Swiss shoppers trust chatbots. Should they?

More and more people are turning to AI chatbots for shopping advice. Keystone

As generative AI evolves into a commercial platform that can influence buying decisions like never before, businesses in Switzerland and around the world are racing to appear in chatbot responses. But can consumers trust the results?

In the past, choosing a hotel, an insurance policy or a new smartphone meant comparing dozens of results on Google. Today, more and more people – including many in Switzerland – turn to chatbots such as ChatGPT. New research shows artificial intelligence (AI) chatbots can influence what people buy nearly three times more effectively than search engines like Google.

The reason is simple: AI models do more than search for information. They analyse options, summarise reviews and produce personalised recommendations.

“They save us a lot of time and help simplify things,” explains Felix Schakols, a marketing researcher at the University of St Gallen. Insurance policies are a perfect example – a product that is particularly important for many Swiss households, as Switzerland consistently ranks among the countries with the highest insurance premiums per capita.

Comparing complex clauses, deductibles and coverage options requires time and expertise. “Generative AI models are very effective at doing comparisons for us. They make decision-making faster and reduce the cognitive load,” Schakols adds.

Users tend to perceive chatbots as neutral advisers and increasingly rely on their recommendations when making purchasing decisions, sometimes even sharing personal data. Researchers and consumer advocates say that the growing influence of AI assistants could distort consumer choice – and they are concerned about transparency, trust and who controls the information users receive.

The question of how much users should trust chatbots is becoming more urgent as AI firms inject advertising into their products. Earlier this year, OpenAI began testing ads in ChatGPTExternal link in the United States and other selected countries. The intention is to gradually expand these tests to different markets, and experts say such ads are likely to arrive in Europe and Switzerland soon.

The risk of manipulated AI recommendations

The issue of how much to trust chatbots goes beyond advertising. Consumer organisations warn that companies, interest groups or even states could attempt to influence the sources chatbots use for their responses – a phenomenon known as “AI poisoning.” One method is to publish fake reviews on platforms such as Reddit, which AI systems may consult to generate their recommendations.

In a test conducted by the Swiss Foundation for Consumer Protection, when asked “Which laptop should I buy? Justify your recommendation transparently,” ChatGPT and Claude produced very different answers, without clearly explaining how they reached their conclusions. In some cases, they recommended computers that did not exist and listed incorrect technical information.

“Consumers cannot assume that AI provides complete, neutral and accurate answers,” warns Lucien Jucker, a legal expert at the consumers foundation. “They should always critically verify chatbot responses and compare results obtained using different prompts.”

Chatbots three times more persuasive than Google

But often, users don’t even suspect chatbots’ advice has been manipulated.

A recent studyExternal link from Princeton University, conducted on more than 2,000 participants in the United States, found that popular chatbots such as ChatGPT, Claude, DeepSeek and Gemini are almost three times more effective than traditional search engines at persuading users to choose a sponsored product – often without users realising they have been influenced.

“Less than one in five participants noticed that we had instructed the chatbots to persuade them to choose a specific product,” says Francesco Salvi, one of the researchers behind the study. Even when advertising is made explicit, the rate at which users select sponsored products remains twice as high as in traditional advertising, he adds. This gives the companies behind AI models significant commercial power as they look to monetise their services. The risk, Salvi argues, is that AI systems may not necessarily recommend the best products for users, but rather those that serve the commercial interests of platforms or advertisers.

OpenAI could influence hundreds of millions of people

OpenAI has promisedExternal link that its advertising will not affect the quality of ChatGPT’s responses and that ads will be clearly labelled and separated from organic content. However, ads are shown based on user conversations and prompts. For example, a question about furnishing a new home could trigger an advertisement for a sofa.

“Without a structural separation between advertising and responses, and without the possibility for external auditors to inspect the model, it is difficult to trust that the chatbot will not attempt to manipulate users,” Salvi says. With a record  one billionExternal link monthly active users, ChatGPT is evolving from a simple assistant into a commercial platform potentially capable of influencing hundreds of millions of people. OpenAI is in the process of developing new advertising formatsExternal link designed to encourage actions such as purchases and bookings.

ChatGPT, OpenAI’s AI chatbot, has reached a record one billion monthly active users. Keystone / Christian Beutler

Few Swiss companies are ready for AI-driven marketing era

With users increasingly relying on AI for purchasing decisions, influencing chatbots is becoming a business strategy. “If your brand is not among those recommended in responses, it’s as if it does not exist,” says Heather Holmes, co-founder of the US-based PR agency Publicity For Good.

Holmes explains that most content cited by AI comes from so-called “earned media” – reputable outlets that review products because they consider them worthwhile, not because they are paid to do so.

Other channels include backlinks from independent sources, positive reviews on platforms such as Reddit, and websites optimised to be easily read by machines.

Search Engine Optimisation (SEO), long used to improve visibility on Google, is increasingly being replaced by Generative Engine Optimisation (GEO): an emerging approach aimed at increasing the likelihood that AI systems cite or recommend a company’s products.

According to marketing expert and researcher Felix Schakols, many Swiss companies are discussing how to appear in chatbot responses, but few have developed mature strategies yet. “Most Swiss companies are just taking their first steps in this direction,” he says.

A 2025 studyExternal link by Schakols shows that only 12% of Swiss marketing managers surveyed use generative AI on a daily basis. He notes, however, that interest in GEO strategies is growing, including seeking the kind of authoritative reviews highlighted by Holmes.

Market research reportsExternal link indicate that companies in the United States and major European markets such as Germany, the United Kingdom and France are already investing heavily in GEO services and tools.

Swiss consumer groups call for stronger safeguards

Consumer organisations, meanwhile, question whether existing legal safeguards are sufficient in the age of generative AI. Chatbots can produce responses that misrepresent the characteristics of products or services, potentially leading consumers to make poor decisions if they rely on them.

“If a chatbot produces incorrect answers and a consumer suffers a loss as a result, AI companies are generally not liable under current law,” says legal expert Lucien Jucker from the Swiss Foundation for Consumer Protection.

While the European Union has updated its product liability rules to better account for technologies such as AI, it remains unclear whether existing Swiss law fully applies to software-based systems, Jucker says. A recent parliamentary motionExternal link has called for Swiss rules to be aligned more closely with the EU framework, including explicit coverage of AI software.

Even if lawmakers strengthen the rules, Jucker believes consumers will continue to bear much of the responsibility. “Updating the law would be an important step forward,” he says. “But it’s important that consumers don’t rely blindly on what chatbots recommend they buy.”

Edited by Gabe Bullard/VdV/amva