The influx of AI agents is quickly creating new ways to autonomously browse the internet and shop online. However, the feature also poses potential challenges for publishers, advertisers, and e-commerce companies — including new problems with how to deal with bot traffic.
Last week, Google debuted a range of new features as part of its release of Gemini 2.0, including a preview of a new agent called Project Mariner that offers to help people with everything from researching and booking trips to shopping for a range of other products.
One demo developed in collaboration with Etsy showed Project Mariner helping to research and buy paint supplies based on the kind of art someone would want to create. Google also pointed out it wouldn’t purchase products without first getting human approval and that it can’t autonomously browse in the background. Although Mariner was designed as a Chrome browser feature, that could change depending on the outcome of its search antitrust trial.
For now, the tool is only a prototype with limited access for beta partners. However, it joins a recent wave of recent AI agent announcements. Last month, Anthropic announced web browsing capabilities to its Claude agent. OpenAI also is reportedly developing a new AI browser project set to launch next year. Yet another is the company behind the internet browser Arc, which is reportedly working on an AI-focused browser to debut in 2025.
While the new AI web browsing tools are still in their infancy, some people already see how it could create issues with how websites identify the bots. Will Google have a new way to self-disclose when a bot visited a website on behalf of a user, and will that ID be different from a person associated with the bot? That could create new problems not just for how publishers measure web traffic. (Researchers have found bots are also getting much better at faking bot-detection tools like CAPTCHA.)
As it gets easier for regular people to run bots with a few prompts, one bot fraud expert said tools for AI agents are making it easier to automate a real browser and create new threats around bot fraud. That could further exacerbate ongoing issues publishers and advertisers are facing with ad fraud and web traffic that leads to lost revenue for media companies and wasted media spend for brands.
Marketers don’t yet seem to fully realize how AI agents will impact their business, for better or for worse, said Amy Holtzer, CMO of the cybersecurity firm CHEQ. However, she said some forward-thinking customers are asking about whitelisting to ensure they don’t miss out on shoppers if they automatically block some types of well-intentioned bots.
“There’s going to have to be a way to resolve all this around identity,” Holtzer said. “If an agent is acting on your behalf, it should carry the same data-privacy preferences as the person using it. And it’s really complicated if you’re not aware of what is identifying you as an individual.”
Other questions also are still unanswered. It’s still unclear what this could mean for retargeting ads. What will identify resolution look like if people use multiple AI agents from various companies, and how will changes affect ad measurement, re-targeting and user privacy?
Autonomous agents also pose questions for transparency around knowing which websites agents visit during a search. That also could pose challenges as companies explore how they can be prioritized by AI agents. Perplexity, which recently debuted its AI shopping assistant, is looking to address this with a new feature announced last week that lets the user choose the websites it’d like to have searched for different queries.
Retailers and publishers will need to rethink their approach to first-party data and transform it into actionable insights through AI agents, said Todd Parsons, chief product officer at Criteo. Instead of relying on banners or video ads, Parsons thinks the future of advertising may lie more in metadata-driven insights that inform and guide consumer decisions in more meaningful ways. That would also move traditional advertising away from push-and-pull models and toward curated, data-driven systems that leverage large language models.
The ability for agents to have multimodal capabilities that understand text and visuals could create new ways for helping people customize what they’re looking for based on personal preferences and the aesthetics of their products. It might also lead to e-commerce companies thinking differently about how they display different products in order to appeal not just to humans but also to bots.
“That is literally nirvana for shopping,” Parsons said. “If advertising changes to serve that multimodality, we’re at a probable future for advertising. But it does assume we’re going to be able to progress past a lot of well wrought workflows and capabilities and technical capabilities pretty fast. And so it’s an exciting time if you think about adoption going quickly.”
The challenge is creating systems that help advertisers and users in a privacy-conscious way while also providing sustainable revenue models, Parsons and others note. Data for product discovery could help with the buyer journey, but that could get harder with direct-response use cases, which is where privacy is often most sensitive. Parsons also wonders if prompt engineering is the right way to tell an AI agent what to discover — or if there’s a better way. He also questioned what AI agents will mean for data collected and shared by Chrome. Will it be trusted by publishers, advertisers and users?
Prompts and Products — AI news and announcements
- AI personalization played a bigger role in this year’s Black Friday and Cyber Monday.
- Omnicom and IPG acquisition could lead to bigger AI investments — and maybe rewards.
- Google announced a number of features including its Gemini 2.0, new features for NotebookLM, a new XR headset in collaboration with Samsung, and updates for its previously announced Project Astra.
- OpenAI continued its “12 Days of Shipmas” with the official launch of Sora and other features.
- Reddit announced a new way to generate answers powered by generative AI.
- Grok, an AI platform owned by xAI, debuted a new AI image generator.
- WPP and Universal Music Group are partnering to develop dynamic music and entertainment solutions for global brands.
- Apple released iOS 18.2 with new Apple Intelligence features along with ChatGPT integrations for Siri and Writing Tools. It also announced the best apps of 2024 including several apps with AI features.
- Paul McCartney spoke out against AI and endorsed laws to protect musicians and creatives from being victims of copyright theft.
- Analyst Benedict Evans released his annual list of predictions for trends that he thinks will take place in 2025.
- Publishers including Civic News and Time released new editorial tools to inform and engage with readers in new ways with AI.