On matters of agentic traffic and declining human readership across the open internet

When traveling with a group of friends to Copenhagen this Fall, for the first time, no one cited Fodor’s, Lonely Planet or even Rick Steves. “I asked Chat” was the common refrain throughout conversation after conversation about what to see and do each day. Even influencer content from YouTube served only to inform follow-up prompts to the LLMs to verify the recommendations and work them into a proposed itinerary.

Are dogs allowed in the Royal Library? Is that smørrebrød place open on Mondays? You see, these elder millennials, despite a decade of attending destination weddings and carrying out adjacent trips to foreign cities, had started using Chat GPT, Gemini and Claude at work over the last year. They had formed new behaviors; this was how they “searched it up” now.

Even the most enterprising of them, an old-school spreadsheet maker, found the AI tools irresistible on some level. Today, when a sea of financial documents needs to be parsed and organized for a job like theirs, the norm has become that their agent handles the legwork. Their preferences are in their claude.md. It has become second nature to open their chat app and start a query. 

“Organize these recommendations, rank them with the ones my friend sent me.”

A helpful itinerary gets produced in the chat UI, along with transit details and reference images. In the model, the traditional opportunity for, say, a credit card company to run a brand awareness campaign targeting high-household income travelers on a list of the top restaurants in Copenhagen is gone.

But those travel websites are still likely to rack up visits from the chat-based queries — just not from human readers. When your agent is “searching the web” for you, they’re also causing ads and rich media on the website to load, setting off tripwires and alarm bells all around the programmatic advertising ecosystem in the form of “Bot-” and “Invalid Traffic” measurements.

These same agents may very well be the ones to recommend a cheap flight route or the best credit card for international travel. But for now, advertisers are still of the mind to claw back ad spend from the publisher who inadvertently found that their readers are now robots.

Publishers used to think of this as “personalized content,” meaning a different homepage layout or dynamic stories recommended at a few break points in a longer article. Now the dynamic aspect that a website operator might want to spend more resources on is determining if display advertising is appropriate at all for a given visitor. An easily-bot-readable form of your article could be entirely more appropriate if you know for sure it’s an agentic visitor, and that you’re not going to miss out on the fleeting chance to have a monetized relationship with a reader.

Perhaps you don’t load your traditional advertising scripts at all if you can tell a reader is agentic, which is a moving target that likely keeps evolving in complexity and reliability to address — even for the likes of Cloudflare, TollBit, Fastly and other content delivery networks (CDNs).

But this is easier said than done when systems like Cloudflare Bot Management say one thing and a third-party measurement paid for by the advertiser could give a different reading, which is a relationship risk between media owners and those that buy their advertising inventory programmatically.

While it’s still being pioneered and will take more of both innovative publishers and experimental advertisers to figure out, a key strategy for ongoing monetization of the open internet could be in embedding sponsored content for LLM consumption into those for-bot versions of articles, typically in markdown or stripped-down HTML, much like the gray box Sponsored stories of today. To get an advertiser’s soundbite into an agentic write-up for a search query, referral link be damned, is a content marketing service.

I get the Chase Sapphire quarterly magazine in print in the mail, and in a series of photos and essays, it makes recommendations for travel destinations around the world that are in vogue this year. A similar list could be repeatedly ingested by crawlers in repetition on all sorts of websites and could end up influencing the chat outputs as effectively as predecessor marketing efforts. The real question is not about the feasibility of such influence, or about the risks of weaponizing such potential for influence, but about how much publishers will ever be paid for any of this. Their continued viability depends on the answer.

This op-ed represents the views and opinions of the author and not of The Current, a division of The Trade Desk, or The Trade Desk. The appearance of the op-ed on The Current does not constitute an endorsement by The Current or The Trade Desk.