Partner Abigail Healey and Lucia Cipolat examine liability for inaccurate AI-generated search results, in the FT’s Agenda

Writing in Agenda, Partner Abigail Healey and Paralegal Lucia Cipolat examine the current forms of redress available to businesses and individuals affected by inaccurate or damaging AI-generated search results, and consider how English law may respond as search engines increasingly generate content of their own.

With artificial intelligence-generated answers fast becoming the first thing users see in search results, a pressing legal question is emerging — who will be held liable when those answers are wrong?

For boards and executives across different industries and jurisdictions, this question increasingly matters.

As generative AI becomes embedded in the tools through which customers, investors and other stakeholders obtain information, an inaccurate AI-generated answer at the top of a search page can cause significant reputational damage before a user ever reaches the underlying sources.

As businesses operate across borders and AI-generated information is distributed globally, questions of responsibility, applicable law and available remedies are unlikely to remain confined to any one jurisdiction. More broadly, as AI becomes a strategic and operational cornerstone across industries, understanding how AI-generated information about a business is produced, monitored and challenged is likely to become an increasingly important governance and reputational issue.

The U.K. Jurisdiction Taskforce’s Legal Statement on Liability for Non-Deliberate AI Harms (the “UKJT Statement”) offers one of the first attempts to answer the question of liability. Its analysis is particularly important where it applies to search engines and the extent to which these companies can be liable for incorrect content.

Although the UKJT Statement addresses English law, the questions it raises are not confined to the U.K. For U.S.-based boards overseeing businesses with U.K. operations, customers or reputational exposure, English-law liability may be directly relevant. More broadly, the developing English approach provides a useful comparison as courts in different jurisdictions grapple with how existing liability rules should apply to generative A.I.

English law already draws a distinction between those who create content and those who merely facilitate its publication. Publication by an individual of words created with the assistance of AI does not disturb the established principle that the individual is clearly the publisher, in that they have “participated in, or authorized, its publication.”

A more nuanced position arises where a search engine has integrated AI functions into its services and content is generated and presented by that search engine, i.e. so-called AI Overviews.

Historically, English courts have generally treated search engines as passive intermediaries rather than publishers in relation to posts made by their subscribers, automatically generated “snippets” of third-party websites and platforms provided for blogs.

Those principles were developed in an era when search engines merely organised information. AI-generated search results, however, raise the question of whether these frameworks still apply or whether the AI function goes beyond merely organising information.

This is no longer an issue only for technology businesses. As generative AI becomes embedded in search, research and other everyday information tools, companies across all sectors may find that AI-generated descriptions of their businesses, products or executives shape how customers, investors and other stakeholders perceive them.

In relation to automatically generated snippets of third-party websites, the English courts have held that Google was acting more like a passive intermediary than an author because it did not exercise effective control over the specific content displayed and had no knowledge of any defamatory material. The UKJT Statement suggests that similar reasoning could arguably be applied where an AI tool generates new statements based on third-party content, rather than directly quoting websites.

AI Overviews, however, now go well beyond a short snippet. As search engines increasingly generate their own summaries, rather than simply directing users to third-party content, the analogy with earlier authorities becomes less convincing.
These questions have moved from theory to reality following a recent decision of the Regional Court of Munich, Germany, concerning Google’s AI Overview feature. The case involved two publishers who sought injunctive relief after Google’s AI-generated summaries falsely associated them with scams, subscription traps and fraudulent business practices. According to the German court, some of the allegations presented by the AI Overview did not appear in the underlying search results at all. This distinction is significant because AI Overviews do more than retrieve information. They synthesise it, often producing factual assertions that appear nowhere in the underlying sources.

The Munich court decision therefore raises the question left open by the UKJT Statement: When does a search engine stop acting as a passive intermediary and become responsible for the content it generates?

Viewed through the lens of existing English authorities, the Munich decision represents a potentially different approach. However, where a search engine presents AI-generated factual assertions directly within its search results, without requiring users to engage with a separate chatbot, the Munich court’s decision arguably reflects the better view.

Even if English courts do take this approach, search engine providers are unlikely to be without defenses. The UKJT Statement also notes that clear warnings that AI-generated content may contain “hallucinations” will influence the extent to which such content is defamatory. There may also be recourse for search engine providers if it can be shown that the search engine was “not the operator who posted the statement on the website.” AI-generated outputs expose an unresolved question: If no human has “posted” the statement, who has?

The implications of these developments also extend beyond defamation laws. AI-generated overviews may process personal data inaccurately, combine information from multiple sources in misleading ways or disclose private information which otherwise may not have been particularly prominent online. Individuals and businesses affected by such outputs may therefore have various legal rights and remedies available to them in order to protect their reputation and information.
For businesses and individuals concerned about the potential harm caused by damaging summaries, the UKJT Statement suggests that English law already possesses many of the tools required to address AI-generated wrongdoing. The more difficult question is whether courts will continue to regard AI-enabled search engines as passive intermediaries once they begin generating factual assertions of their own.

The Munich decision distinguishes generative AI from traditional search technologies. If other jurisdictions follow a similar path, the legal protections that have long underpinned search engines may prove less secure in the age of generative AI.

Abigail and Lucia’s article was published in Agenda, an FT Specialist publication, 14 September 2026.