The EU AI Act Regulates How AI Is Built, Not Whether It Lies About Your Brand

Brands large and small that are clamoring to be discoverable in search powered by artificial intelligence (AI) have at least one big thing to keep in mind first.

When Columbia’s Tow Center asked eight AI search tools who wrote a news story, who published it, and where to find it, the tools got it wrong more than 60% of the time. The customer asking those same engines about your brand tomorrow has no way to tell.

You’d assume that when an AI engine decides which brands to name, it favours the accurate ones. That’s a common misconception. Industry research suggests that one of the strongest predictors of being cited isn’t solely accuracy or traffic, but sheer frequency of mentions across many platforms, with the most-cited pages often carrying less traffic than the ones engines skip.

Essentially, the machine rewards being talked about, not being right. And that’s now your problem, because this is fast becoming how customers meet your brand.

While the EU AI Act has just rolled out another phase of enforcement in August 2026, the regulations don’t address how information is cited. Instead, the Act governs how AI is built and disclosed, including forcing chatbots to identify themselves and AI content to be labeled as such.

With neither the regulator nor the engines accountable for how a European brand gets represented, the only party left with both the incentive and the ability to keep that representation honest is the brand itself.

The exposure is reputational

Imagine a game of broken telephone, where a message gets whispered down a line and comes out the other end scrambled. The search engine version of this can be damaging. The LLM reads what’s out there about your company and hands the customer a garbled version on the very first pass. 

In broken telephone, you expect the message at the end to be wrong. Here, the customer gets one sentence about your brand and will take it at face value.

We’ve submitted to its impacts with the overarching statement, “It reflects whatever it was fed.” A simple shrug, accepting the terms that come with using the tool. But there’s real stakeholders on the ground working to fix how AI misrepresents at a larger scale. 

“A model isn’t neutral. It inherits the blind spots of whatever it reads,” says Larry Adams, the founder of Chromatics AI, a Netherlands-headquartered company involved in creating Aisha, an AI assistant built to represent Black history and experience that mainstream models handle poorly.

“For a brand, that translates to reputational exposure. Every gap you leave is one the machine fills on your behalf, from whatever else it happens to have absorbed, and you don’t get a say in what it grabs,” he tells Entrepreneur Europe. “The only move you have is to leave a clean, consistent version of yourself sitting there for it to find first.”

At the individual level, Europeans at least have some recourse. Vienna-based digital rights group NOYB filed GDPR complaints against OpenAI over factual errors, including on behalf of a Norwegian citizen after ChatGPT told him he had been convicted of murdering his two sons. It was a fabrication mixed with real details, like the correct number of children and his hometown.

However, the parts of European law that do reach AI-generated falsehoods stop at the individual; everything the machine invents about a brand sits outside that protection.

You can’t make an LLM value truth, but you can make your brand hard to garble

An analysis of 75,000 brands found that the ones mentioned most often across the web get more than ten times the AI citations of the tier below them. The pages these models lean on, furthermore tend to have less traffic and fewer links than those they skip. 

What predicts a citation isn’t only the ranking signals SEO was built to optimize. How often and how consistently a name turns up across trusted places is now doing much of the work. The confidence an engine has in a brand is built from mentioning frequency and source quality together, not backlinks alone.

“The big driver has been a massive shift in clicks coming from search. Branded search has forever been an amazing driver of traffic,” notes Brian Yamada, Chief Innovation Officer at VML and WPP Enterprise Solutions, in conversation with Entrepreneur Europe.

“So, figuring out what’s changing, how they’re showing up now, and what they need to do about it is massive. This is similar to the early days of social media monitoring, where the platforms and the models are changing pretty quickly, and the tools to measure them continue to evolve.”

Tameem Rahman, founder and CEO of search marketing agency Kingmaker Search, sees the discipline itself changing shape from the inside.

“The question isn’t just ‘How do we rank?’ anymore. It’s ‘What source material is AI reading and citing, and how do we build our clients’ presence within it?'”

Rahman was careful to note that traditional SEO fundamentals haven’t disappeared, and that genuinely useful content that ranks well is still likely to become AI source material. But he argues that ranking is now only one part of a wider job. 

“Don’t just optimize your website to be found. Build a credible presence everywhere AI might look when deciding who you are, what you’re good at and whether you’re worth recommending.”

What it responds to is something more mechanical, information that’s clean to pull out and easy to check, and when you give it that, you aren’t earning the machine’s respect so much as lowering the odds it mangles you. The goal was never to be trusted by the LLM, it’s to be too legible to misrepresent.

And now in the direction agentic commerce is heading, managing this is becoming even more vital for B2C brands.

Machines shopping on customers’ behalf, Yamada adds, don’t all behave the same way. Once agents become the primary interface, showing up in that moment of inquiry, which is when the consumer is actually asking, is what separates visible brands from invisible ones. 

“We’re finding the commerce agent assistants to be extraordinarily product-detail focused, much more math-oriented and specification-driven, whereas the LLMs will typically be a little more narrative in their generated answers.”

What this looks like in practice for founders

Building for this environment isn’t just a technical exercise. It’s a series of choices about what a brand wants AI to know about it and how it wants those interactions to feel.

“Part of it is trying to understand what interaction we’re trying to power. Part of it is how we design it in a way that people will use it, find it, and find the interaction valuable,” Yamada explains.

“Is there something out of the box that does that, or is it something we need to custom develop, tapping into an API or multiple models to create something more unique?”

Ultimately, it’s about balancing what the business is trying to achieve, how differentiated they want to be, the user experience, whether it actually delivers value to people, and budget versus feasibility. 

In this sense, it doesn’t feel like traditional marketing at all. But it’s the only part of the pipeline a brand still controls.