Tech giants are trying to clean up the internet of AI junk content

The worst kind of AI junk is already flooding Google searches with glue pizza recipes, stuffing Amazon product listings with fake biographies, and flooding Facebook feeds with pictures of Prawn Jesus.

This digital cesspool has spread across tech platforms, cheapening meaningful content and polluting the information ecosystem, while chatbots and video generators continue to advance.

Now, Silicon Valley wants to clean up the mess.

Music streaming platform Spotify said it removed 75 million bulk uploads, duplicate songs and other "junk" music last year - a sizable portion of its total library. LinkedIn said in July that "Tackling AI spam is a top priority for all of us," and developed a button for users to report it. Researchers at YouTube parent company Google described how the video platform purged 130,000 channels with poor-quality content in six months.

Across the internet, tech companies are engaged in a generative waste cleanup, and in many cases they have been inundated with the same junk content that their AI tools spawned in the first place. Mobile messaging apps, restaurant review sites, academic archives and dating services are all eradicating or degrading AI-generated spam — many of them relying on the same technology to accomplish this cleanup.

People are already tired of AI’s impact on the environment, employment and mental health. Aidan Walker, a researcher and content creator at the Carnegie Endowment for International Peace, a global affairs think tank, said that AI spam content is just another "festering wound."

"Users are gradually becoming more and more dissatisfied with this," he said. "The atmosphere of public opinion has changed significantly, so these companies are finally starting to face these problems."

Generative AI can produce increasingly realistic images, audio, and videos, but only some of them are spam. Tech companies are mostly targeting the kind of low-quality clickbait that’s easy to make, spread, consume, and forget about — just as fast fashion has squeezed out haute couture and pink meat scraps have replaced steak.

If left unchecked, the online synthetic swamp could "crowd out human creativity, corrode cultural values, and undermine public discourse," European researchers wrote last year.

AI spam has become so common that dictionary publisher Merriam-Webster selected it as its word of the year in 2025. French streaming service Deezer said this spring that nearly half of the tracks updated to its system every day were generated, many of which had never been listened to by users. Among the content uploaded to Apple Music, AI-generated music accounts for more than one-third.

Instagram head Adam Mosseri said late last year that tracking "real media will be more practical than tracking fake media." His observation comes shortly after Meta (the parent company of Instagram) launched an all-AI video stream that many derided as a showcase for spammy content.

Many of the tech companies now trying to manage AI spam are the same companies that facilitated it. In July, Meta pledged to support a European agreement to make AI-generated content more identifiable through detection, labeling and annotation. That same month, the company also released an image generator that automatically uses public Instagram accounts as a reference for the tool. (After three days of public backlash, Meta removed the feature.)

“AI spam is a double-edged sword because many platforms get a boost in engagement because of it,” said Paul Shovlin, an AI expert at Ohio University. “But then it becomes a burden because there’s so much spam that it’s now difficult for users to find the kind of quality content they want.”

Google, for example, has gone all-in on AI, connecting its Gemini model with email, mapping and travel services, and search. According to an experiment last year by online video editor Kapwing, more than 20% of the initial short videos pushed to new users on YouTube were AI-generated spam.

Kapwing estimates that one of the most popular YouTube spam channels earns $4.25 million per year. (The channel has since been suspended from YouTube’s monetization program.) YouTube CEO Neil Mohan wrote in a blog post that “managing AI spam” is a top priority this year.

Recently, researchers at Google considered this challenge. Rather than evaluating videos individually, they looked at upload speeds, duplicate templates and other patterns that could indicate coordinated abuse. They found 50,000 account groups and removed them. However, some digital media experts worry this approach could go too far and penalize legitimate creators who publish large amounts of content.

"While academic research helps advance industry understanding, it does not reflect the many layers of defenses we use to enforce our policies," YouTube said in a statement. "This includes strong safeguards to ensure that well-intentioned creators using AI tools are not compromised."

Other companies are turning to various strategies to combat AI spam, including mobilizing their own users to help identify it. In October, Pinterest updated its system to allow users to limit the number of AI-generated posts they see. TikTok said it is testing a feature in the United States that allows users to adjust the amount of AI content they see. (The company also says it has automatically flagged more than 3 billion AI-generated videos.) TikTok removed more than 377,000 videos for violating its AI policies in the first three months of this year.

Nearly half of posts with more than 50 words on Elon Musk’s social media platform X were created using AI, according to data from AI detection startup Pangram. Nikita Beer, who stepped down as head of

Substack CEO Chris Best said last month that "it's increasingly difficult to tell what's real on the internet," and that "platforms that reward false content will lead to a race to the bottom." The messaging platform has announced the launch of its own AI detection tool for users, powered by Pangram, which is designed to scan replies, comments and posts longer than 100 words and determine whether they may have been written by humans or with the help of AI.