Scores documents using update signals: how often the content changes, how substantively, and how the change pattern relates to query freshness sensitivity. Foundational freshness-aware ranking that distinguishes living documents from stale ones.
Patent Overview
- Inventor
- Jeffrey Dean, others
- Assignee
- Google LLC
- Filed
- 2003
- Granted
- 2012-02-07
The Challenge
The Challenge
Some queries demand fresh results (news, events, releases); others reward established content (definitions, biographies). The scoring layer needs to distinguish which queries deserve freshness boosts and which documents deserve freshness credit.
- Stale Documents Misrank For Fresh Queries — When users query a current event, year-old documents that lead by raw relevance score under-serve intent. Freshness must enter the score.
- Trivial Updates Don't Make Content Fresh — Pages that timestamp every load or shuffle ads look 'updated' to a naive system. Update signal must distinguish substantive from cosmetic change.
- Freshness Sensitivity Varies By Query — Some queries are deeply freshness-sensitive (news); others are not (definitions). Per-query freshness weighting is required.
- Update Velocity Carries Information — Rapidly evolving topics produce rapidly updating documents. Update velocity itself is a quality and topicality signal.
- Manipulation Resistance Matters — If the system rewards updates blindly, sites will fake updates. Substantive-change detection plus per-domain trust gating prevents gaming.
Innovation
How The System Works
The system tracks per-document content changes over time, distinguishes substantive from cosmetic updates, computes update-velocity signals, classifies queries by freshness sensitivity, and applies a freshness-weighted score.
- Track Content Versions — Per crawl, store versioned snapshots. Diff between versions captures change magnitude.
- Distinguish Substantive From Cosmetic — Filter out timestamp-only, ad-only, and trivial changes. Surface real edits to body content, structure, and meaningful metadata.
- Compute Update Velocity — Per document, calculate update frequency and magnitude over rolling windows. Capture both burst and steady-update patterns.
- Classify Query Freshness Sensitivity — Per query, infer freshness sensitivity from query patterns, click behavior, and explicit topical signals. Output is a per-query freshness weight.
- Score Documents With Freshness — Multiply base relevance by a freshness factor proportional to query sensitivity and document update velocity.
- Apply Trust Gating — Per-domain trust attenuates freshness rewards. Low-trust domains earn less from frequent updates, preventing thin-update gaming.
- Decay Old Documents Appropriately — For freshness-sensitive queries, older documents decay; for evergreen queries, age is not penalized.
Freshness Is Query-Dependent
The patent's load-bearing idea is that freshness cannot be a global boost. It must vary by query and by domain trust. Per-query freshness sensitivity plus per-document update velocity combine into a tunable freshness factor.
Substantive Updates Win
Cosmetic edits and timestamp tricks don't earn freshness credit. The system rewards documents whose content actually changes in ways that matter.
- Query-Dependent Sensitivity — News queries reward fresh; definitions don't. Per-query freshness sensitivity is the gate.
- Substantive Change Detection — Diff filters cosmetic edits. Only meaningful content changes count as updates.
- Trust-Gated Reward — Per-domain trust attenuates freshness rewards. Low-trust domains earn less, preventing update-spam gaming.
Technical Foundation
Technical Foundation
The patent specifies the version store, diff classifier, update-velocity calculator, query-sensitivity classifier, freshness combiner, and trust gate.
- Version Store — Per-document content snapshots indexed by crawl time. Enables diff against any prior version.
- Diff Classifier — Categorizes changes as substantive or cosmetic. Filters timestamp-only and ad-shuffle changes.
- Update Velocity Calculator — Per-document, computes update frequency and magnitude over rolling windows. Outputs velocity score.
- Query Freshness Classifier — Per-query, infers freshness sensitivity. Output is a per-query freshness weight.
- Freshness Combiner — Multiplies base relevance by a freshness factor proportional to query weight and document velocity.
- Trust Gate — Attenuates freshness rewards by per-domain trust. Prevents low-trust domains from gaming update boosts.
The Process
The Process
Update tracking runs continuously; freshness application runs per query. Per-document velocity scores cache in the index; rankers consume.
- Crawl Document — Crawler fetches latest version. Version-store records snapshot.
- Diff Against Prior — Diff classifier categorizes change. Substantive changes count; cosmetic don't.
- Update Velocity Score — Velocity calculator updates per-document velocity score. Cached in index.
- Receive Query — Query arrives. Freshness classifier outputs per-query freshness weight.
- Score Candidates — Per candidate, base relevance times freshness factor (velocity * query weight).
- Trust-Gate Adjustment — Per-domain trust attenuates the freshness factor. Low-trust domains earn less.
- Sort And Return — Sort candidates by combined score; return top-N.
Quality Control
Quality Control
Update-based scoring is a prime target for manipulation. The patent specifies safeguards.
- Substantive-Change Filter — Diff classifier filters trivial edits. Cosmetic changes don't accumulate velocity.
- Trust Gating — Per-domain trust attenuates freshness rewards. Low-trust domains can't game freshness without first earning trust.
- Velocity Bounds — Velocity score is bounded. Spam-pace updates don't earn unlimited boost.
- Query-Classification Calibration — Per-query freshness sensitivity calibrates against click and dwell data. Mis-classifications surface as user-engagement regressions.
- Evergreen Protection — Evergreen content (definitions, tutorials) is not penalized for stability. Age-decay applies only to freshness-sensitive queries.
Real-World Application
Content-update scoring underpins every modern freshness layer. The primitives appear in news ranking, top-stories carousels, and the per-query freshness modeling that all major search engines deploy.
- Per-query Freshness Sensitivity — Each query carries its own freshness weight. News queries reward freshness; definitions don't.
- Substantive Update Filter — Only meaningful body and structure changes count as updates. Cosmetic edits are filtered.
- Trust-gated Reward Calibration — Per-domain trust attenuates freshness boosts. Low-trust domains can't update-spam their way up.
Why Real Updates Beat Date Tricks
Diff classifiers filter timestamp-only and ad-shuffle edits. Faking a publication date doesn't earn velocity score. Genuinely revising content does.
Why Evergreen Content Stays Strong
Per-query freshness sensitivity is the gate. For evergreen queries, stability is not penalized. The strategic lesson is to identify which content benefits from updates and which doesn't.
<\/section>What This Means for SEO
What This Means for SEO
This patent scores documents by how substantively and how often their content changes, gated by per-query freshness sensitivity and per-domain trust. SEO implication: meaningfully revise content that serves freshness-sensitive queries, but do not waste effort faking updates on evergreen or low-trust pages.
- Substantive Edits Only — A diff classifier filters out timestamp-only changes, ad shuffles, and trivial edits. Republishing with a new date or reshuffling a sidebar earns no freshness credit; rewriting body content, structure, or meaningful metadata does.
- Update What Freshness Queries Reward — Freshness sensitivity is per-query. News, events, and releases reward recent updates; definitions and tutorials do not. Identify which of your pages target freshness-sensitive intents and concentrate update effort there.
- Evergreen Content Is Not Penalized For Stability — Age decay applies only on freshness-sensitive queries. A stable, authoritative reference page is not punished for sitting still, so do not churn evergreen content just to look active.
- Update Velocity Carries Topicality Signal — Rapidly evolving topics produce rapidly updating documents, and the velocity is itself a quality and topicality signal. On fast-moving subjects, a genuine cadence of real updates helps you keep pace.
- Trust Gating Stops Update Spam — Per-domain trust attenuates freshness reward, so a low-trust site cannot thin-update its way up. Build trust before expecting frequent updates to convert into ranking.
- Velocity Is Bounded — The velocity score is capped, so spam-pace publishing does not earn unlimited boost. There is no benefit to mechanically editing pages at high frequency past the threshold.
- Real Revision Beats Date Tricks — The strategic move is to genuinely improve a page when the topic has moved on, not to manipulate a published-date field. Editorial substance is the only input that accumulates velocity.