AI slop websites rise fivefold with GEO optimization

Data: DoubleVerify; Chart: Erin Davis/Axios

AI "slop" sites, or domains with low-quality AI-generated content, rose more than fivefold in the second quarter of this year compared with the same period a year prior, according to new data from ad measurement firm DoubleVerify. Traffic to those types of sites quadrupled in that time.

Why it matters: SEO's incentive problems have been exacerbated in the AI answer engine era with brands and publishers competing to influence what chatbots recommend.

As AI citations become more valuable, companies have an incentive to optimize content for machine consumption, which low-cost content mills could exploit.

By the numbers: Ad requests associated with low-quality AI-generated content surged 354% between January and September, per DoubleVerify.

There were more than 3.2 billion ad requests tied to AI slop webpages in the first nine months of 2026.

Zoom in: Generative AI is making it easier to operate content farms designed to capture advertising revenue.

DoubleVerify identified a network of more than 200 made-for-advertising websites filled with AI-generated articles. Some pages featured more than 50 slides, each packed with as many as eight ad banners that refreshed every few seconds, CMO Dan Slivjanovski tells Axios.

The surge in activity toward the end of the year resembles historical patterns of advertising fraud as operators seek to capitalize on rising holiday ad spending, he adds.

Zoom out: AI platforms are moving to establish foundational infrastructure to prevent spam, scams and misinformation from flooding their chatbots.

Google has teamed with Shopify and a slew of online retailers to develop a Universal Commerce Protocol standard for AI shopping.

OpenAI and Stripe have built an Agentic Commerce Protocol to secure checkouts.

Microsoft has piloted a premium content marketplace for publishers. Amazon is eyeing a similar effort.

Yes, but: Those partnerships may establish guardrails for large portions of the transactional web, but the universe of content created to influence large language models is growing rapidly as bad actors use AI to manipulate consumers.

Earlier this year, a marketer Deana Burke went viral for showing how easily LLMs are manipulated. She spent $11.25 to create a website designed to promote a fake deodorant brand to AI agents that ultimately recommended it over real, established products.

"There is a problem with stuff that's created just for bots," said Joe Marchese, an investor and author of "Human Dot AI," a forthcoming book about the AI attention arms race. "How can you have a trillion-dollar industry that can be fooled by an $11 website?"

The big picture: The rush to appear in AI answers risks re-creating the content arms race that plagued traditional search.

Kantar CEO Paul Zwillenberg says brands often initially focus on producing more content, but the marketing data and analytics company advises clients to prioritize quality and relevance.

"It's not about volume. It's about value," Zwillenberg says. "It's about the quality of the interaction."

Between the lines: Brands must optimize for multiple AI models that weigh sources differently.

Appearing more frequently also doesn't equate to appearing more favorably.

In a snapshot analysis of AI-generated airline recommendations, Kantar found that one airline appeared more often but was discussed less favorably than a competitor with fewer mentions.

Reality check: The extent to which low-quality content is making its way into AI-generated answers is unclear.

DoubleVerify's figures measure ad requests associated with low-quality AI-generated pages, not whether those pages are influencing AI recommendations.

What to watch: Medium offers one warning sign.

CEO Tony Stubblebine says roughly half of the 25 Medium stories most frequently surfaced in Google's AI search summaries had been excluded from Medium's own recommendation algorithm.

"That stuff is like the most generic content mill things you could imagine," Stubblebine tells Axios.