The generator is only as good as what you tell it
Menuistia uses the same kind of models everyone else is talking about. The useful part is not the model. It is that the recipe comes out already bent around your kitchen: diet, allergies, time, season, and the food in front of you.
It is not magic, and it should not sound like it
You give it a pile of ordinary facts. Someone is vegetarian. Someone cannot eat nuts. You like slow Sunday cooking but need Tuesday to be fast. It is September, so the tomatoes are good and the asparagus is a memory. You want a soup, or a traybake, or you do not care as long as it is dinner.
From that it writes a recipe, or several, and a shopping list for what you do not already have. If you hand it a photo of a book page, a link, or a plate you liked in a restaurant, it tries to land on that dish and then adjust it. A restaurant photo is a sketch. Treat the result as a starting point you can taste and correct.
What to put in, if you want it to be useful
Preferences you will actually keep. "I like Thai" is weaker than "no coriander, yes lime, not too hot". An avoid list beats a cuisine label. Servings matter, because a recipe for four fed to two people is how leftovers become guilt.
If you only type "chicken dinner", you will get a chicken dinner that could have come from anywhere. That is not a failure of the tool. It is an empty brief.
Check the obvious things
Salt, heat, and anything medical should stay in your hands. A generated recipe can mis-count, suggest a substitution you dislike, or sound confident about a bake that wants exact ratios. Read it the way you would read a note from a keen friend: probably fine for a skillet dinner, worth a second look before you commit to a cake for other people.
The shopping list is where you catch the drift. If it asks for six herbs for a Tuesday pasta, delete them and ask again with "pantry only".
Related guides
Get cooking!