AI has become the newest gateway between brands and their audiences. Increasingly, customers aren’t visiting your website first, or reading your latest press release. They’re asking ChatGPT, Perplexity, Claude, or another large language model (LLM) to tell them what they want to know.
At the end of last year, more than a third of respondents to one survey (37%) said they turn to AI rather than traditional search engines when initiating a general search. And the answer they receive could influence whether they buy from, invest in, or even trust you at all.
The main problem with that? Your brand may not have written that answer – and it may be wholly inaccurate.
This was the discussion between Dr Cecilia Dones and Iesha White at MAD//FEST London 2026. The audience were transfixed – and possibly a little horrified – because, as an industry, we’ve spent years learning how to manage what people say about our brands. Refining our craft. Adjusting to trends. But it’s becoming clear that we need to learn how to manage what AI says about them instead.
The new reputation challenge
Both speakers shared a simple exercise they’d recently undertaken, which involved asking AI systems to describe themselves and their organisations. The results ranged from amusing to concerning, including incorrect executive names, outdated information, and fabricated affiliations.
These weren’t isolated glitches, they were reminders that generative AI is probabilistic, not authoritative. For brands, the commercial implications can be huge.
When hallucinations become business problems
It’s easy to dismiss AI hallucinations as harmless mistakes until they cost real money. One example shared during the session was of a US solar installation company that reportedly began losing contracts after AI-generated summaries incorrectly suggested the business had been involved in lawsuits and malpractice. None of the claims were true, yet prospective customers acted on the misinformation before speaking with the company.
Whether through accidental hallucination or poor source aggregation, the outcome was the same: damaged trust and lost revenue. With AI as the first point of contact between consumers and brands in more and more cases, such examples reinforce that inaccuracies present genuine risks rather than just technical issues.
Data poisoning is surprisingly easy
Perhaps the most striking insight from the session concerned ‘data poisoning.’
Unlike humans, LLMs don’t truly understand truth. They identify patterns, repetition and statistical likelihood. Research cited during the discussion suggested it can take remarkably little misleading information to influence how an AI system characterises a brand. One Cornell study found that as few as 13 words posted on Reddit could alter an LLM’s perception of a company, with tainted results potentially appearing within 24 hours.
Another experiment demonstrated how repeatedly posting a fabricated claim, in this case the claim that a professor had won a hot dog eating contest, was quickly incorporated into AI-generated biographies as fact.
The experiment itself was, of course, harmless, but the implications were anything but.
A legal grey area
For years, social media platforms argued they were simply hosting user-generated content, but generative AI changes that equation. Rather than simply displaying existing content, LLMs synthesise, summarise and generate new responses based on numerous sources. That distinction is becoming increasingly important in legal circles.
The discussion highlighted emerging legal cases, including proceedings in Germany examining whether AI providers could be held responsible when generated summaries harm businesses through misinformation. Similar debates are beginning to emerge elsewhere as regulators grapple with where accountability should lie.
While the legal landscape is still evolving, it is clear that brands cannot assume someone else will manage AI accuracy for them.
So what should brands do?
The advice from the panel was refreshingly practical.
First, audit your AI presence regularly. Ask multiple LLMs about your company, products and leadership team. Different models use different data sources and update at different speeds, meaning you’ll likely receive different answers.
Second, don’t ignore inaccuracies. Report them where platforms allow corrections and engage with providers, particularly if you’re already investing advertising spend with them.
Third, double down on good communications. High-quality journalism, strong public relations, authoritative websites and consistent brand storytelling have never been more valuable. AI systems learn from the public information ecosystem. If your organisation isn’t actively telling its story, AI will assemble one from whatever signals it can find.
Your reputation now has an AI audience
For decades, brand reputation management focused on journalists, search engines, review platforms and social media. Today, AI is a whole new audience to consider. Every AI-generated summary becomes another touchpoint between your organisation and potential customers and, while sometimes it’s accurate, sometimes it’s incomplete. Occasionally, it’s plain wrong.
Reputation management has entered a new era, where monitoring what machines say about your brand is becoming just as important as monitoring what people say. The fact that LLMs are becoming powerful intermediaries for brand perception may be an uncomfortable reality for many advertisers, but if false information can be amplified so easily, organisations must consider AI manipulation as part of their reputation management strategy.
If there was one takeaway from the chat, it was this: AI models need access to accurate, trustworthy information about an organisation. So brands have to structure their content so that facts surface instead of any misinformation. Their reputation depends on it.