X has no 'Made with AI' toggle
X's help pages do not describe a composer AI toggle. The 90-day figure is real and scoped to unlabeled AI war video, not a general labelling regime.
There is a claim circulating that X added a post-level "Made with AI" toggle, and that failing to use it on synthetic content about contested topics triggers a 90-day suspension from revenue sharing. It gets repeated in creator newsletters, in threads about platform compliance, and in checklists that people are genuinely building their workflows around.
The bundle is the problem. One piece is a composer control that X's help pages do not describe. The other is a 90-day number that is real and narrowly scoped, then stretched until it sounds like a general AI-labelling regime. Trace the toggle and it often lands on a Grok reply rather than a help page. A model built by a company is not that company's policy desk. Trace the 90 days and it lands on a March 2026 product announcement about unlabeled AI war video, not on a checkbox.
This matters because the failure mode generalises. Policy advice is now largely written by systems that don't distinguish a rule from a neighbouring rule, and creators are downstream of that.
What X actually publishes
The published surfaces relevant to AI content and monetization, as far as X's own help domain goes, are these:
| Page | What it covers |
|---|---|
| Synthetic and manipulated media policy | Media that could deceive and cause harm may be labelled or removed |
| Original Content Rewards | Payout program requirements — Premium, age, country, followers, impressions, original content |
| Content monetization standards | What disqualifies an account from earning, including bot-inflated engagement. AI-generated videos of armed conflict must carry a clear disclosure; realistic video-game depictions of armed conflict must be labelled synthetic |
That's the set. None of them describes a per-post AI disclosure control in the composer. A 90-day revenue suspension does exist, but not as a general penalty for unlabeled synthetic posts. In March 2026, X's head of product, Nikita Bier, announced a 90-day Creator Revenue Sharing suspension — permanent on a further violation — for AI-generated armed-conflict video posted without a disclosure that it was made with AI. TechCrunch and The Guardian reported it. The current monetization standards still require that disclosure. Original Content Rewards eligibility text does not attach a 90-day clock to unlabeled AI as a class.
The absence is genuinely notable, because it's not the norm. Meta applies an "AI Info" label triggered by industry-standard metadata or creator self-disclosure at upload. YouTube requires disclosure for realistic altered or synthetic content in three specific cases and provides the control to do it. Pinterest goes furthest, applying "AI modified" labels from metadata, owner disclosure, and classifiers that detect generated content without obvious markers. Against that backdrop, "X has a toggle too" is a very easy thing to assume — which is probably why the claim spread.
The operational consequence is small and specific: on X, disclosure is something you write. There is no checkbox that does it for you, so if you want a post disclosed, it goes in the post text, the video itself, or your profile.
Why a chatbot reply is a worse source than a blog post
A secondhand blog post at least has a byline and a publication that can be wrong in public. A generated reply has neither, and it has two properties that make it uniquely bad as a policy source.
It produces the shape of an answer regardless of whether one exists. Asked "does X have a Made with AI toggle," a model that hasn't got a documented answer will still return a fluent one. Absence and presence look identical on the way out.
Specificity is mixed with neighbouring facts. "90 days" is a real number from a real announcement. The tell is the rest of the sentence: a composer toggle, and a penalty that applies to any unlabeled synthetic post about a contested topic. That is how a scoped armed-conflict rule becomes a general AI-labelling regime in one hop.
Grok being X's own model makes a fluent mix-up worse, because it adds an illusion of insider access. It has the same relationship to X's help pages that any model has to any documentation: it has read text about them, and it can produce more text.
A four-step check for any secondhand platform claim
This takes about three minutes and catches nearly everything.
Find the primary URL, on the platform's own domain. Not a news article, not an aggregator, not a screenshot.
help.x.com,support.google.com/youtube,help.instagram.com,values.snap.com. If nobody citing the claim can produce that URL, it is a rumour, and you shouldn't build a workflow on a rumour.Isolate named UI and specific numbers. Claims about a button, toggle, or menu item that exists in the product, and claims about a duration, threshold, or count. Both are trivially checkable. Named UI is what most often gets invented; numbers are what most often get copied from a neighbouring rule. A UI claim you can verify by opening the composer. A number you can only verify against the page that states it.
Check the provenance chain to its origin. Follow the citations backward until they stop. A surprising share of confidently-repeated platform "policy" terminates at one unsourced post, one chatbot reply, or one corrected article. If the chain ends anywhere other than the platform, you've found your answer.
Test internal consistency against what the platform does publish. A claimed penalty should fit the enforcement structure the platform actually describes. If the published consequences are "labelled or removed" plus a disclosure duty on AI armed-conflict video, and the claim introduces a composer toggle that appears on no help page, that mismatch is evidence against the toggle — not against every number sitting next to it.
Applied to the circulating bundle: no primary URL for a composer toggle, a named UI element you cannot find, and a 90-day figure that is real but scoped to unlabeled AI-generated armed-conflict video. The toggle fails the check. The number survives, narrower than the thread that cites it.
This isn't unique to X
Absence-of-a-rule is a genuinely common finding once you check, and it's usually more useful than the rule people assumed. The UK's advertising regulator, for instance, has no blanket AI disclosure rule — the obligations that apply come from existing rules about misleadingness, not from a new AI-specific one. Same structure: everyone expects a dedicated rule, the dedicated rule doesn't exist, and the actual obligations sit somewhere less convenient.
There's a second pattern: rules that exist but do less than people fear. YouTube's disclosure requirement is real — and YouTube's own page states that disclosure does not limit audience reach or monetization eligibility. Creators avoid labelling out of a belief in a reach penalty the platform explicitly denies. That's the mirror image: an invented cost attached to a real feature, instead of an invented feature attached to a real platform.
Both errors come from treating platform policy as folklore rather than as documents you can open.
What to do on X in the absence of a toggle
Nothing here is complicated, it just has to be done by hand.
Decide the scope first. Per-post disclosure and profile-level disclosure aren't interchangeable — the two kinds of AI disclosure post works through when each is appropriate. On a platform with no post-level control, a profile-level line is a reasonable default for an account whose output is mostly generated.
Write the line, don't bury it. If disclosure is going in post text, it needs to be legible without expanding anything. Writing a disclosure line nobody scrolls past covers the craft; the short version is that it belongs early and in plain words.
Keep the manipulated media test separate. Disclosure and the deception-plus-harm test are different instruments. Labelling reduces the chance a viewer is misled; it doesn't authorise fabricating a real person's statement. Both apply, independently.
Don't let the absent toggle stop cross-platform hygiene. If the same asset ships to five destinations, four of them may have their own controls. The cross-platform labelling checklist is the practical version, and how Meta decides to show an AI label explains the automated path that fires without your input. Whatever handles posting and scheduling moves the file; it does not make the disclosure decision.
Understand what an AI label actually is. The AI content label glossary entry draws the line: a platform badge is a product feature, and legal disclosure obligations exist independently of whether any platform ships one. Satisfying a toggle says nothing about a jurisdiction's requirements. The absence of a toggle doesn't excuse you from them.
FAQ
So is it definitely true that X has no AI toggle?
Nothing on X's published help pages describes a post-level composer control. Platforms ship features without immediately papering them, so open the composer and check the help pages yourself before treating either the claim or this correction as settled. Do not take the absence of a toggle as the absence of disclosure duties: AI-generated videos of armed conflict still need a clear disclosure under the monetization standards.
Should I disclose AI use on X anyway?
If your content is generated and your audience would reasonably want to know, yes — and on X that means writing it into the post or your profile. The absence of a control is not a signal that disclosure is unwelcome, just that it isn't automated.
Why do so many creator resources repeat the toggle claim?
Because it's plausible, several other platforms have exactly that feature, and the correction is less interesting than the claim. Once something enters the secondhand layer it propagates on plausibility, and each repetition adds apparent confirmation without adding any evidence.
Is there a fast version of the four-step check?
Ask for the URL. If the person making the claim can't produce a page on the platform's own domain, everything else is optional — you already know you're looking at a rumour rather than a rule.
X's help pages for the synthetic and manipulated media policy, Original Content Rewards, and the content monetization standards are the authoritative text on everything above. This post is secondhand too; apply the same four steps to it.