What Every Platform Now Does About AI Content.

Seven of the platforms most brands publish on now hold a stated position on AI-generated content, and they differ.

Some label it. Some hand the viewer a dial. One has stopped recommending it. Others decide whether it earns anything.

That makes AI less of a productivity decision and more of a per platform compliance question. Lift output across every surface you publish on and you are under a different rulebook on each, where the rule that costs real money is rarely the one anyone noticed first.

The three things a platform can do about AI content

Platform rules on AI come in three shapes, and they cost you very different things.

• Label it. The post runs as normal and viewers see a marker, sometimes with provenance data attached.
• Refuse to recommend it. The post stays up and your followers may still see it, but the algorithm stops carrying it to anyone else.
• Refuse to pay for it. The post runs and earns nothing, and that judgement usually lands on the account, not the post.

Most teams react hardest to the first and plan least for the other two, which is backwards. A label is a presentation cost. Losing recommendation or payout eligibility changes the numbers, and the fix is the discipline that made AI useful in the first place: keep it in the repeatable parts of the job, not the parts that carry the brand’s voice.

Pinterest labels it, then hands the viewer the dial

Pinterest adds an AI modified label when its systems detect that AI was used to generate or change a Pin, or when the owner discloses it. Detection reads IPTC photo metadata and runs classifiers that pick up content carrying no obvious markers. Labels can be appealed through support.

The sharper change sits on the viewer’s side. Pinterest’s tuner lets people see less GenAI content on eligible image Pins in the categories most prone to it, including art, beauty, fashion, home decor and health. It runs on desktop and Android, with iOS rolling out.

So Pinterest does not demote AI content globally. It lets each user demote it for themselves, which for a brand in beauty or interiors is a firmer ceiling than any label.

YouTube keeps the label and the money apart

YouTube asks creators to disclose content that is meaningfully altered or synthetically generated and looks realistic: a real person appearing to say something they never said, footage of a real place or event altered, or a realistic scene that never happened. Plenty is exempt, including beauty filters, colour adjustment, special effects, captions, idea generation, using AI to build an outline, script, title or thumbnail, and cloning your own voice for a voiceover.

Disclosure itself is free. YouTube’s stated position is that it will not limit a video’s audience or affect its eligibility to earn money. Consistently failing to disclose is what costs you, and YouTube can apply the label itself, remove the content, or suspend the channel from the Partner Programme.

The money rule is separate. In July 2025 YouTube renamed its repetitious content policy to ingenuine content: content should not be mass-produced, generic, repetitive or manipulative, and AI-generated content built on unoriginal templates with none of the creator’s own perspective fails it. Partner Programme membership belongs to the channel, so the exposure is never just one video.

TikTok is labelling at scale, and testing a dial of its own

TikTok requires people to label AI-generated content containing realistic images, audio or video. It was the first video platform to implement C2PA Content Credentials, so AIGC made elsewhere is labelled automatically when that metadata travels with the file.

The scale matters. In July 2026 TikTok said it had labelled over 3 billion videos as AIGC, using Content Credentials, creator labels and invisible watermarking, a watermark only TikTok can read and harder to strip out.

It is also testing an AIGC setting inside Manage Topics, which TikTok describes as helping people tailor the range of content in their feed rather than removing it, alongside better detection of AI-spam accounts. Testing is not shipped, and that dial turns both ways.

Snapchat has drawn the hardest line

In July 2026 Snap said wholly AI-generated videos are no longer eligible for recommendation on Spotlight. Content enhanced or edited with Snapchat’s own AI creative tools stays eligible and carries transparency indicators.

That is a distribution rule rather than a labelling one, and the strictest of the seven. Recommendation is how Spotlight reaches anyone beyond your existing followers, and rewards follow views, so for fully AI-made video the earning route closes with it.

Meta labels the AI and demotes the unoriginal

Meta requires people to disclose, through its AI-disclosure tool, when they post organic content with photorealistic video or realistic-sounding audio that was digitally created or altered, and says it may apply penalties when they do not. Its AI info label appears when Meta detects industry standard AI indicators or when someone self-discloses, and for content only edited with AI the label sits in the post’s menu.

Meta labels and keeps synthetic media rather than removing it, unless it breaks another rule.

The distribution rule is about originality rather than AI. In March 2026 Meta said Facebook is prioritising original creators: content filmed or produced by the account counts as original, third-party footage counts only when the creator adds fresh information, analysis or substantial improvement, and accounts posting mostly unoriginal content can be deemed non-recommendable and demonetised.

Those two rules catch different failures, and the second is where volume-first AI output usually dies. When reach drops on one surface and nobody can say whether it is the policy or the work, that is a platform optimisation problem, not a reason to post more.

LinkedIn treats it as a disclosure question

LinkedIn’s community policies tell members not to share synthetic or manipulated media showing a person saying or doing something they did not, without clearly disclosing that the material is fake or altered. Images and video signed with C2PA Content Credentials also carry a provenance icon, which can include an assertion about whether AI was used.

That is a deception rule, not a distribution one. The risk on LinkedIn is reputational before it is algorithmic, and a B2B audience does not need the platform’s help to punish a synthetic-looking case study.

On Reddit the binding rule is the subreddit’s

Reddit is the outlier, because the rule deciding whether your post survives is usually written by moderators rather than by the platform. Those rules spread fast: an academic study of 227,737 subreddits found the share carrying AI rules doubled between July 2023 and November 2024, reaching 17.1% among the largest 1% of communities, and roughly 55% of them are flat bans rather than disclosure requirements.

Read the sidebar before you post, community by community. The penalty here is not a label, it is removal by a person who made their mind up about AI content long before you arrived.

What good looks like when every platform differs

You do not need seven AI policies. You need one internal standard that clears the strictest platform you publish on, and one check at the point of publishing.

1. Record what the AI actually did on each asset: idea, script, edit, voice, or wholly generated. The platforms split on that distinction, so your records should too.
2. Treat wholly generated as its own content class with its own approval. That is the class that loses recommendation and payouts.
3. Disclose at the source. Credentials that travel with the file are what several of these platforms read to label the post correctly.
4. Keep human origin in the work that carries the account: filmed, observed, argued. Original perspective is what every one of these policies protects.
5. Review payout rules quarterly and label rules when they move. Labels alter presentation. Payout rules alter revenue.

If approvals are informal, this is where it breaks first. Adding the AI question to a checklist takes an afternoon, and the free social media approval process template saves inventing the process from scratch.

Next step

None of this makes AI unusable. It makes unmanaged AI expensive, because the cost lands on distribution and payouts rather than a warning nobody reads.

If your output has gone up and your reach has not, NBK can help find the constraint in the system before more content gets poured into it.