Algorithm updates

Under the Hood: how to check if X is limiting your clip account

X's Under the Hood pilot, launched August 13, 2026, shows the labels that limit your posts' visibility. Who gets it, what each label does, what it hides.

Updated October 4, 20269 min read7 sources

In short

Open X's Under the Hood page (x.com/i/jf/under_the_hood). X launched it as a pilot on August 13, 2026 for a test group of accounts; Gigazine reported the conditions as an account at least one year old with 10 or more posts in the previous month. The report covers one month and lists each visibility label on your posts and account, how many posts or days it covered, and its effect. Since September 18, 2026 it also shows posts withheld in a country after a legal demand.

On this page

Under the Hood replaces guessing about a "shadowban" with a label list
Who can open the report
What the monthly report contains
The labels that matter most for clip accounts
Where the labels come from
"Hidden from non-followers" is the typical clip-account limit
What the report does not show
What to do if your report shows a label
Why this matters beyond one account
Questions
Sources

Under the Hood replaces guessing about a "shadowban" with a label list

Clippers often suspect a shadowban: a quiet limit on reach with no notice. On X the honest answer used to be that you could not check. On August 13, 2026, X published visibility code in its xai-org/x-algorithm repository and paired it with a transparency tool called Under the Hood.

The tool shows the labels on your account and posts. A label is a tag that automated systems or reviewers attach, such as SPAM_HIGH_RECALL. X's visibility system reads labels and decides, for each viewer, whether to show a post, show it behind a warning, or drop it.

The README explains why X built the tool. Some labeling rules and the Grox prompts (instructions for X's language-model classifiers) are not published, to reduce gaming. Instead, X shows the outcome: which labels hit your account. X says this lets you see whether labels were applied manually outside automated systems, and match each label to the published code.

For the TikTok, Instagram and YouTube side of this question, see shadowbans.

Who can open the report

The page is at x.com/i/jf/under_the_hood. The README calls it a pilot and says X will expand availability over time. If you are not in the pilot, the page shows this text: "We're testing this new feature with a small group to collect feedback."

Gigazine reported on August 14, 2026 that the pilot goes to a randomly selected test group of accounts that meet two conditions:

The account is at least one year old.
The account made 10 or more posts in the previous month.

The page code in the repository matches this: it renders a checklist of eligibility checks, and its sample data uses these same two checks. A new clip account fails the age check by design. Our reading: if you run a young account, you cannot use this tool yet, so keep an older account active if you want this view.

Gigazine quoted Keith Coleman, X's VP of Product, saying the tool lets users download the statistics as JSON (a plain data file). The page code shows the download button on the web only, not in the apps.

What the monthly report contains

Each report covers one calendar month. The page code and its sample data show these parts:

Part	What it shows
Period	Start and end date of the month, with time zone
Report generated	When X produced the report
Posts in period	How many of your posts the report counted
Post labels	Each label on your posts, with the number of labeled posts, the total, and a percentage (for example, 3 of 40 posts)
Account labels	Each label on the account, with the number of days it was active in the month (for example, 5 of 31 days)
What it means	X's plain description of each label
Effect on visibility	X's description of what the label does
Legal withholding (since September 18, 2026)	Posts withheld in a country after a legal demand, and which country

The report gives totals per label, not a list of which posts were labeled. If 3 of your 40 posts carry a spam label, you learn the count, not the 3 post IDs. Our reading: compare the month's analytics. Posts with views mostly from followers and almost none from For You are the likely ones.

The labels that matter most for clip accounts

The repository contains the exact text the report shows for each label, in underTheHoodLabels.strato. These are the ones a clip account is most likely to meet. The effect column is X's wording, shortened.

Label	Applies to	Effect X states
SPAM_HIGH_RECALL	Post	Hidden from recommendations to non-followers
SPAM	Post	Post is not shown on X
MALICIOUS_URL	Post	Hidden from recommendations to non-followers
DO_NOT_AMPLIFY	Post	Hidden from recommendations to non-followers; being replaced by MALICIOUS_URL
is_dmca	Post	Hidden from recommendations to non-followers; media unavailable where seen
Dmca	Post	Post is not shown on X
FOSNR_ABUSE_INSULTS	Post	Hidden from recommendations to non-followers, with a visible limited-visibility label
SpamHighRecall	Account	All the account's posts hidden from recommendations to non-followers
ImpersonationHighPrecision	Account	All posts hidden from recommendations to non-followers
AbusiveHighRecall	Account	All posts hidden from non-follower recommendations; security challenge required
DoNotAmplify	Account	All posts hidden from non-follower recommendations pending review
Compromised	Account	All posts hidden from non-follower recommendations; password reset required
ReadOnly	Account	Read-only until the violating post is removed

The list also covers adult-content, violent-content and other policy labels, which matter less for podcast and stream clips. Note the two copyright labels: a successful copyright report can strip the media from a clip or remove the post, which ends its campaign views. Impersonation is a risk for fan accounts that use a creator's name and photo; X's parody, commentary and fan rules are summarized in Can clippers earn on X?.

Where the labels come from

The README describes a labeling path that runs apart from the feed request. Knowing its parts helps you guess what triggered a label:

Rules on live events. A rule engine reacts as events happen: on this event, if these conditions hold, apply this label. Some of these rules are published; some are withheld.
Account models. One batch job, agatha, labels an account from how others respond to its posts: blocks, reports and spam reports relative to likes. Another, bdsm, reads the sequence of actions an account takes over time to spot inauthentic behaviour.
Content classifiers. Grox and the media models classify posts as they are published, for spam, adult content and violent media.
Enforcement service. A separate service reads account scores and can label an account or its posts, challenge it, or suspend it.

Our reading for clippers: the agatha input means that a clip which draws many blocks and reports compared with likes can hurt the whole account, not only that post. Bait hooks and misleading captions carry this risk.

"Hidden from non-followers" is the typical clip-account limit

Most labels above share one effect: the post is hidden from recommendations to people who do not follow you, while your followers still see it. The README explains why. Visibility filtering has a set of rules that apply only when a post is recommended from an account the viewer does not follow, and those rules can only drop. The same post is allowed to a follower.

For a clip page that lives on For You reach, this is close to a full stop. Our reading: one account label such as SpamHighRecall, active for a few days, would explain a sudden drop to follower-only reach during those days.

Ranking and visibility are separate in X's design. A label does not lower your score; it removes the post for a group of viewers after ranking. No amount of engagement fixes a label. For how the score itself works, see what X's ranking weights reward.

What the report does not show

Under the Hood is narrower than its name. Four gaps matter for clippers:

Only listed labels appear. The repository describes the label file as the "public label and takedown allowlists". A label outside that list is not shown. For example, X's code applies a COPYPASTA_SPAM label to clusters of duplicate text, and that label is not in the report's list. This is our reading of the code.
No ranking information. The report does not show your scores, the new-author boost or the out-of-network discount.
No payout information. Impression reassignment for reuploads and Original Content Rewards decisions are not part of it. See reupload detection on X.
Monthly, after the fact. You see last month, not today. A label that hit you this week shows up in next month's report.

What to do if your report shows a label

Download the JSON on the web each month. Keep a record. A change from month to month tells you more than one report.
Read the effect line first. "Hidden from recommendations to non-followers" means your followers still see the post; "not shown on X" means nobody does.
Clear account labels that ask for an action. Compromised asks for a password reset. AbusiveHighRecall asks for a security challenge. ReadOnly lasts until you remove the violating post.
Stop the pattern behind spam labels. Common clipper triggers are posting the same clip and caption across accounts, mass posting in bursts, bought or traded engagement and link-heavy posts to doubtful sites. See flagged views and view bots and running several accounts. This list is our reading of X's spam rules, not a list X published.
Appeal through X's normal channels. The repository shows that when an enforcement is overturned on appeal, the automated system does not re-apply that action for a fixed period, set per appeal queue.
Tell the campaign. If a label covers days when your campaign posts went live, report it to the campaign manager with the JSON. Low views during a labeled period are not a content problem.

Why this matters beyond one account

Under the Hood changes how X disputes can work. Before August 2026, a clipper with a sudden view drop had only guesses. Now an eligible account has a dated record that names a label and quotes X's stated effect.

The limits are real: a pilot with random selection, a one-year age gate, monthly totals and an allowlist of labels. Treat the report as evidence, not as the whole picture.

Questions

Is Under the Hood available to everyone on X?

No. As of October 4, 2026 X calls it a pilot. Gigazine reported that it goes to a random test group of accounts at least one year old with 10 or more posts in the previous month. Others see a message that X is testing the feature with a small group.

Can I download the report in the X app?

The page code in X's repository shows the Download button only on the web. Open x.com/i/jf/under_the_hood in a browser to get the JSON file.

If my report shows no labels, does that mean nothing limits my reach?

No. It means none of the labels on X's published list hit you that month. Ranking, the out-of-network discount, labels outside the list and reupload reassignment are not in the report.

Does a label mean my account is suspended?

No. Most labels limit who sees your posts. Suspension is a separate enforcement action. The SPAM, Dmca and legal withholding labels do stop a post from being shown, worldwide or in named countries.

What changed on September 18, 2026?

X added legal compliance to the report. It now shows whether your account or posts were withheld in a country after a legal demand, and which country.

Why is a copyright label in a visibility report?

Because a copyright report changes who sees the post. X lists one DMCA label that hides a post from non-follower recommendations and makes its media unavailable, and another under which the post is not shown at all.

Sources

We checked every claim on this page against these sources on the date above. Platform rules and rates change often. Found an error? Tell us.

In short

Questions

Sources

Related guides