Agentic AI: what should brand owners’ trademark teams be thinking about?

Everyone is talking about agentic AI, but what impact will it have on brands? This article looks at some trademark questions arising from the growth of agentic AI and identifies topics for brand owners and technology providers to plan for.

An agentic AI system uses language models to carry out a task, acting flexibly, learning from experience, and making decisions with limited human intervention. For example, a shopping agent is an AI system that carries out some or all of a purchase on a consumer’s behalf, triggered by an everyday-language request the person types or speaks. The consumer might type: “Find me the best-reviewed waterproof running shoes under £120 and order a pair in size 9.” The agent carries out the task independently and is able to query product databases, consider price and reviews, choose a product, and finalise the purchase without the consumer ever browsing.

Thinking forward to a world where significant numbers of day-to-day purchases are occurring in a fully automated agent-to-agent setting, what does that mean for brands? Some brand owners have expressed concern that brand reputation might become less relevant, particularly in lower-value transactions. For example, if an AI purchasing agent is acting on instructions to “On the first of the month, buy the cheapest 2kg box of non-bio washing powder available with free UK delivery”, what role are brands playing? There are competing schools of thought on this question: it would seem plausible that only a consumer who is already brand-agnostic will ever use a brand-agnostic prompt.

But if the prompt given to a purchasing agent does contain a brand, what happens if a similarly branded product (e.g., a dupe) arrives on the doorstep instead? The ‘average consumer’ here is likely the person supplying the prompt rather than the shopping agent itself, but in that case are the usual characteristics attributed to such a consumer (‘reasonably well informed and reasonably observant and circumspect’) still apt? In any event, confusion (by a human) may only arise once the product is delivered.

Following the Iconix case in the Supreme Court, we know that in the UK post-sale confusion can be actionable in a trademark infringement claim, but what about other jurisdictions? And what if that product is in fact a counterfeit? With both dupes and counterfeits, the brand owner’s best recourse is almost certainly going to be tackling the websites at source, and the agentic AI transaction is an additional way that such problems will first come to a brand owner’s attention.

With ever-increasing volumes of internet search traffic flowing through AI-powered search, and increasing numbers of internet users using generative AI (GenAI) to interact with online content, it has become important for brands to ensure they remain discoverable and visible to GenAI systems. This is even more important where agentic AI systems are researching and making purchasing decisions based on what they can access – and transact with – online. Generative engine optimisation (GEO) is already being used by many brands as a legitimate way of structuring digital content and managing online presence to improve visibility in responses generated by GenAI systems. (It is reminiscent of search engine optimisation, which is structuring such content to boost visibility in orthodox online search requests.)

However, new ways of communicating with consumers tend to bring new possibilities for bad actors to exploit well-known brands for their own benefit. Agentic AI is no exception. It is still early days, and no doubt such bad-faith strategies are still developing, but brand owners should be astute to attempts by counterfeiters or dupe sellers to free-ride on their brand (e.g., manipulating chatbots) to drive GenAI engagement or agentic AI transactions. We know from Google France and the cases that followed that machine-only uses of a trademark (e.g., a party bidding on a keyword with a search engine) can still infringe.

Through regular monitoring, brand owners can check whether illegitimate websites are being surfaced or interacted with in response to good-faith GenAI search queries and instructions given to agentic AI. Search and agentic AI providers will, of course, themselves also be astute to bad-faith attempts to influence outcomes, just as, in the past, search engine operators identified and restricted ‘keyword stuffing practices for orthodox search.

In terms of competitor-on-competitor activity, new kinds of tools have appeared that let a brand check how often it is mentioned by AI systems compared to its competitors. This has led some brands to write direct comparison pages (e.g., Brand A versus Brand B), because this is the type of content AI systems most often pull from when someone asks for a comparison between two products. Often, these comparisons include statistics, as part of GEO techniques, and they can be amplified by dedicated comparison sites. All of this is ripe for trademark infringement allegations where the comparison does not comply with comparative advertising rules.

The topics raised here give brand owners plenty of food for thought. And that is without even considering the role of AI-powered summaries of reviews/recommendations (e.g., what should a brand do if incorrect pricing or features are mistakenly attributed to its product in a summary). With the current pace of change, it may not be long before we see more courts grappling with these issues.

The author thanks Hiba Abdi, a trainee at Bird & Bird LLP, for her assistance with this article.