Why Publishers Are Adding Invisible Ads for AI Agents

For as long as there have been ads on the internet, marketing has targeted human eyeballs. But as AI agents crowd onto the web, advertisers and publishers are beginning to deliver ads only AI can see, sparking consternation among some of the agents’ developers, who fear marketers could try to sway agents to take actions without their users’ knowledge.

The first publisher to experiment with these types of invisible ads was venerable Time magazine, which continues to generate a significant portion of its revenue from ads on Time.com. Those ads—for Ally Bank and Project Management Institute, a trade group for project managers—first showed up online in August and were developed in a partnership with Mobian, a New York–based ad tech startup. The ads were designed to give agents information about the institute’s expertise in project management and Ally Bank’s online banking services and ATM network.

Following the ads’ debut, Time has had over 30 meetings with brands about doing similar ads for AI agents, and several new advertisers are launching them now, said Time Chief Operating Officer Mark Howard.

Since Time launched its ads for agents, 10 more publishers have started running such ads, said Mobian CEO Jonah Goodhart, who previously sold Moat, a data analytics startup, to Oracle for nearly $1 billion in 2017. Those other publishers are mostly niche sites. One is Daily.dev, a news website for developers, which has signed a one-year revenue-sharing deal with Mobian. Its ads for agents launched around September. Goodhart hopes more than two dozen publishers will be running such ads by December.

These ads for AI agents exist on a separate, text-only website version designed to be easily read by the agents. A site using these ads, like Time.com, identifies an agent as soon as it arrives and serves it the stripped-down version of the site rather than the one human visitors see.

Paul Bannister, chief strategy officer at Raptive, a New York digital ads company, said he is planning to begin pitching ads for AI agents to brands in the next couple of weeks. And over at PMG, roughly a fifth of the Texas-based digital marketing agency’s 60 or so clients plan to run Mobian’s agent ads over the next month, according to a person close to PMG. If those clients go ahead and launch such ads, the agency expects to know by early next year just how successful they are at getting agents to pay attention to those brands.

The past year has brought a frenzy of interest around agentic AI—software that can autonomously carry out tasks on behalf of people. Much of the technology’s development has focused on coding and office tasks, but attention has lately turned to agents people can use for everyday tasks like shopping. In September, Meta Platforms launched its Muse agent, and Meta CEO Mark Zuckerberg recently said shopping is the most popular use of the app. A few weeks after Muse launched, OpenAI debuted a competing tool, Dots. Another competitor is Instinct, from a San Francisco–based startup by the same name that was founded last year and is valued at $10 billion.

The enthusiasm around agentic shopping has drawn a mixed reaction from web retailers. Amazon, for instance, has blocked agents from its site. Other companies, including Walmart and Expedia, have struck partnership deals with Meta to allow it access to their sites.

Agentic shopping is very much in its infancy. While 80% of consumers in the U.S., Japan, Mexico, Brazil and the U.K. use AI as part of their shopping experience, only 7% feel comfortable letting it place orders without their approval, according to a study published in late October by Riskified, a New York firm that sells software to prevent online fraud.

Still, publishers and advertisers are eager to figure out how to push their brands to agents. Many of them have been energized by how AI has overturned the digital ads world, with chatbots replacing internet searches and direct visits to websites, the two main traditional avenues for selling ads online. At the same time, publishers are trying to figure out exactly how to commercialize those ads and how much a brand is willing to pay for ads visible only to agents compared to traditional ads meant for humans.

Anthropic, OpenAI and Google have all held informal discussions with publishers about ads aimed at agents, according to two people familiar with these discussions. In those conversations, the companies have expressed their worry that marketers could include hidden instructions with an ad—perhaps telling an agent to ignore a competitor’s product—and they want the content clearly labeled as an advertisement, these people said. (Anthropic, OpenAI and Google wouldn’t comment for this story.)

Some advertising executives worry they could be criticized for using agent-targeting ads and have drawn a comparison between such ads and cloaking, a tactic in which a website shows search engines different content than it shows people visiting the site in order to manipulate search results. Google has long treated cloaking as a form of spam. The comparison irks Goodheart, the Mobian CEO, who insists that humans and AI see the same advertising content.

Nimrod Kramer, CEO of Daily.dev, thinks brands will self-police themselves and won’t use the agent-targeting ads to deceive consumers and their agents. He believes such deception might lead the companies developing the agents to blacklist those brands, effectively banning them “from one of the most important sources of traffic over the next 10 to 20 years.”

Within two weeks of Time’s launch of its agent ads, Perplexity said it began blocking those ads from influencing its AI agents and search results, calling the practice deceptive.

After Time began its experiment, the Interactive Advertising Bureau, the chief trade group for online publishers and ad agencies, polled publishers and advertisers about their thoughts on the ads. More than half of the respondents said they felt it was “too early to tell,” while more than a quarter said they viewed the ads as deceptive or manipulative.

“The Time experiment struck a nerve,” said John Dawson, vice president of strategy at Jellyfish, a London-based digital marketing company. “We’re at an inflection point where advertisers need to think about messaging to machines in some pretty explicit ways.”

To figure out how much traffic comes from agents, four publishers have signed up to use a new tool from Resemble AI, a startup best known for deepfake detection software used by The Wall Street Journal and The New York Times. None of those publishers has started using the tool yet, which will be released on Thursday, said Resemble CEO Zohaib Ahmed.

The tool identifies whether a website visitor is a person or an agent based on behavior such as mouse movements and clicks. In internal testing, it caught 98.8% of agents designed to look human, Ahmed said. If the websites see a large amount of AI traffic, it could spur them to add their own ads meant for agents, he said.

Right now, publishers that want to include ads for agents aren’t sure exactly when the best moment is to show them to agents. Some ads are shown to an agent when it is carrying out initial shopping research, while others are shown just as it is completing a purchase.

While agent ads remain experimental, publishers are feeling pressured to figure out what will work best whenever they look at the numbers of bots visiting their site: It’s a veritable onslaught.

When Daily.dev’s Kramer first heard about Time’s experiment with such ads, he checked to see how many bots were crawling his site. “I was shocked,” he said. “It was millions and millions.”

Alix Coutures is a reporter covering digital advertising focused on OpenAI, Google and AI startups, based in New York City. She can be reached at [email protected] or on Signal at Alix_coutures.49