An AI account on social media: what gets banned and what does not
A great deal is said about bans for AI content and there are no specifics. The truth is not what people think: the fact of generation is not itself a violation. Below are five things accounts genuinely lose over, and three causes of falling reach unrelated to bans.
A great deal is said about bans for AI content and there are no specifics. The truth is not what people think: the fact of generation is not itself a violation. Below are five things accounts genuinely lose over, and three causes of falling reach unrelated to bans.
The fear that stops people starting
Somebody has learned to make a character and is ready to run an account — and stops on one question: will I be banned.
The question is reasonable. A great deal is said about bans for AI content and there are almost no specifics: some claim the platforms are fighting generation, others that nobody cares.
The truth is in between, and it is not built the way people think. Platforms do not ban you for content being generated. They ban specific things, and generation itself is not on the list.
Let us go through what is genuinely dangerous.
What is not a violation
Let us remove the main fear straight away.
Generated content is permitted. Images, video and text created by models are published legally on every major platform.
A virtual character is permitted. A fictional character running an account is a long-established format.
Labelling. Platforms ask for synthetic content to be marked, and some do it automatically. That is a formatting requirement rather than a ban: labelled content is not automatically downranked and not blocked.
The practical conclusion: the fact of generation is not a problem. The problems begin in the same places they do for everybody else.
Five things accounts genuinely lose over
1. Somebody else's face
The most dangerous and the most common.
Using a celebrity, a public figure or somebody else's photograph from the internet in a generation is not allowed. That is not a question of platform rules — it is a question of a person's rights to their own image, and the complaint comes from a lawyer rather than an algorithm.
What is legal: your own face, a character you trained on your own photographs — not stock people and not somebody else's faces.
For commercial work that is the decisive point: you hand a client a clip and do not have to think about whose face is in it.
2. Passing synthetic off as real where that is deception
The distinction is fine and worth understanding.
A virtual character advertising a product is fine. A virtual character presented as a real buyer with real experience — "I bought it and it helped me" — is already a false review.
The boundary is simple: the format may be staged, the factual claim may not. You cannot attribute to a non-existent person an experience that never happened.
3. Volume and sameness
Here the platforms react firmly, and not because of AI.
An account publishing dozens of near-identical pieces a day falls under spam filters regardless of what made them. A pipeline creates the temptation to post a lot — and that is the fastest way to lose reach.
What to do: three or four publications a week. By my observations the relationship between frequency and engagement is negative — the more often, the worse the response. So the constraint here coincides with what is more effective anyway.
4. Inflating figures and automating actions
Publishing through a service is fine. Automatic follows, likes and comments from your account are a direct and punishable violation.
The difference: publishing your own automatically is allowed, interacting with other people's automatically is not.
5. Ordinary violations that have nothing to do with AI
Medical promises, financial guarantees, prohibited product categories, somebody else's music without rights. Everything live bloggers get banned for.
Generation neither softens nor tightens those rules — they are the same for everyone.
What to do about labelling AI content
A practical approach that settles the question.
Label it. The platforms have a labelling tool and using it does not worsen distribution. There is no point hiding it: it is detectable anyway, and the consequences of concealment are worse than of an honest label.
Do not turn it into a subject. A label is administrative information rather than content. An account that explains in every post that it was made by AI is selling AI rather than its own product.
For commercial work, agree it with the client in advance. Who applies the label, who is responsible for compliance with advertising requirements. That is part of the handover rather than a detail.
What actually lowers reach
Separately, because this gets confused with bans.
The account was not banned but stopped being shown. There are usually three causes, and none is about generation.
Sameness. Five publications from one piece of raw material must differ not only in form but in angle. Otherwise the viewer sees one post cut into pieces.
No response. Material that is not saved and not shared stops being shown by the platform. A share is the scarcest reaction and the most honest signal.
Synthetic text. Not the picture but the caption. The same turns of phrase, an even lifeless rhythm, "in today's world" — a reader recognises that instantly and scrolls on. A separate cleaning step for machine clichés is obligatory.
The last is the easiest to fix: the cleaning can be a separate step: ask Cerebrum or an assistant to reread the text and strip out typical machine phrasing.
How to run an account safely: the short code
The face — only your own, or one you trained.
Staging is allowed, false facts are not.
Three or four publications a week, not thirty.
Publishing automatically yes, interacting automatically no.
Label it, but do not make it the subject.
Clean the texts of machine phrasing.
Six rules cover almost every real risk.
What the platform offers for an AI account
Training your own consistent character in Gen AI — a face you hold the rights to.
Cleaning off AI clichés — as a separate step: Cerebrum or an assistant rereads the text and strips out typical machine phrasing.
The "Marketing" category assistants — 26 of them, including social media work and brand management. They help with what matters more than the rules: getting the content shown at all.
Registration is free and opens three days of full Basic access — all the assistants, the prompt texts and six foundational express courses.
Express courses on image and video generation (Midjourney, ChatGPT Images & Nano Banana, Veo 3) — Plus plan and above, or $9 one-off.
Where to start today
One action. Check your existing or planned account against one point — the first.
Where your character's face came from. If the answer is "found it on the internet" or "took it from stock", that is the one point of the six that can cost you not reach but a legal complaint. It is changed in one evening: train your own character on your own photographs.
The other five rules are about the account growing. The first is about it existing at all.
What on the platform is built so as not to expose you
Training your own character in Gen AI — the character is trained on your own photographs, so it is not stock people or somebody else's faces, and no rights questions arise.
Analysing somebody else's clip in Cerebrum is worth setting up so that the structure is copied rather than the person: a successful video is broken down into its hook, its script frame and its techniques. The author's face and voice are not copied.
Where permissible automation ends
Publishing your own content on a schedule is allowed — the platforms' own official tools and scheduled-posting services do that.
Automating actions in other people's accounts is not, and that is exactly what platforms ban. The boundary runs here rather than at the volume of publications.
What helps avoid restrictions
Three things from practice, besides the above.
Consistency instead of randomness. An account where the character changes from post to post reads as a warehouse of generated material. An account with one recognisable face reads as a person's account. A trained character or the free “Series with a consistent character” skill (it works in Cerebrum and through the AI SKILLS connector, Full plan; generations are paid from the balance) solves that.
Publishing through official channels. Publish through services that work with the platforms' official APIs (for example the Ayrshare skill for social posting) rather than through emulating user actions. It is the second mechanic that most often leads to restrictions.
Density and rhythm. A hundred clips a day is technically possible — and it is also the fastest way under the filters. A pipeline gives volume, but it has to be published at a human rhythm.
What to take on the platform for this task
- AI social media manager — social audit, content, handling hate and reputation; assistant, from Basic
- SMM strategist — SMM strategy for public communications; assistant, from Basic
- Brand manager — brand management and positioning; assistant, from Basic
- Series with a consistent character — a series of scenes with the same hero; a free skill (works in Cerebrum and through the AI SKILLS connector, Full plan; generations are paid from the balance)
- Ayrshare — social publishing through an official API; skill
What to read next
Rights to a face in advertising — the one point that costs a legal complaint.
What an AI blogger earns from — what the account is run for.
The cold top of the funnel — what works for reach here.
How to find trends — what to post about so it gets shown.
AI video generation in 2026 — the full guide to video models.