About Our Scoring

The Som Score is a proprietary composite rating on a 100-point scale that represents our calculated estimate of a wine's critical consensus. It aggregates scores from established industry publications, professional critics, and community platforms into a single, comparable number.

We source critic scores and reviews from publicly available industry publications and community platforms. Our data collection operates in two tiers:

Tier 1 — Direct Source Retrieval: We query established wine databases and review aggregators for published critic scores, including Wine Spectator, Wine Advocate, Wine Enthusiast, James Suckling, CellarTracker, and Vivino community ratings, among others.

Tier 2 — AI-Assisted Web Search: When direct sources do not return results, we use AI-powered web search to locate publicly available scores from critic publications and retailer listings. This tier uses Google Search grounding to find real, published data — it does not fabricate or estimate scores.

Every score we collect is attributed to its original source. We do not create, invent, or hallucinate critic scores. If we cannot find published data for a wine, it is marked "New to Som" rather than assigned an estimated rating.

After collecting raw scores from multiple sources, we use AI to assist in calibrating and normalizing the data. Different publications use different scales, methodologies, and grading philosophies. Our calibration process accounts for these differences to produce a meaningful composite.

What AI does: AI assists in processing and normalizing scores across different rating scales (e.g., 5-star community ratings vs. 100-point critic scales), detecting outliers, and weighting sources based on their historical reliability for specific wine regions and styles.

What AI does not do: AI does not generate, fabricate, or predict scores. The underlying data always originates from human critics and community reviewers. AI is a tool in our pipeline — not the source of truth.

Our proprietary Sommelier AI layer applies sommelier-informed heuristics to calibrate the composite score. This step considers factors that raw score aggregation may miss:

Regional expertise — certain critics are recognized authorities for specific appellations and their scores may carry more weight for those regions. Vintage context — the Sommelier AI considers vintage quality and producer reputation when calibrating between sources. Style alignment — community ratings and professional critics sometimes diverge on style preferences; the Sommelier AI balances these perspectives.

The Sommelier AI uses models trained on sommelier knowledge and wine industry standards. It is not a panel of human sommeliers reviewing each bottle individually, but rather an automated system informed by professional sommelier expertise.

Every Som Score includes a confidence indicator that reflects the breadth and reliability of the underlying data:

High confidence — Multiple established critics and publications have reviewed this wine. The score is well-supported by diverse, independent sources.

Medium confidence — Some professional reviews exist, possibly supplemented by community ratings. The score is reasonable but based on fewer data points.

Low confidence — Limited data is available. The score should be treated as a rough estimate rather than a definitive assessment.

When no reliable data can be found, we do not assign a score. These wines appear as "New to Som" until sufficient data becomes available.

We believe in being straightforward about what the Som Score is and what it is not:

It is a useful tool for comparing wines, discovering highly-regarded bottles, and understanding critical consensus at a glance.

It is not a substitute for your own palate, a professional appraisal, or investment advice. Wine is subjective, and no algorithm can capture the full experience of tasting.

Our scoring methodology is continuously refined as we incorporate new data sources and improve our calibration models. If you believe a score is inaccurate, you can use the Refresh Scores button on any wine's detail page to trigger a fresh data collection.

Beyond scoring, Som uses AI to power several features throughout the app:

Tasting Profiles — AI-generated flavor descriptors (nose, palate, finish) based on grape variety, region, and vintage characteristics. These are estimates informed by wine knowledge databases, not from tasting the actual bottle.

Food Pairings — AI-suggested food pairings based on the wine's style, body, and flavor profile. These are recommendations, not guarantees of a perfect match.

Drink Windows — AI-estimated optimal drinking periods based on vintage, grape variety, region, and producer track record. Actual aging potential varies by storage conditions and individual bottle variation.

Recipe Generation — AI-created recipes designed to complement specific wines. These are generated suggestions, not tested recipes from a professional kitchen.

All AI-generated content in Som is clearly identified and provided for informational and entertainment purposes. For authoritative guidance, we recommend consulting a professional sommelier or wine advisor.