chard: Google's Page-Level Content Quality Model

What is chard?

chard is Google's page-level content quality model. It rates the quality of individual pages, calibrated against the judgments of human quality raters that Google trains according to the Quality Rater Guidelines (QRG). The name 'chard' — a leafy green vegetable — follows Google's pattern of naming systems after foods, similar to 'tofu' and 'kale'.

The Google API leak reveals two chard fields: chardScores (the quality scores themselves) and chardVariance (a confidence measure showing how certain the model is in its verdict). Together, these two fields provide both a quality assessment and a reliability indicator for that assessment.

chard is specifically designed to evaluate content quality, not link quality or user behavior. It assesses the text, structure, and substance of a page — whether the main content is helpful, whether supplementary content adds value, and whether the page demonstrates the E-E-A-T signals described in the QRG.

The calibration chain: IS-Calibration → chard → NSR

chard is fed by IS-Calibration (System 12) — the human rater system that provides ground truth for quality models. This means chard's quality scores are ultimately derived from human judgments: raters evaluate pages according to the QRG, and chard learns to predict what a rater would say about a new page.

chard feeds into NSR (System 7), the normalized site rank. This creates a three-level quality chain: IS-Calibration provides the ground truth (human judgments), chard produces page-level quality predictions, and NSR aggregates these into site-level quality scores. Each level transforms the signal: rater judgments become model predictions, and model predictions become site-level aggregates.

The chardVariance field is architecturally significant. A high variance score means the model is uncertain about its quality assessment — perhaps because the page type is unusual, the content is ambiguous, or the page doesn't match the training data well. This uncertainty can be used downstream: NSR may weight high-variance chard scores less heavily, or flag them for additional rater review.

YMYL: Your Money Your Life pages

chard is especially relevant for YMYL (Your Money Your Life) pages — pages that could potentially affect the user's health, financial stability, or safety. The QRG sets higher E-E-A-T standards for YMYL pages, and chard's calibration reflects this: a YMYL page must demonstrate higher expertise, authoritativeness, and trustworthiness than a non-YMYL page to receive the same quality score.

YMYL classification is not binary — it's a spectrum. Medical information about treating a disease is clearly YMYL. A recipe for chocolate cake is clearly not. But financial tips, legal information, and news about current events fall on a spectrum that chard must navigate.

The chard + YMYL combination creates a specific risk for publishers: content that is technically accurate but lacks the E-E-A-T signals expected for its YMYL category will receive a lower chard score. A medical article written by a non-expert, even if factually correct, may be rated lower than one written by a board-certified physician — because the QRG instructs raters to evaluate the author's credentials for YMYL content.

chard in the ranking architecture

chard operates at the document level (Reach: Doc) — it produces quality scores for individual pages, not for entire sites. This is architecturally distinct from Panda (site-level) and Q* (site-level): chard evaluates each page on its own merits, while Panda and Q* aggregate across the site.

The E-E-A-T dimension mapping for chard is Expertise (Exp: primary) — the primary signal chard contributes is whether the page demonstrates genuine expertise. The indirect Trust signal (T: indirect) comes from the fact that quality content is more trustworthy, but trust is not chard's primary function.

chard has no upstream dependencies beyond IS-Calibration (fedBy: [12]). This makes chard a focused, single-source model: it takes rater calibration as input and produces page-level quality predictions. The simplicity of this architecture — one input, one output — is characteristic of quality models that need to be transparent and debuggable.

Implications for SEO practitioners

chard's calibration chain means that content quality is ultimately defined by the QRG. The most effective SEO strategy for chard is to understand what the QRG describes as high-quality content and to produce content that matches those criteria — not to try to reverse-engineer the model.

The chardVariance field has an important implication: pages that are unusual or don't fit common patterns may receive high-variance scores. This is not inherently bad — it means the model is uncertain, not that the page is low quality. But it does mean that unusual content may take longer to rank, as the system needs more data to confidently assess it.

For YMYL content, the QRG's emphasis on author credentials means that authorship signals matter. Bylines, author bios, and author pages that demonstrate expertise are not just good UX — they are signals that chard may use to assess whether the page meets YMYL E-E-A-T standards.

The Hoax detection component means that content that is factually incorrect or misleading — even if well-written — will be penalized. This is particularly important for news publishers: a well-written article that contains factual errors will receive a lower chard score than a less polished but factually accurate article.