AI slop is eating the world

July 25, 2025 Yanac

Privacy-focused DuckDuckGo rolled out a new feature this month that lets users hide AI-generated images in search results. When I posted this news on social networks, my followers lit up, and a consensus quickly formed that this was the most welcome new search feature of the year.

Here’s how it works. When you use DuckDuckGo to search, then click on “Images,” you’ll see a prominent “AI Images: show” menu. Click on it to choose “Hide,” and AI slop will mostly vanish from the results. To make the change permanent, find a “Hide AI-Generated Images” switch in DuckDuckGo’s search settings, or save noai.duckduckgo.com as your main search link.

The ad-free paid-search service Kagi, which costs $5 or $10 per month depending on the plan, also offers AI filtering. (Full disclosure: My son works at Kagi.)

After searching in Kagi “Images,” users should see an “AI Images” drop-down menu with “Any,” “None,” and “Only” options (the “Only” option excludes most non-AI-generated images). Users can manage AI images by opening Settings, going to the Search tab, then AI tab, and toggling “Exclude AI-generated images.” That setting blocks or down-ranks images from sites known to flood the web with machine-made content. As of this month, however, the filter is the default.

You can use operators like “ai:none” or “ai:only” in the search bar. Kagi labels AI images with a small badge.

DuckDuckGo uses open-source blocklists from uBlockOrigin and uBlacklist to identify AI images, while Kagi bases its identification on the source website, flagging images from sites known to produce or post AI images.

Neither catches all AI.

AI-generated ‘user’ reviews kill the value of reviews

AI slop is quickly taking over just about every kind of site that offers content, including supposed user-generated content. Last year, for example, the number of fake user reviews written by AI exploded in app stores and online retailers.

The digital ad fraud detection company DoubleVerify found more than three times as many fake, AI-generated app reviews in 2024 as the year before. Its investigation uncovered that over half of all reviews on a popular streaming app were AI fakes.

Apps with thousands of five-star reviews showed repetitive phrasing, usually a sign of machine-generated text. The problem was not limited to a single platform. These fake reviews appeared across mobile stores and smart TV app platforms, too.

In 2024, a fake AI-generated review cost as little as $2.23. Some AI-written reviews revealed their origin, like when a reviewer wrote “I’m sorry but as an AI language model…,” copying a standard generative AI reply. But most relied on vague or highly positive phrases repeated in dozens or hundreds of posts.

Two things are obvious. First, the number of fake reviews this year will be much higher than last year and keep growing. Second, when (not if) AI-generated fake reviews overwhelm human reviews, user reviews become worthless.

Sites can either kill reviews altogether or weed out fakes. There is no third option.

AI-generated ‘creative’ content sidelines content creators

Content consumers — people who read things, watch videos and listen to music — may one day not care whether people make the stuff of our culture or machines do. But for now, I suspect, they do care.

New songs were uploaded to Spotify this month by singer Blaze Foley. Trouble is, Foley died in 1989.

The songs were AI-generated but remained public for days before being removed after complaints. Foley’s label, Lost Art Records, called the incident “harmful” and said it could have been stopped easily if Spotify wanted to do so. But the only reason Spotify removed them at all is that Foley has fans and a company invested in the singer’s catalog.

Velvet Sundown, an act made entirely by software, crossed one million monthly listeners in July after climbing Spotify’s charts with tracks like “Dust on the Wind.” Velvet Sundown’s debut album came out in June. The band has no members, just a text bio describing them as a “synthetic music project.” Fake bands are rising in the charts on Spotify — made using tools like Suno and Udio, which can churn out new songs in less than a minute from a simple prompt.

American music producer, multi-instrumentalist, songwriter, audio engineer, and YouTube influencer Rick Beato showed on YouTube how easy it is to invent a fake musical artist and churn out songs using AI.

On Deezer, which tracks AI-generated song numbers openly, more than 20,000 new fully AI-generated tracks were uploaded per day by spring 2025, making up more than 18% of all new content. It’s less than 1% on Spotify, which tries to remove AI content, especially when it impersonates the music of real people.

Needless to say, AI-generated songs will continue to increase in both number and sophistication.

Music is a powerful and universal element of human culture that can deeply influence emotions, mental well-being, identity, and personal growth. Who wants to turn this powerful art over to machines?

I think it’s clear that music services need to prepare for a future where AI-generated uploads overwhelm music made by people.

AI slop is everywhere

In 2025, fully AI-generated content is growing fast and reshaping how we get information and entertainment. Entire articles, blog posts, product descriptions, and video scripts come straight from AI models like ChatGPT, Jasper, and Gemini, producing polished text in dozens of languages within seconds.

Video creation can be almost fully automated on platforms like Runway, Synthesia, Pika Labs, and D-ID. Users input a text prompt or script, and the AI builds videos with virtual actors, stock scenes, or animations.

Social media is being overwhelmed with AI-made images and short videos from platforms like MidJourney and DALL-E, creating memes, GIFs, and visuals without any human artist involved.

While banning all AI content is probably undesirable (and impossible), the sites and services that curate or host content need to help users discern the real from the fake, the human-generated from the machine-generated. I think people will demand it.