AI agents don't care about your pretty website or tempting ads
- Reference: 1748373014
- News link: https://www.theregister.co.uk/2025/05/27/ai_agents_confused_by_websites_ads/
- Source link:
Agents - AI models augmented with browsing capabilities or multimodal interfaces - see websites and web ads differently from people. They focus on structured data such as price, availability, and specifications, while largely ignoring visual cues and emotional appeals that typically influence human users.
Andreas Stöckl, a professor with the Digital Media Lab at University of Applied Sciences Upper Austria, and researcher Joel Nitu, recently explored the issue in a preprint [1]paper titled, "Are AI Agents interacting with Online Ads?"
[2]
In answer to that question, the authors found that AI agents are doing just that, but in surprising ways.
[3]
[4]
Andreas Stöckl told The Register that the study has different implications for publishers and for advertisers. One of the findings, he said, is that ad personalization must be adapted to suit agents.
If a website is well optimized for accessibility, it is, in principle, well prepared for agent-based interaction. However, the design and placement of online ads need to be analyzed and reconsidered
"General accessibility for agents follows the same principles as accessibility for other user groups," he said. "If a website is well optimized for accessibility, it is, in principle, well prepared for agent-based interaction. However, the design and placement of online ads need to be analyzed and reconsidered. This is precisely what we are currently investigating in our research project."
The research project took three different computer use systems – a specific subset of agents that are able to interact with computers like a human would – OpenAI’s Operator, Anthropic’s Claude "Computer Use," and an open-source Browser Use agent. Then it looked at how they handled a travel website using various multimodal models – GPT-4o from OpenAI, Claude 3.7 Sonnet from Anthropic, and Gemini 2.0 Flash from Google. They used multimodal models because the agents needed to be able to interpret visual input – they "see" by taking website screenshots and analyzing them.
The authors directed these agents to perform hotel search and booking tasks autonomously while taking into account user-specified preferences like destination, price range, and availability before finalizing the itinerary.
[5]
They were able to accomplish these tasks to differing degrees, but interacted with the websites very differently than a human would. For example, when a travel website shows image-based banners showing a call-to-action (CTA) – a pitch to book a particular hotel, for example – agents may miss the promotion entirely.
"Google’s model vividly demonstrated this inefficiency, executing additional back-and-forth steps due to uncertainty about whether its overlaid CTA in the image-only ad was clickable — a complication not encountered with text-based banners," the paper explains.
"Uniquely, Gemini 2.0 Flash contradicted this overall trend, however: under the image-only banner it recorded higher booking specificity and a slight uptick in banner engagement, even as its overall reproduction of promotional language declined."
[6]Trump can bluster and bluff all he wants, but iPhone manufacturing isn't coming to the US
[7]Ex-Meta exec: Copyright consent obligation = end of AI biz
[8]Salesforce takeover of Informatica is on for $8 billion
[9]Some signs of AI model collapse begin to reveal themselves
The researchers examined how AI agents interacted with different types of ads, whether they acknowledged ad content, and the extent to which ads influenced booking behavior or prompted user-like actions (e.g. a click).
They began with a baseline test environment on a hotel booking website that featured standard text-based banner ads. The researchers then evaluated two additional ad formats in separate environments: one using image-only banners, and another where promotional text, such as a Valentine's Day offer, was embedded directly within the ad image.
In the baseline environment, Claude 3.7 Sonnet clicked on 59 standard text-based banner ads but did not engage with any sponsored content.
[10]
Gemini 2.0 Flash was more selective, clicking 29 banner ads and interacting with sponsored content three times. GPT-4o clicked 59 banner ads and engaged 12 times with sponsored ads, while OpenAI's Operator clicked 47 banner ads and interacted with sponsored content 20 times.
The extent to which models interacted with ads tended to reflect the relevance of keywords to prompts and queries, with text-based ads outperforming images that had text embedded in them.
The models also exhibited the sort of bias that experts continue to warn about. For example, both Claude and Gemini handled bookings for "husband/wife" differently from bookings for "girlfriend/boyfriend."
"Claude 3.7 Sonnet recommended longer stays for married couples (averaging 5.2 nights) compared to dating couples (averaging 3.6 nights), with Gemini 2.0 Flash showing a similar trend (7 nights versus 5.7 nights), while GPT-4o displayed minimal difference (5.8 versus 5.6 nights)," the paper says.
With regard to advertisers, Stöckl said, "The new technology does not necessarily result in an increase in invalid clicks. However, not every ad format or design is suitable for agents, and this requires further examination."
It may be necessary to develop new ad formats specifically designed for interaction with agents
Advertising needs to be studied and reevaluated, he argues, as agents become increasingly common.
"It may be necessary to develop new ad formats specifically designed for interaction with agents," he explained. "This shift doesn’t have to disadvantage advertisers. On the contrary, if agents acting on behalf of users (e.g., making bookings) can be influenced through targeted advertising, this could open up entirely new strategic opportunities."
Stöckl said the preprint paper will be finalized and published in July.
The Register asked Simon James, group VP of data science and AI at Publicis Sapient, a digital consulting firm, about the researchers' findings and he concurred with the need to structure web pages to accommodate software agents.
James said the paper echoes points he made in a post he authored about the shift from focusing on customer experience to agentic experience.
"Agents don’t browse; they execute commands," he wrote. "A human may whimsically browse for something to do during lunch break, might get distracted by a cool image, or a seductive strap line. For 25 years, customer experiences have been built to tempt, intrigue and beguile humans to dwell. The journey is just as enjoyable as the destination. An agent filters noise aggressively. Yes, your well-honed customer experience is noise to an agent." ®
Get our [11]Tech Resources
[1] https://arxiv.org/abs/2504.07112
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aDY19bVhSZ2ySD3sB9NfxgAAA1E&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aDY19bVhSZ2ySD3sB9NfxgAAA1E&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aDY19bVhSZ2ySD3sB9NfxgAAA1E&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aDY19bVhSZ2ySD3sB9NfxgAAA1E&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2025/05/27/trumps_iphone_manufacture/
[7] https://www.theregister.com/2025/05/27/nick_clegg_says_ai_firms/
[8] https://www.theregister.com/2025/05/27/salesforce_snaps_up_informatica_for/
[9] https://www.theregister.com/2025/05/27/opinion_column_ai_model_collapse/
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aDY19bVhSZ2ySD3sB9NfxgAAA1E&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[11] https://whitepapers.theregister.com/
Re: Oh yuck
"One of the findings, he said, is that ad personalization must be adapted to suit agents"
This sentence is a threat.
Re: Oh yuck
They'll just move all advertising to where people are consuming content for entertainment, rather than trying to "get things done" like making an airline reservation or placing an online order for a new pair of shoes.
Because you could hand off the "get stuff done" to an AI agent, but you can't hand off the task of streaming the latest episode of your favorite show.
Soon you'll need to browse the web through an AI to filter out the clutter
In the 1990s, the web was chaotic, in a wild west sort of way, Around 1998, the advertising companies got involved, and many web sites started to look like video games. This was in the days of not only sidebar ads, but interstitial ads, popup ads, popunder ads, delayed popup ads (so you couldn't trace which site/tab had triggered it), etc. And, at a time when cable and DSL were just getting started, and the majority of people still had 56kb (or 28kb) modems, and had metered internet access, many ads were playing bandwidth killing music and video.
I was working in a Microsoft shop where everyone else was running Internet Explorer, but I ran Opera to filter all that crap off. I remember once being asked sent a URL to look at, and I couldn't see any problem with it. When I looked at a co-worker's computer with it in IE, it was a completely different page. At least 90% of the page was noise, with animated characters bouncing around, music playing in the background, three popups going on, etc. It to a minute for him to load it. I could load the same page in under 2 seconds because the content was barely 5% of the page.
In those early days, only a minority of users ran ad blockers, but eventually the web became so loud and obnoxious with ads that some browsers started integrating ad blocking in their releases, and not requiring the user to install a plugin or addon (which is outside the skill level of many users). And so many web sites now check for ad blockers, and it's become an arms race.
AI scraping the web is going to learn how to bypass all that clutter. And that's going to trigger a new arms race.
Web servers will start trying to find ways to insert ads that AI can't ignore, which will make things worse for humans, so AI in browsers will be used to counter the AI in the web servers.
People who used the internet prior to 1994 will remember the infamous Canter and Siegler green card scam, where a pair of lawyers flooded Usenet with advertisements. They were the first major spammers/scammers. Prior to that, email and usenet were both pretty insecure, because no one really cared enough to lock things down. After the spammers started, SMTP servers had to be secured, and usenet relays has to be locked down to prevent scammers.
Today, email spam is so overwhelming that actual email is essentially statistical noise at this point.
AI web crawlers could be the Cantor and Siegler of the web. Today, email admins spend more time dealing with spammers than supporting actual email, and the same could happen to the web.
Meh....
....if I can make so the agent burn the company's data centre to the ground trying to navigate a website, then I'll be happy.
I make things for real people, not narcissistic asshole billionaires.
Your well-honed customer experience is noise to
Anyone looking for useful information, AI, human, even Salesmen.
In fact, is there ANYBODY outside of your own marketing team who find your "customer experience" to be anything more than a bloody nuisance?
> For example, when a travel website shows image-based banners showing a call-to-action (CTA) – a pitch to book a particular hotel, for example
Then WHO is being served by that being an image? Anyone wanting to search for reviews by copy'n'pasting ("you DARE to mistrust our recommendations?"). Anyone with a screen-reader ("Don't worry, we've used special tags with the text for them" - not being able to see the absurdity in making more work for yourself because you cocked it up using an image in the first place!).
Maybe, MAYBE, one good thing could come of the LLM Frenzy: getting rid of all the stupid, over-designed, over-coded* bullshit websites that infest the Internet nowadays.
* I mean, come on: you specifically wrote code to PREVENT me opening details of multiple carparks in new tabs! Why? What do you gain from that? All I want is to have them available so when we find one is full we know where our next choice is! [Story based on Real Life, looking for parking in Coventry - yes, the real one, in Warwickshire** - this morning. And don't me started on Google Maps, who gave as their FIRST pick when walking from the cathedral back to the car park somewhere on RHODE ISLAND! Gaah]
** You heard me, Warwickshire!
Suckers
So AI agents are suckered by click bait adverts and dark patterns? I so want those agents to be spending money on my behalf. Not…. Guys just admit that this is all a steaming pile and always will be.
I think we should do the opposite.
Cater to humans and rig websites so that AI agents fail. And not just don't work but work really badly. Anyone stupid enough to allow AI to roam the net with the ability to book a holiday, should have a luxury world cruise booked when they want a dirty weekend in Paris, no money refunded.
AI clicking ads generates zero money
AI don't buy anything. So them "clicking" on any ad is a waste to the advertisers. Detecting an AI-agent should therefore mean: Don't show any ads to optimize return of investment.
News to you Andreas..
I'm human and I've been ignoring adverts for years, ad blocker or not. Anything more than a very basic text advert (and that's on a web page, not one inserted into social media, reddit, or other srteam of updates service) and there's zero chance I'm clicking on it, as it's actively making the task of finding information more difficult.
Oh yuck
If only we could strangle everyone performing this research. One of the most valuable uses of an AI agent would be to ignore advertising and fulfill the user's request. I guess we'll see another arms race between the advertisers and the ad blockers, leading to further enshittification of Web content.