Computes a per-site quality score from two ratios: queries that refer to the site versus all queries the site appears for, and queries that lead to user selection of site resources versus queries the site is associated with.
Patent Overview
- Inventor
- Navneet Panda
- Assignee
- Google LLC
- Filed
- 2012-09-28
- Granted
- 2015-05-12
- Application Number
- US 13/631,492 (related)
The Challenge
Quality Needs A Number, Not Just A Label
The Panda ranking framework needs a concrete numerical site quality score that the runtime ranking can consume. Hand-classifying sites as high or low quality does not scale. The system needs to derive a continuous quality score from observable signals in query and click data that captures whether real users find the site to be a satisfying destination versus a transient stop.
- Need A Continuous Score, Not Binary Labels — Binary quality classification cannot capture the gradient between great sites, decent sites, and mediocre sites. A continuous score lets the ranking apply graded penalties or boosts.
- Query Behavior Reveals Site Standing — How users behave with respect to a site in search behavior is the strongest available quality signal. Queries that refer to the site, and selections of the site's resources, reveal the audience's revealed preference.
- Two Independent Ratios — One ratio measures whether the site is queried for directly; another measures whether the site is selected when it appears in results. Both are needed because each catches a different failure mode.
- Score Must Be Robust To Volume — Sites of very different sizes need comparable scores. Using ratios (not raw counts) normalizes across scale.
Innovation
Two Ratios From Query Behavior
The system determines a first count of unique queries received by the search engine that are categorized as referring to a particular site, and a second count of unique queries associated with the site (queries that were followed by user selection of the site's resources). The quality score is computed from these two counts (typically as ratios against denominators like total queries the site appeared for). The score is the per-site quality value Panda consumes.
- Identify The Site — Establish the site boundary (typically domain). All counts will be computed relative to this site.
- Count Unique Referring Queries — Determine the number of unique queries received by the search engine that are categorized as referring to the site. A query refers to a site when it names the site directly or unambiguously targets the site.
- Count Unique Selecting Queries — Determine the number of unique queries associated with the site, where association is defined by the query being followed by user selection of a search result that identifies a resource in the site.
- Compute Quality Ratios — Form ratios from the counts: referring queries over total queries the site appeared for; selecting queries over queries the site appeared in. Both ratios should be high for a quality site.
- Combine Into Quality Score — Combine the ratios into a single quality score per site. The score is a scalar consumable by ranking.
- Refresh Periodically — Recompute the score on the configured refresh cycle so the signal tracks the site's evolution.
Two Behavioral Ratios Per Site
The quality score is built on the two ratios that capture how the audience treats the site: do they search for it (referring), and do they pick it when they see it (selecting). The combination is the per-site quality signal Panda uses.
Referred AND Selected
A quality site is one that users search for by name AND select when they see it in results. Either ratio alone is weaker; both together is the signal.
- Referring Query Ratio — Fraction of queries that name or unambiguously target the site. Captures brand-level demand.
- Selecting Query Ratio — Fraction of queries the site appeared in that were followed by user selection of the site. Captures the audience's preference among alternatives.
Technical Foundation
Counts And Ratios
The score is derived from per-site counts of two query types.
- Referring Queries — Queries that name the site or unambiguously target it. Brand-name queries and site-specific queries dominate this category.
- Selecting Queries — Queries that resulted in the user clicking on a search result from the site after the result was presented.
- Quality Score — Combined scalar from the referring and selecting ratios. Consumed by ranking as the site's quality value.
Key Insight: The two ratios are complementary. Brand-targeting (referring) queries reveal name-level demand. Selection (selecting) reveals competitive preference. A site can have one without the other (a popular brand whose pages don't actually satisfy users, or a useful site nobody knows by name). The quality score requires both.
<\/section>What This Means for SEO
What This Means for SEO
Site quality scoring is the per-site number that drives much of modern ranking. Knowing the two ratios reveals the leverage points for moving the score.
- Brand Search Volume Is A Quality Signal — Searches that name your brand or domain contribute to the referring query count. Brand-building (PR, content marketing, audience development) feeds the brand-search demand that drives the referring ratio.
- Click-Through Rate Per SERP Position Matters — When your result is presented and the user clicks, the selecting ratio strengthens. Above-average CTR for your position adds to the quality score; below-average CTR drags it.
- Snippet And Title Quality Drive Selection — Selection happens after the user reads your title and snippet. Both should be compelling and accurate. Misleading or boring snippets sacrifice the selection signal.
- Existing In Many SERPs Without Selection Hurts — If your site appears for many queries but doesn't get selected, the selecting ratio drops. Better to rank for fewer queries with high selection rates than to rank for many queries with poor selection.
- Audience-Defined Targeting Beats Keyword Stuffing — A site that targets queries its audience actually searches for produces both referring queries (brand) and selecting queries (CTR). Keyword stuffing produces appearance without selection and dilutes the score.