In May 2026, the Regional Court of Munich ordered Google to stop its AI Overviews from making false claims about two German publishers, casting them as running scams and subscription traps that appeared in none of the cited sources. Google argued it was merely surfacing third-party content. The court disagreed, holding that because the AI writes and synthesizes its own text, the output is Google's own statement, and Google is answerable for it.
It is the first ruling of its kind, and the lesson is hard to miss. When AI speaks about a company, someone is accountable, and increasingly that someone is the company itself.
Nor is this only European. In 2024, Canada's Civil Resolution Tribunal held Air Canada liable when its chatbot invented a bereavement refund policy, rejecting the claim that the bot was a separate entity and that the real policy sat elsewhere on its site. Cases across the United States and Australia are testing where responsibility falls, with mixed outcomes; in Walters v. OpenAI, the AI company prevailed. Yet the reflexive defenses, that the AI said it, or that a disclaimer was showing, are proving fragile.
Even sophisticated firms are exposed. In 2025, Deloitte partially refunded the Australian government after a report for a federal department was found riddled with fabricated quotes and citations, including an invented passage attributed to a judge, the hallmarks of AI hallucination. NeuroRank groups these failures into one lens, ORHL, for Omitted, Replaced, Hallucinated, and Zero Leads. it gives you a clear view of the risks that pull you down, and in the recommendations it specifies the fixed that reduce the hallucinations.
The lawsuits are the visible edge of the problem. The larger risk is quieter, and it does not need a courtroom. Every day, AI models describe your brand to people who are deciding whether to buy from you, and when the description is wrong, the damage is immediate and unseen. A model can leave you out of an answer where your product is the obvious fit. It can name a competitor in your place. It can invent a false detail about your pricing, your policy, or your quality, and send a customer away with an impression you never get the chance to correct.
Three things follow from this, and every brand should sit with them.
First, AI hallucinations are real and frequent. The cases above are not exotic edge cases. They involve a global search engine, a national airline, and a Big Four consultancy, each tripped up by AI stating things that were simply untrue. At the scale these systems now operate, confident errors are not rare. They are a constant background rate.
Second, people act on what AI tells them, and few stop to check. The answer arrives with the tone of a settled fact and the ease of a single reply. Most people do not open the sources. They read the summary, form a view, and move on. A false statement inside an AI answer is trusted by default.
Third, the cost is yours, and it is commercial long before it is ever legal. A hallucination about your brand rarely ends in a lawsuit. It ends in a customer who trusts you a little less, or who never becomes a customer at all. That is the real exposure, not legal fees but eroded trust and revenue that leaves without a sound. No AI company will carry that cost for you, and no disclaimer will refund it. It is yours to protect.
Put together, the cost of leaving AI answers ungoverned works at several levels, and it helps to picture them as an inverted funnel, ordered by how often each one happens.
At the top, widest and most frequent, sit plain factual errors: a missing product model, a feature described as absent when it exists, an out-of-date price. Small on their own, they are relentless, and they strike the buying funnel directly. A buyer comparing options is quietly steered elsewhere, your organic preference slips, and the paid clicks you are still buying land on a decision the AI answer has already shaped against you. That is a double hit to revenue, earned and paid at once.
Below that, less frequent but heavier, sits reputation. A hallucinated negative review, an invented verdict on your product's performance, a misread of sentiment repeated as fact. This does not only cost a sale. It moves conversion rates, shapes how the market perceives your brand, and near the top of an organization can travel further, into how investors, acquirers, and analysts read you, with consequences for a valuation, a merger conversation, an IPO narrative, or a share price.
At the narrow base, rarest but most severe, sits the error that ends up in the cases above: a false claim of wrongdoing, a fabricated safety or compliance failure, a damaging tie to a scandal that was never yours. Uncommon, but each one can trigger legal exposure, regulatory attention, and a board-level crisis far out of proportion to how often it happens.
The pattern is consistent. The errors that occur most often drain you quietly and continuously, in revenue. The errors that occur rarely hit hard and fast, in reputation and standing. Left ungoverned, AI answers expose you at both ends of the funnel at once.
None of this is inevitable. A brand's information in AI can be governed, in the same deliberate way a business already governs its search presence, its public messaging, and its financial disclosures. It means knowing, precisely and continuously, how ChatGPT, Gemini, Claude, and Perplexity represent your brand across every ORHL failure, tracing each error to its cause in the content the models learn from, and correcting it at the source, with clear ownership. That discipline is what NeuroRank was built to make possible.
The cases will keep coming, and the law will keep catching up to the technology. But no business should wait for a courtroom to learn what its AI answer says. The exposure is live now, in thousands of quiet conversations a company never sees. The brands that come through this well will decide, deliberately, to know what AI says about them and to take responsibility for correcting it. Your brand's information is now yours to govern. If you do not, an AI model will do it for you, and it will not always tell the truth.
By : Ambika Sharma, Founder and Chief Strategist at Pulp Strategy Communications, and Product Architect of NeuroRank
