The question in the topic title is the wrong shape. AI already writes posts, cuts videos, and fills a Shorts feed. That is not the same as replacing the people who still have to be believed.
The useful split in 2026 is four jobs: produce the artefact, verify that it is true, distribute it into a feed, and stand behind it when it is wrong. Mix those jobs and you get a slide that says "AI will replace creators." Keep them separate and you can ask who still gets paid.
This post is a map of those jobs. It is not a eulogy for YouTube, and it is not a product roundup.
What "replace" actually means on the internet
Three different systems get lumped together.
1. Production at unit cost near zero. Text, stills, voice, and short video can be emitted without a crew. That is real. It is also the part that was already cheap: recaps, listicles, thumbnails, "what happened today" voiceovers, faceless explainers.
2. Distribution that rewards volume. Feeds do not pay for authorship. They pay for the next swipe. A model that can publish fifty variants of the same claim will beat a person who publishes one, if the job is only to occupy the slot.
3. Trust and liability. Someone still has to be the name when the clip is fake, the quote is invented, or the "news" channel is a farm. Platforms have started to label and demonetise the worst of this. Labels are not a replacement for a reporter. They are a warning sticker on a feed.
If you mix those three, you get hype. If you keep them separate, you can ask a better question: did this system change what got made, what got believed, or only what got shown?
The feed is already synthetic in places you can measure
Kapwing's YouTube test is the honest number, not a vibe. On a new account, 104 of the first 500 Shorts (21%) were classified as AI-generated. Another 165 (33%) were "brainrot" — low-effort loops whether or not a model made them. That is a new-user feed, not "all of YouTube." Treat it as a density sample.
The same lab looked at popular channels: 278 in a top-100-per-country sample uploaded only AI slop. Combined reach in that slice: on the order of 63 billion views and 221 million subscribers, with estimated annual revenue around $117 million. Those are Kapwing's estimates, not YouTube's books. The direction is still the point: the cheap video job scaled.
TikTok is worse on a cold start. Kapwing's June 2026 pass found 294 of 500 (59%) For You videos on a fresh account classified as AI slop. Kids' content was the worst category they scored (57%). That is not "creators were replaced." That is a distribution system that will take whatever fills the slot.
YouTube's response has been policy, not a ban. It has taken down inauthentic channels, started labeling video that is meaningfully AI-altered, and in July 2026 tightened monetisation on repetitive, rage-bait, and AI-character spam. Neal Mohan called managing slop a 2026 priority while the company ships its own generation tools. Do not read that as "AI video is over." Read it as: the platform will host the cheap job, and it will try not to pay for the worst of it.
Newsrooms did not get replaced. Commodity news got cheaper.
The Reuters Institute Digital News Report 2026 (16 June, 48 markets) is the grown-up dataset.
For the first time globally, social media and video networks (54%) beat publishers' own sites and apps (51%) as a way of getting online news. 77% watch online news video in a given week. AI chatbots as a news source moved from 7% to 10% in a year — growing, still small, and (per Edelman's read of the same report) users click through to originals less than they did from search.
Creators are in that stack. About 27% say they get news from creators. Only 3% rely on them as the only source. Creators win on ease and tone. They lose on trust. That is a job split, not an extinction.
What newsrooms are actually doing with AI is closer to triage than replacement: keep the work a chatbot cannot do (investigation, presence, a name that can be sued), and stop over-producing the recap a model can emit by 7 a.m. The danger Agnes Stenbom Swedling flagged at the Reuters Institute in May 2026 is not "a model in the CMS." It is an industrial value system where speed outweighs public value until the human is a proofreader for a machine that already chose the story.
Search is the other squeeze. Reuters Institute analysis in 2026 put a one-third drop in Google search traffic to publishers in the year to November 2025, with executives expecting more. If the answer is assembled in a chatbot, the page that did the reporting is optional. That replaces a referral, not a correspondent.
Four jobs, keep them separate
A short test for any "AI will replace creators" pitch:
1. Produce. Can a model emit the artefact cheaper than a person? For recaps, thumbnails, B-roll, and first drafts: yes. For a visual investigation, a comedy voice that is actually yours, or a source who will only talk to a named reporter: no. 2. Verify. Can it tell you the plant never existed, the quote was never said, the face is a composite? Mostly no. This is why "real actors, AI script" farms showed up after platforms started labeling obvious avatars — the tell moved. 3. Distribute. Can it occupy the slot? Yes, at a volume no newsroom can match. That is a feed problem, not a talent problem. 4. Stand behind it. When the clip is wrong, who answers the email? A channel with 221 million slop-subscribers in a sample is not a newsroom. It is inventory.
The software lesson is the same as DRS in sport: a review budget on irreversible publishes, and a human who is allowed to spend it. Fully automatic is a demo. A labeled farm is still a farm.
What to do with this if you make things
If you are a creator: do not compete on unit cost with a model. Compete on the job that still needs a name — a beat, a face people can recognise as not-a-template, a correction, a source. Use the model for the cheap jobs you already hated.
If you run a publisher: stop paying senior rates for commodity recap. Keep paying for the work a chatbot will not be sued for. If your CMS can emit 40 local variants overnight, you still need someone who knows which variant is false.
If you build software: log the job (the claim, the source, the publish), not the vibe. Put a review budget on anything that can ship to a stranger. Do not call a voiceover farm "a newsroom."
What not to do
Do not say "AI replaced creators" because a Shorts feed is 21% synthetic on a cold start. Do not say "YouTube banned AI." Do not treat a chatbot summary as a citation. Do not pad a like list, and do not pad a byline.
The internet's content creators were never one job. AI took the cheap ones. The expensive ones got more expensive, because they now have to prove they are not the feed.
No comments:
Post a Comment