Your doctor recommends a medicine. The advice is sound, the treatment works. Then you learn that the company that makes it pays him a bonus for every prescription.
Nothing about the medicine has changed. And yet something has broken. You can no longer know whether the advice was for you, or for him.
A confidant like no other
You don't talk to ChatGPT the way you read a newspaper. You confide things in it: a symptom that worries you, a money problem, an argument, a hesitation before voting. For many people it has become a kind of confidant, available at any hour, who does not judge.
That confidant will now display advertising. OpenAI is already testing ads in the United States, and has just announced a new image-based format. The company promises that advertising does not influence the answers.
Let us take that promise seriously. Let us suppose it is kept perfectly, answer after answer. Is the problem solved?
What advertising changes, even without cheating
It gives the service an interest in keeping you there. A service funded by advertising makes more money when you spend more time on it. Nothing forces it to distort an answer to achieve that. It is enough to be a little more agreeable, a little more flattering, a little more captivating. We have already shown why AIs tend to tell you you're right: advertising does not create that tendency, but it rewards it.
It gives the service an interest in knowing you better. An advertisement is worth all the more when it is targeted. And an intimate conversation is the best source of targeting there is.
It moves the line, slowly. Television, search engines and social networks all began with sincere promises. Then, year after year, the boundary between service and advertising slipped, without any single day marking the break.
In 2004, the head of TF1, Patrick Le Lay, summed up the advertising-funded television model with a frankness that caused a scandal. What his channel sold to a major drinks brand, he explained in substance, was "available human brain time": programmes designed to prepare the audience to receive advertising messages. The phrase remained famous because it said out loud what the model implied. An assistant funded by advertising also sells something very precious: our attention, at the moment when we are most trusting.
The argument on the other side
The other side has to be heard too, and it is a serious one.
Advertising makes it possible to offer a free service to those who cannot pay. Without it, the best assistant might become a privilege reserved for subscribers. And newspapers have lived on advertising for two centuries without their articles necessarily being bought: they have learned to separate editorial content strictly from the advertising department.
So the question is not "advertising or not". It is: what protections?
What would make trust possible
Adverts that are always clearly labelled, and never mixed in with the answer itself. Outside checks, capable of verifying that answers are not influenced, rather than a simple promise. Real transparency about what the advertiser knows about you. And the option, for anyone who wants it, to pay not to be targeted.
These are the rules we eventually imposed on the press, on television and on online advertising. They took decades to arrive. For assistants, they have yet to be written.
What we take away
A confidant can be paid. Doctors, lawyers, advisers all are. What matters is knowing by whom, and being able to verify that this payment does not weigh on the advice.
As long as we cannot verify it, the best protection remains the simplest: remembering, before telling everything, that there may be someone else in the conversation.