You ask an AI what to eat, it gives you a plan in four seconds, the plan looks professional and you follow it for the next month. Nobody told you the answer was convincing rather than correct. That is the distinction nutritionists have now started making out loud.

The Spanish nutritionist Aitor Sánchez puts the problem in one sentence: artificial intelligence can give a persuasive answer that is not necessarily the most appropriate one. He adds something even more specific - these systems tend to flatter and to find a positive angle even when the recommendation has no scientific basis. Then it gets something wrong and simply walks it back.

That "walking it back" is the key detail. When the mistake belongs to an algorithm, it simply corrects itself on the next question. When the mistake is in your diet for three months, nobody walks anything back.

What the AI does not know about you. It does not know your medical history. It cannot see your blood results. It does not know your relationship with food, whether you have restrictive diets behind you, how much stress you carry, or whether the plan it hands you can be sustained for more than two weeks. Allergies and intolerances often do not enter the calculation at all, and the information can be out of date or simply wrong.

Nutrition coach Clau Gago draws the line in a sensible place: the tool is useful for organising ideas and explaining a concept. The problem starts the moment that quick answer begins to replace personalised healthcare.

Spain's General Council of Dietitians and Nutritionists goes further, saying that AI-generated meal plans amount to professional fraud, that such guidance is neither precise nor safe where health is concerned, and that the field needs regulation.

Here the question is simpler, because the queue is shorter. Before there was a chatbot, the meal plan came from a neighbour, an aunt or a Facebook group. Now it comes from a screen that writes better than all of them and sounds more certain than any doctor. Does that make it correct, or only more convincing?