A smoke detector that goes off every time you make toast is just noise. After the tenth false alarm you stop running for the door, and the day there’s a real fire, nobody moves. The social platforms have built exactly that with AI labels. TikTok, Facebook and Instagram all stamp posts as AI-generated now, mostly automatically, and they’re catching real artwork, real photographs and video of real hands. The machine can’t tell toast from a fire, and it can’t be argued with either.
A label that tells readers to skip you before they’ve read a word is a verdict, handed down without a trial. Nobody asked how the post was made, what the tool did, or whether a human wrote every sentence of it. A machine guessed, the guess went on the post, and the readers believed the guess.
What happened to one author’s book cover on TikTok
An author who has published for years posted a short TikTok in a big indie group. It was nothing more than a Kindle screenshot over an animated version of his book cover. The artwork was painted by an artist he’s worked with since before AI image tools existed.
TikTok labeled the video AI-generated. He appealed. The appeal went nowhere. Then TikTok removed the post for violating its Community Guidelines, without saying which one. He appealed that too, and the rejection came back in the same minute he submitted it.
The replies were full of the same story. One author’s video of her own hands turning book pages got the AI label. Another had a pillow on her sofa flagged. Images made in Canva have been labeled on Facebook and Instagram. Somebody got a strike on TikTok for stealing his own work.
Why does TikTok label real artwork as AI-generated?
Mostly because of what’s hidden in the file.
Many editing tools now write a tag into the file’s metadata when any AI feature touches it. The industry standard for this is called Content Credentials, and TikTok and Meta both read it. The platform never looks at your art. It reads a tag, and the tag doesn’t say how much AI was involved. It just says some was, and the platform turns that into a label on the finished post, as if a machine made all of it.
The likely culprit in the cover video was the animation. Most of the apps that turn a still image into a moving one use AI to do it, so the animation step stamped the whole file. The painting underneath was human, and the label can’t tell the difference. Canva works the same way: use one of its AI features on a single element and the file goes out marked. One small step in a long process decides what the whole post gets called.
Where there’s no tag, the platforms guess, and the guessing is worse. That’s how a sofa pillow gets flagged.
Meta has already been through this once. In 2024 it put a “Made with AI” label on real photographs that had only been retouched, photographers revolted, and Meta renamed the label “AI info.” The name changed. The machine didn’t.
Who decides an appeal of an AI label?
As far as anyone can tell, another machine.
Nobody reads an appeal in under a minute.
That should bother every writer more than the label itself. You can prove the artwork’s origin, name the artist, show the dated files and the invoice, and none of it matters, because nobody is looking. Machines are labeling the work, and machines are hearing the appeals. A system like that can only ever agree with itself, and the person it’s wrong about has no one to tell. That’s a rotten way to treat people who make things for a living.
Is there a way to reach a human at TikTok? Authors in the thread said Reddit sometimes turns up workarounds the official process won’t tell you about. For most writers posting a book video, there’s no path to a person. You can spend a week fighting it, or you can make another video.
An AI label that fires on everything is a smoke detector that goes off for toast. After a while nobody runs, and the label still costs you readers. – Richard LoweShare on X
Was it the AI label or the caption?
Probably the caption.
Several authors in the thread pointed out that the post’s text excerpt described a character beating up his younger brother. Violence words, sexual words and body words get videos pulled on TikTok all the time, and authors who write romance or thrillers learn to disguise them in captions. TikTok allows AI content. It doesn’t allow what its filters read as violence, and a scene from a thriller looks a lot like violence to a filter that can’t tell fiction from a threat.
So the label and the removal may be two different machines making two different mistakes. That doesn’t make it better. It makes it harder to know what to fix.
What does an AI label cost a writer?
Readers, mostly. A label sits on top of the post, and plenty of people who see it scroll past without reading. Some of them were never going to buy, and some were exactly the reader you wanted. You’ll never know which, because the platform doesn’t tell you how many people saw the label and left. You just see the numbers drop and you’re left guessing why.
It also hands ammunition to the people who spend their days hunting for AI. I’ve written about why AI detection can’t be evidence and about the pile-on over restored family photos. A platform label gives that crowd something that looks official. A post stamped “AI-generated” by TikTok reads like proof to someone looking for a fight, even when the machine was wrong.
And the worst part is that nobody can opt out of the judgment. You didn’t choose the label, you can’t remove it, and you can’t explain it to the reader before they’ve already decided.
Why can’t platforms tell AI writing from human writing?
Writing is even harder to label than pictures, because a block of text carries no metadata at all. There’s no tag in a paragraph saying which tool touched it. So a platform that wants to label writing has two choices: ask the writer, or guess from the words.
Guessing from the words is what AI detectors do, and they’re bad at it. They flag formal writing, careful writing and writing by people whose first language isn’t English, and they miss plenty of machine text that’s been lightly edited. I went through what those scores are worth in what an AI detector score on your manuscript is worth. The short version is: not much. A score from one of those tools is a probability dressed up as a finding.
Then there’s the question nobody at the platforms will answer. Spell check uses AI now. So do grammar tools, the suggestions in Word and Google Docs, and the cleanup filters in every photo app. Where’s the line between a writer who used a tool and a post that was “written with AI”? The platforms don’t say, the labels don’t explain, and the writer finds out which side of the line they landed on when the readers stop coming.
Should authors put a “No AI” badge on their book covers?
Some authors have decided to answer the labels with labels of their own. An author in an indie promotion group asked whether readers would like a “No AI” badge on her covers. The replies were close to unanimous: don’t.
The reasons were good ones. A badge is a claim, and anyone can make it, including the people it’s meant to expose. One commenter pointed out that the badge itself looked machine-made. Another had written in her author’s note that a book wasn’t AI, and a reviewer called it AI anyway, author’s note and all. The people who see AI everywhere won’t be talked out of it by a sticker.
Some will read the sticker as a confession.
It also costs something. A cover has a few seconds to sell a book, and a badge spends part of that on an argument most readers aren’t having. Plenty of readers say they don’t care how a book was made, they want a good book. A badge tells those readers the author is worried about something they never thought about.
If you want a mark that means more than your own say-so, the Authors Guild runs a Human Authored certification that registers the book and lets registered authors print its logo. It’s still a promise and not a test. Nobody can test a novel for AI. The platform labels have the same flaw.
What should authors do when a platform labels their work as AI?
Credit the human. If an artist painted your cover, name them and tag them in the caption every time you use the art. Readers who see the credit know who made the art, whatever the label says.
Tell the truth about the tools. If an app animated the cover, say so: the art is by this artist, the animation used an AI tool. Readers forgive a tool. They don’t forgive feeling misled, and the honest caption takes the sting out of the label before anyone can use it against you.
Keep your originals. Save the dated source files, the artist’s invoice and the drafts. You’ll rarely need them for a platform, but you might need them for a reader, a reviewer or a retailer someday.
Check the caption before you blame the label. If a post comes down, look for words a filter would read as violence, sex or self-harm.
Then move on. Make the next video. A week spent arguing with an appeals process nobody is reading is a week you didn’t write.
And put your best work somewhere no platform can stamp it. Your own website doesn’t add labels, doesn’t run a detector over your posts, and doesn’t hide you from readers because a filter got nervous. Social platforms are rented space, and the landlord just started writing warnings on the doors. Post there for reach, but keep the real home where you hold the keys and nobody else decides what your work is called.
The detector that cried wolf
Platforms say the labels are about transparency. A label that can’t tell a painting from an animation filter, or a pillow from a prompt, is a guess printed in a font that looks like a fact.
The place nobody can label your work is your own site. More on the rest of this fight is in the AI and Writing Hub and the AI Derangement Syndrome series.
If you want to use these tools in your work without handing anyone a reason to come after you, look at my AI services.
Keep making toast. Just don’t expect the detector to know the difference.
