Can AI reshape trade mark practice – promise, pitfalls and the average consumer of the future

The MARQUES panel on external factors changing our work also turned to the potential of artificial intelligence to transform trade mark practice. Dev Gangjee explored how AI could help fill gaps in the international system, improve trade mark processes, and even challenge the foundations of trade mark law - who is the average consumer when an AI agent is doing all of your purchasing for you?

AI tools in trade mark practice

AI already promises to support trade mark offices and practitioners in a range of tasks:

• Detecting fraudulent filings - particularly when providing evidence of use

• Assisting with specification updates and approvals

• Conducting goods and services comparisons

• Performing mark-for-mark analyses

• Machine translation

But how reliable are these systems in practice?

The challenges of image search

A recent study tested the ability of EUIPO and BOIP systems to identify earlier figurative marks. The results were sobering.

Initially EUIPO performed very well. When tested using a dataset of EUIPO Opposition Division decisions, it retrieved the relevant opponent's mark 62.32% of the time. The opponent's mark was typically ranked at 4th place on the list of problematic earlier marks. However, the EUIPO's system was likely trained on the EUIPO data so testing based on that same data yielded an inaccurate impression of good performance.

When the EUIPO's image search was tested based on Benelux IP office (BOIP) decisions, the hit rate dropped to 26.9% when considering the first 1,000 results and 15% (within 50 results - BOIP's limit).

So, whilst these tools are an exciting development, they are currently less effective than expected, and headline figures can mask the true picture.

AI as a trade mark examiner

Other AI models have been created to predict inherent distinctiveness. When tested using 1.5 million US word marks and 2.2 million office actions, one machine learning system reached 86% accuracy when tested against recent applications. For clear-cut cases (very distinctive or clearly descriptive marks), accuracy was above 90%.

However, for borderline marks – the kind that keep examiners and lawyers busy – performance dropped sharply. Here, the AI’s lack of nuance and over-confidence may mislead rather than assist.

The message is clear: AI can triage and support, but human judgement remains essential.

The “average consumer” in an AI world

Perhaps the most profound question posed was whether AI will alter the very functions of trade marks. Trade marks protect the signals brands send to consumers: origin, consistent quality, advertising, investment and communication.

But what happens if the “consumer” is a shopping bot or agentic AI? Already, retailers such as Walmart are rolling out AI assistants that build shopping lists and complete checkouts based on natural language prompts. Smart devices may reorder supplies without human input.

If technology makes purchasing decisions directly, does the trade mark still serve its traditional functions? Or must we reconceive the “average consumer” as AI-assisted?

Autonomy and adoption

Studies show that (human!) consumers value autonomy. They may welcome recommend systems that suggest options, but resist fully autonomous systems that make purchases without their involvement. The trade-off depends on context: a passionate runner may trust an AI app to order the optimum running shoes, but shoppers resist ceding control over identity-linked choices. (An exception to this tends to be where price comes in. Consumers are happy to identify the product and then leave it to the AI to buy it at the optimum time.)

This creates both a design challenge for technology providers and a legal challenge for trade mark law. Will dilution, rather than confusion, become the primary battleground? Will competition law or platform regulation take on issues of bias and manipulation?

Copyright and AI authorship

The panel also touched on copyright. Human authorship has always been the foundation of protection. If AI outputs are imitative and derivative but mass-producible, should they qualify for rights? Granting protection risks flooding the system with machine-generated works.

Some jurisdictions are already requiring disclosure of AI use when registering and relying on works which were AI created (or AI assisted), with legal sanctions for misrepresentation.

What does this mean?

AI is set to transform trade mark practice – from searching and examination to the very conception of the consumer. But it is not a panacea. Current tools often overpromise, and human oversight is critical. The rise of agentic AI could shrink the space for brands to influence human choice, forcing us to revisit core doctrines.

At the same time, copyright law faces pressure to preserve the human core of creativity. Across the board, AI is less about replacing people and more about rethinking legitimacy, accountability and the functions of IP law.

To find out more about the issues raised in this blog contact Rosie Burbidge.