AI tool limitations

Knowing what these tools do not do is what makes them safe to use. This page is deliberately a list of boundaries.

They do not set prices

No tool sets your glass, bottle, portion or menu prices. The wine tools can fill market references — supermarket price, and minimum, average and maximum restaurant price — but those are internal guidance, never shown to guests.

Those references do not know your city, your rent, your service or your cellar. Treat them as one input among several.

They do not assign allergens

Allergens are never generated, inferred or imported. They are assigned per item, from the recipe, by a person who can be held responsible for the answer.

A generated description mentioning an ingredient is not an allergen declaration.

They do not publish

Nothing is published by a tool. Results wait on screen; you review and save, and publication remains a separate status change that carries every language.

They do not know your kitchen

They work from two things: what you type in the brief, and the business description saved in the setup assistant. They cannot know:

• the recipe as your kitchen makes it;
• what your supplier delivered this week;
• your portion sizes;
• your house terminology;
• what you have run out of.

Everything specific has to come from you.

They can be confidently wrong

The characteristic failure is a well-written record about a different wine, or a plausible ingredient that is not in your dish. Confidence in the output says nothing about its accuracy.

The riskiest cases are the ones where a database would not be expected to have data: small producers, private labels, house wines, local dishes.

Coverage is uneven

• Widely distributed wines are well covered.
• Small and local producers often are not.
• Handwritten and low-quality photographs produce more corrections.
• Very short or generic names are ambiguous.

The description writer produces Spanish

The Write content with AI action drafts its text in Spanish. For a menu in another source language, or for other published languages, use it as a starting point and translate the result.

The photo tool reads wine lists

Image analysis is specific to wine lists. There is no photo import for food menus, no dish recognition, and no image generation or retouching anywhere in the product.

A food menu that exists only on paper is typed into a spreadsheet and pasted into the bulk table.

They cost credit, and reruns cost again

Every generation consumes POM AI credit, with an estimate shown first. Regenerating to see a different result costs the same as the first attempt, which is an argument for reading the first one properly and editing it.

They are optional

Every AI tool can be turned off under Enabled tools, and a Fenómenus site works completely without them. They save time on research and drafting; they are not required to run a menu.

What they are genuinely good at

Stated fairly, because the boundaries above are not an argument against using them:

• turning a name into a structured wine record;
• completing technical detail across a long list;
• reading a printed wine list into rows;
• producing a first draft when the blank field is the obstacle;
• translating a finished description into a working draft.

Used with review, they turn days of work into hours. Used without review, they publish confident mistakes on a public menu.