How Label Scanning Works

Label scanning is Sommo’s front door. You point a camera at a bottle and, a moment later, you have the producer, the region, the grape, a tasting profile, a drinking window, and food pairings. The interesting part is not that it reads the label. Plenty of apps do that. The interesting part is what happens between the photo and the answer, and the rule that stops the app from making things up.

What you see

Open the camera and you have three ways in:

• A bottle label. The everyday case. Front label, one tap, full identification.
• A wine list. Menu mode reads a printed restaurant list and scores the entries so you can order well.
• A shelf. Point at a row of bottles in a shop and Sommo reads what it can.

You start with five scans free, for the life of the install. That is enough to scan a few bottles at home, see whether the answers are any good, and decide for yourself. Premium removes the cap.

What happens when you scan

Underneath the single tap, a short pipeline runs:

1. Text comes off the image. The label photo is read for words: producer, cuvée, vintage, appellation, alcohol.
2. The wine gets identified and enriched. Sommo AI takes that text and works out which wine it is, then fills in the things a label does not print: typical structure, the region’s character, and what the vintage means for how the wine should taste.
3. The answer is grounded. Before Sommo names a producer or a region, the result is checked against a library of real wines, wineries, regions and grapes that I maintain by hand. This is the step most “AI wine” features skip, and it is why Sommo does not invent estates that have never existed. I wrote about why this matters in the truth about AI wine apps.

Results are cached so a re-scan of a wine you already saved is instant. If you switch the app’s language, the wine’s facts stay the same, the proper nouns never get translated, and only the written description is produced fresh in your language.

What Sommo returns

A finished scan hands you a small, structured record rather than a paragraph of prose:

• A tasting profile: body, tannin, acidity, sweetness and the rest, shown as bars so you can read the shape of the wine at a glance.
• A drinking window: four dates that say when the wine opens up, when it peaks, and when it starts to fade. How those are estimated has its own page.
• Food pairings drawn from the wine’s actual structure, not a generic “reds with meat” rule.
• A quality read to set expectations.

From there a scan is a starting point, not a dead end. Save it to your journal to record what you actually thought, or add it to your cellar to track the bottle and get pairings from your own collection.

Menu and shelf modes

Scanning a single label answers “what is this wine?”. A restaurant changes the question to “which of these should I order?”. Menu scoring reads the whole list, scores each entry, and surfaces alternatives, so the work happens before the waiter comes back. It is a Premium feature, and it is deliberately a different flow from single-label scanning because the decision is different.

Where it stops

Honest is better than impressive, so here is where scanning struggles:

• Damaged or obscured labels. Torn, stained, or glare-blown labels lose text, and missing text means a weaker identification.
• Very obscure producers. A tiny grower with no presence in the library is the hardest case. When the model cannot ground a name, the app says it is unsure rather than guessing one.
• Non-vintage and multi-vintage wines. Many Champagnes and fortified wines carry no vintage, so vintage-specific detail does not apply.
• Non-Latin scripts and heavy stylisation. Ornate or non-Latin typography is where text extraction is least reliable.

In all of these, the app is built to admit uncertainty. A blank where a guess would be is the honest answer. The broader list of what Sommo can and cannot know lives on the limitations page.

See it in the app

The full product walk-through, with screenshots and the exact free-tier terms, lives on the AI wine scanning feature page. For a longer read on how scanning fits a beginner’s first week with the app, see the wine label scanning guide on the blog. The glossary entry on tasting notes explains the structure those tasting-profile bars are built from.

Frequently asked.