Review AI wine data

Every AI tool in Fenómenus leaves its result on screen before saving. This page is the checklist for that moment, and the argument for never skipping it.

Why review is not optional

A wine list is a claim of expertise. A generated description is fluent, confident and plausible — which means an error in it does not look like an error. The failure mode is not gibberish; it is a beautifully written note about a different wine.

Guests do not verify wine lists. Sommeliers and suppliers do.

The five-point check

For each wine, against the physical bottle:

1. Producer — the house that actually made it, not a similarly named one.
2. Cuvée — the specific wine. Producers make several with related names.
3. Vintage — the year on the bottle.
4. Region — the denomination on the label.
5. Grape — the varieties, and their spelling matching the rest of your list.

If any of the five is wrong, the record is about another wine. Correct it or discard it; do not keep the parts that sound right.

Then read it as a guest

• Does the description say something true and useful, or is it generic praise?
• Is it in the voice of your list, or noticeably different from the wines you wrote by hand?
• Are the tasting notes plausible for that wine and that vintage?
• Do the Parker and Peñín scores match a published rating for that exact vintage, rather than another release of the same wine?
• Is the pairing suggestion something your kitchen actually serves?
• Is the length consistent with the rest of the list?

A menu where twenty wines read one way and two hundred read another is visibly machine-filled.

Prices are always yours

No AI tool sets your sale prices. The market references it can fill — supermarket, minimum, average and maximum restaurant price — are internal guidance, never shown to guests.

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

Watch for confident detail on obscure wines

The riskiest records are small producers, private labels and local wines. A tool that cannot find the wine may still produce a complete-looking sheet.

Rule of thumb: the less you would expect a database to know about a bottle, the more carefully you read what came back.

Review in batches you will finish

Process twenty to fifty wines at a time. A four-hundred-wine run produces a review that nobody completes, and unreviewed content published on a wine list is a credibility problem, not a productivity gain.

What review does not have to mean

You do not have to rewrite everything. A realistic standard:

• Verify the five identity fields on every wine.
• Read every description that will be published.
• Rewrite the descriptions of the wines you actively sell — the by-the-glass selection and the recommendations.
• Accept the technical sheet on reference wines that guests rarely ask about.

After the review

1. Save only what you have read.
2. Spot-check ten saved wines against their bottles.
3. Confirm no hand-curated value was overwritten.
4. Confirm prices are unchanged.
5. Confirm an unavailable rating stayed empty rather than becoming zero.
6. Publish, then read the list signed out on a phone.
7. Regenerate the PDF.

The standing rule

Anything an AI tool produces is a draft until a person has confirmed it. Fenómenus enforces the pause — the results wait on screen — but the judgement is yours, and it is the part that cannot be automated.