The French grant more credibility to search engines and generative AI, but their trust remains low when it comes to the protection of their personal data. According to an Ipsos study conducted for Eskimoz in early 2026 among 1,000 French people, 65% of them find the content provided by search engines credible, an increase of 9 points in one year, and generative AI reaches 35% (+8 points). Regarding personal data protection, however, no platform exceeds 35% trust, and generative AI plateaus at 20%.
Believing Is Not Entrusting
Credibility and trust are often confused, although they represent two different commitments. Credibility means believing what a platform claims, trust means letting it act on one’s behalf, with one’s data, money, and preferences. According to the analysis, platforms have won the first battle, but the second remains to be won.
This gap is nothing new. Researchers Patricia Norberg, Daniel Horne, and David Horne coined it the ‘privacy paradox’ as early as 2007: people claim to be wary of platforms, yet accept terms and conditions without reading them. For twenty years, this compromise worked. With agentic AI, meaning AI agents that act for the user, the exchange changes nature: one no longer consults, one delegates. An agent that books a ticket or compares three insurance policies needs the user’s bank card, address, and history.
George Akerlof demonstrated in 1970, with his ‘market for lemons,’ that when a buyer cannot verify what they are being sold, the entire market can shrink. In a process where the human no longer sees the intermediate steps, trust ceases to be an add-on and becomes, according to the column, the only guarantee.
Advertising, an Invisible Tax on Trust
Still according to the Ipsos study, 59% of French people say they are bothered by advertising on video platforms, an increase of 7 points in one year. The annoyance drops to 38% on search engines and stands at 29% for generative AI, which is still little monetized through advertising. The analysis distinguishes three advertising regimes: ‘response-ad,’ displayed when the user expresses an intent, thus tolerated; ‘intrusion-ad’ from videos and social networks, which interrupts and irritates; and the ‘advice-ad’ of AI assistants, the most delicate, because a sponsored recommendation that blends with the response creates a doubt that affects the whole.
The image of a pharmacist receiving a commission on every recommended box illustrates this risk: one would question not just one product, but all past recommendations. The problem is that the players who most need trust to get agentic AI off the ground will also be those with the most interest in monetizing it.
How to Be Judged Legitimate?
A window remains open. For products, generative AI remains marginal, with 4% of searches, but it reaches 10% for services, and 12% among 18-34 year olds. The analysis sees this as a weak signal: AI enters through services, where advice matters more than price. Online commerce offers a precedent. In the early 2000s, many refused to enter their bank card number online, and e-commerce progressed thanks to a trust infrastructure: the browser padlock, strong customer authentication for payments, and the right of withdrawal.
For brands, three tasks are proposed. Explain the origin of information, how a recommendation is built, and what is sponsored. Reassure by collecting less data and clearly stating why the data that is kept is retained. Demonstrate through guarantees, a return policy, and verified reviews—all evidence that AI agents will be able to read and compare. The author bets that the first player to make data respect a commercial argument will capture a large share of agentic commerce.
