, so people who don't follow you stop being shown it. You can still see it yourself.
Why is my tweet marked probable spam?
We couldn't find how that section is filled in the published code we searched. Replies that repeat themselves or arrive fast are common suspects, but that's informed guessing.
How do I see if my account has a label?
Open Twitter's Under the Hood page at x.com/i/under_the_hood. It's a pilot, so your account may not have access yet.
Can I post the same tweet twice?
Twitter's published code labels clusters of duplicate text, but how similar or how recent is not public. Rewrite repeats around a new angle.
Do links get you labelled as spam?
The published rules target links rated low or bad quality. A link to a reputable site isn't what they describe.
Is "Like if you agree" against Twitter's rules?
Engagement bait is a named spam policy in the published code. Which phrases count isn't published, so drop the ask to be safe.
Does VoiceMoat stop me posting a risky tweet?
No. It warns you and you can still post. You decide what goes out.
AI disclosure
This article was drafted with AI assistance and edited by Prateek Singh. Twitter's rules were read in its published code on 9 November 2026, at commit 3b6919f. Under the Hood details come from third-party reports, named in the text. VoiceMoat is our product.
Keep reading
Growth
Twitter shadowban test: how to check for a search ban, and what the platform documents
A Twitter shadowban test comes down to one thing you can verify from outside: whether your posts still appear in search. This guide shows the one-minute manual test, compares the free checkers, and explains Twitter's Under the Hood label report. It also covers what the open-source code says about visibility filtering, the six published causes of a search ban, and why nobody can tell you how long one lasts.
9 min read
·
Industry
Twitter engagement is down in 2026. Here is what the data actually shows.
Twitter engagement is down in 2026: the median brand rate fell from 0.029% to 0.015% in a year. This data read checks every benchmark against its source, then shows what Twitter's open-source ranking code says can lower reach, from the 48-hour window to visibility labels. It ends with how to tell a sudden drop from the slow trend.
14 min read
·
Industry
AI slop: the quiet marketing crisis nobody wants to name
AI slop is the average-quality, voice-flat, fluently-incoherent content that now floods every marketing channel. It is the quiet crisis of 2026: nobody wants to name it because too many teams are producing it. Here is what AI slop actually is, why marketing teams keep shipping it, and what the alternative looks like for creators who want to keep their audience.
13 min read
·
Growth
How to increase Twitter impressions without resorting to generic content
Most impression guides hand you a list of templates and call it a strategy. The templates work briefly, then plateau, then erode the audience you built. Here's how to grow impressions on signals that keep working: your voice, your timing and what Twitter's ranking actually counts.
7 min read
·
From the Twitter algorithm series
The Twitter/X algorithm in May 2026: why voice wins under the new transformer
Read ›
