How we test BBQ sauces

Discovery

We start from a hand-curated brand list — currently 79 manufacturers, chosen to cover the high-intent BBQ-sauce keyword clusters (sugar-free, Carolina, Alabama white, Korean, Japanese, Kansas City, Texas, and the major brand-name searches). Every brand's homepage is verified before it enters the list.

For each brand we then either crawl the storefront's product index directly (Shopify products.json, sitemap, or category page), or auto-detect a catalog URL when the homepage isn't itself a catalog (e.g. McCormick-owned brands hosted on the corporate site).

Classification

Every discovered URL is classified by an LLM gate against a strict schema: is this a BBQ sauce, or is it merch, a rub, a marinade kit, a gift card, or a bundle of multiple SKUs? The gate runs against page metadata plus a short snippet of the page body, so we don't pay extraction cost on URLs that won't ever be exported. Bundles and variety packs are flagged separately and excluded from consumer-facing rankings because they have no single-bottle ingredients or nutrition to verify against.

Extraction

Each classified sauce runs through a multi-step extract agent. The product page HTML is parsed for structured data (JSON-LD, microdata, Shopify variant payloads) for unambiguous fields — brand, name, GTIN, product images. Ingredients and nutrition come from the published label text where the merchant exposes it, or from OCR on the label image when they don't.

Label data and the Open Food Facts fallback

We always prefer the manufacturer's own published label. When the merchant doesn't publish one and OCR cannot recover one from the product photograph, we fall back to Open Food Facts, a community-maintained product database. Open-Food-Facts data is contributor-supplied and can lag the current bottle, so every field sourced this way is tagged with the external status and shows a citation linking back to the upstream entry on the sauce page. Sauces with no first-party label and no usable OFF entry render the corresponding section as “not yet captured” rather than as a fabricated value.

How heat, sweetness and smoke are scored

Each sauce carries three 0–5 scores for heat, sweetness and smoke. The scores are derived from the sauce's data, not from first-hand tasting. An LLM reads the published ingredient list, label callouts (“extra hot”, “sweet heat”, “competition-style”), and the brand's product description, then emits a score on a fixed scale against a controlled vocabulary in our schema.

The scale anchors:

• Heat (0–5) — 0 is no heat at all, 2 a mild background warmth (cayenne in trace amounts), 3 a clear kick noticeable on the first bite, 4 hot enough that most people will reach for water, 5 a dedicated hot sauce (habanero, scotch-bonnet, ghost, reaper). Signals: capsaicin sources in the ingredient list, their position (earlier → more), and any explicit Scoville claim on the label.
• Sweetness (0–5) — 0 is bone-dry (vinegar-forward Carolina styles), 2 a balanced sauce, 3 the Kansas-City baseline, 5 a candied glaze. Signals: the position of sugar/molasses/honey/maple in the ingredient list, total sugars on the nutrition label when available, and label terms like “sweet”, “honey” or “sticky”.
• Smoke (0–5) — 0 is no smoke character at all, 2 a hint of smoked paprika, 3 a clear wood-smoke note, 4 strongly smoky, 5 a pit-smoked or liquid-smoke-forward sauce. Signals: smoked paprika, chipotle, natural smoke flavor, and label terms like “hickory”, “mesquite” or “applewood”.

These scores are estimations from data, not sensory findings. A sauce that lists chipotle deep in the ingredient list will score higher on smoke than one with none — but the only way to know how it actually behaves on a brisket is to taste it. Treat the meters as a comparison aid across the catalog, not as a promise about a specific bottle.

How we score web opinions

The “What reviewers say” block on each sauce page is AI-synthesized from third-party sources, not our own first-party review. For each sauce we search the open web for independent coverage — major retailers, food blogs, BBQ forums, recipe sites — and an LLM aggregates the cited passages and ratings into a structured digest: a consensus rating, overall sentiment, pros, cons, and direct quotes attributed back to the source URLs.

Quality checks that ship with every digest:

• Sources are required. A digest only renders when we have at least one verifiable source URL. The displayed count is the number of distinct sources backing the rating, not a free-floating “reviews found” number that could be inflated upstream.
• Quotes link to originals. Every highlighted quote in the digest carries a direct link to the page it was pulled from, so readers can audit the synthesis themselves and check that the LLM hasn't drifted from what the source actually said.
• No verbatim quotes from restricted platforms. Reviews on social networks and large retail sites (Amazon, Walmart, Target, and others) inform the aggregate rating, sentiment, and pros/cons, but we never republish their review text verbatim. Highlighted quotes come only from blogs, forums, and other editorial sources, and a deterministic filter enforces this after synthesis.
• We don't claim it as our own review. The block is labelled “From around the web”, and the consensus rating is presented as an aggregation of third-party opinion, not as a BBQ Sauce Scout star rating. First-party ratings, when they exist, come from the comment form lower down the page.

Brand enrichment

Brand pages get a separate enrichment pass that gathers a short editorial summary, founding facts, and notable distinctives from the brand's own homepage, its about page, and its Wikipedia article when one exists. Every field is nullable — if the sources don't clearly state it, the brand page shows nothing rather than something the LLM invented.

Ranking

Guide rankings use the criterion the guide is named after — lowest sugar for keto guides, lowest carbs for low-carb guides, web-opinion consensus rating for style guides. Sauces that don't meet a guide's quality gate (verified or OCR-captured nutrition and ingredients where the gate requires it) are excluded entirely.

All sauces ship with a "captured at" timestamp visible on the sauce page, and every guide shows a "last updated" date driven by the newest capture in its pick set. Formulations change; we re-crawl the source brands periodically and label drift will resurface as updated rankings.