Patent US7346839B2: How Google Uses Time Against Link Builders

The Historical Data Patent

If you've read my On-Page SEO guide, this patent sits on the other side of the equation — not what's on the page, but what points to it and how that changes over time. It is one of the foundational off-page patents, and the 2024 Google API leak contains attributes that strongly correspond to every major mechanism.

The Honest Hedge

Every analysis has a threshold where certainty ends and inference begins. Here's where that line falls for this patent:

---

---

Patent Metadata

Eight inventors. 56 claims. 472 citing families. Filed the last day of 2003, this patent is older than YouTube, Chrome, and the iPhone — yet I still see its fingerprints in client data every month. The inventors aren't just any Google engineers: Matt Cutts was head of Google's webspam team for a decade. Jeffrey Dean is the architect behind TensorFlow and two of Google's core infrastructure systems. Paul Haahr was the VP of Search Quality who later testified in the DOJ antitrust trial. When these three co-author a patent, it becomes infrastructure.

---

What This Patent Does (Plain English)

Most search patents tackle one thing. This one tackles everything that changes over time — which is almost everything. At its core, it describes a system that monitors and scores documents based on their history. Not what the page says today, but how what points to it has changed, how the domain has changed, and whether the patterns look natural.

Here's what the system does:

1. Monitors when links appear and disappear — tracking the dates and the rate of change
2. Weights links by freshness — newer links from recently updated pages carry more signal than stale ones
3. Detects velocity anomalies — distinguishing between a topical spike (CDC during an outbreak) and a purchased link campaign
4. Catches rotating links — sites that swap out their "featured link" daily get weighted lower than consistent editorial links
5. Monitors DNS records — frequent registrar changes, name server swaps, and bulk registrations flag a domain as potentially illegitimate
6. Detects expired/doorway domains — throwaway domains used for spam get identified by their registration patterns
7. Flags ranking jumps — a document that suddenly ranks across many queries could be topical or an attempt to spam

---

Link Velocity: The Core Signal

The phrase that launched a thousand SEO blog posts comes from this patent:

This is important: the patent explicitly distinguishes between spikes that are topical and spikes that are manipulative. It doesn't say "all fast link growth is bad." It says the system monitors the rate and decides.

The patent describes three specific aspects of link velocity monitoring:

1. Link Appearance and Disappearance

Every link gets a timestamp. Google tracks not just when links appear, but when they disappear — and it draws conclusions from both signals. A page that steadily loses inbound links over time is flagged as stale.

2. Rotating Links Detection

This one is subtle. If a page changes which sites it links to on a daily or weekly rotation, the freshness of each individual link doesn't save it — the rotation pattern itself is a negative signal. This catches paid link placements that rotate sponsors, blogroll-for-sale schemes, and any arrangement where links are temporary by design.

I've seen this firsthand. When I was early in my career and didn't know better, I bought 12 PBN backlinks with a variety of keyword-rich anchor text. The provider fed them over about 21 days. By the time the last link was live, the page that was sitting at position 9 started dropping. I built a couple of legitimate guest posts, and it eventually stabilized around position 21. I got hit on both velocity of link acquisition and anchor text velocity simultaneously. That experience taught me something this patent makes explicit: the system doesn't look at any single signal in isolation.

3. Link Staleness and Disappearance

This creates a cascade effect. If a page accumulates many inbound links but then those linking pages themselves become stale, the target page loses equity from two directions: the direct loss of links, and the devaluation of the remaining ones because the linking pages are now considered stale themselves.

---

Freshness Weighting: Not All Links Age the Same

Most SEOs think about link building in terms of quantity: more links equals more authority. This patent introduces a variable most ignore — time decay.

The patent goes further with a genuinely elegant observation:

Read that again. The patent is saying: if a page gets updated but the link to your site remains unchanged, that's a stronger signal than a newly placed link. The link has survived editorial review. Someone updated the page and decided the link was still worth keeping. That persistence is a freshness signal — the link gets treated as fresh because the containing page was recently updated.

This flips the common SEO narrative. Among the most valuable links are those that survive editorial updates — not because age alone matters, but because persistence through changes is itself a freshness signal. A link that's been in a Wikipedia article for eight years, through hundreds of edits, carries an implicit freshness advantage that no freshly built guest post can match.

---

Domain-Level Signals: DNS, Name Servers, and Doorway Domains

This is where the patent gets genuinely surprising. Most SEOs associate US7346839B2 with link velocity. But a large section of the patent is dedicated to domain infrastructure monitoring — signals derived from DNS records, hosting history, and registration patterns that are independent of page content.

DNS Record Monitoring

Frequent registrar transfers, WHOIS changes, and hosting swaps all factor into a domain legitimacy assessment that operates independently of content quality. For practitioners who buy and sell domains, this isn't theoretical — it's the reason why some domain acquisitions require a stabilization period before the content starts performing.

Name Server Profiling

This is reputation by association. The other domains on your name server affect how Google evaluates your domain. If your name server's portfolio is a mixed bag of legitimate businesses — law firms, restaurants, tech companies — that's a positive indicator. If it's predominantly one registrar, commercial keyword domains, or recently created bulk registrations, that's a negative flag. Your DNS neighborhood matters.

Doorway Domain Detection

I've bought my fair share of expired auction domains over the years. Here's what I've learned that this patent helps explain: there's a critical distinction between expired and auctioned. An auctioned domain changes hands, but it doesn't go through the full expiration cycle. An expired domain actually lapses — the registration drops, it sits in limbo, and then someone re-registers it. The success rate of bringing a truly expired domain back to life is dramatically lower than an auctioned one. The topic of the new site needs to be nearly one-to-one with the old site. With auction domains, you can get further from the original topic, but the same principle applies: the more the new site resembles the old one, the more likely the historical equity survives.

---

Ranking Jump Detection

Again, the patent uses the same framework as link velocity — a sudden change could be legitimate or manipulative, and the system's job is to distinguish between the two. Like the CDC/SARS example for link spikes, a page that jumps across many queries during a major news event behaves differently from one that jumps because someone purchased a batch of exact-match anchor text links.

I've seen clients receive a large spike of referring domains inside a 30-day window, and the traffic immediately dropped — no core update, no algorithm change, nothing else happened. The only variable was a flood of backlinks. When I dug deep into those backlinks, the majority were from extremely questionable sources with rich anchor text. I performed surgical disavows on every one of those properties, and all of them recovered without exception.

---

Historical Data SEO Implications: What This Means for Your Link Building

1. Velocity Alone Isn't the Trigger — Context Is

The patent's CDC example shows that Google distinguishes between topical spikes (viral content, product launches, PR campaigns) and manipulative ones. The patent itself mentions checking "news articles, discussion groups, etc." on the theory that spam documents won't be mentioned in the news. Modern Google likely extends this to a far broader set of corroborating signals. When something launches on Product Hunt and goes viral, Google can see the Product Hunt page mentioning your brand, the referral traffic, and the backlink spike all happening simultaneously. Cross-referencing entities is trivially easy for Google. Faking that constellation of signals — buying social signals, listicle placements, Facebook ad traffic to justify the links — requires more effort than building something that actually earns the attention.

2. Link Persistence > Link Volume

The freshness weighting mechanism rewards links that survive editorial updates over links that are simply new. A five-year-old link from a regularly updated resource page is worth more than five one-month-old links from sites nobody maintains. Build the kind of content that people keep linking to — not the kind that gets linked once and forgotten.

3. Domain Infrastructure Is an Independent Signal

Your DNS history, hosting stability, and name server neighborhood contribute to a domain legitimacy score that operates independently of your content quality. If you're acquiring domains, allow a stabilization period. Avoid rapid registrar changes. And check what else your hosting provider is hosting — your DNS neighborhood is part of your reputation.

4. Rotating Link Placements Are Devalued

Sponsored sidebar links that rotate monthly, blogroll exchanges that shuffle, and any link placement that changes regularly gets weighted lower than a consistent editorial link. If you're paying for link placement, a permanent editorial mention is worth more than any rotating sponsorship, even if the sponsorship generates more raw link count.

5. The Disavow Tool Still Works — When It Needs To

For older link profiles with pre-2022 toxic backlink spikes, surgical disavows still have a demonstrable effect. I've recovered every site where I identified and disavowed the velocity-triggering links. The historical velocity system described in this patent creates the conditions that make disavow effective — when the damage is from identifiable, time-stamped link patterns rather than from modern ML-based evaluation.

---

Google API Leak Cross-Reference: AnchorSpamInfo and RegistrationInfo

The 2024 Google API leak — first reported by Rand Fishkin and investigated by Mike King at iPullRank — revealed attributes across 8 patent mechanisms, with 5 direct thematic matches, 2 extensions, and 1 gap. This is one of the strongest patent-to-API correspondences in the entire leak:

Patent Mechanism	API Attribute	Alignment
Link appearance/disappearance dates	firstseenDate / creationDate / deletionDate in AnchorsAnchor	✅ CONFIRMED
"Legitimate docs attract links slowly"	phraseAnchorSpamDays / phraseAnchorSpamRate in AnchorSpamInfo	✅ CONFIRMED
Link freshness weighting	firstseenNearCreation boolean in AnchorsAnchor	✅ CONFIRMED
DNS record monitoring	RegistrationInfo.createdDate / expiredDate	✅ CONFIRMED
Doorway domain detection	expired boolean per anchor	✅ CONFIRMED
Ranking jump detection	Q* scoring system (quality rater integration)	🔶 API EXTENDS
Rotating links detection	droppedRedundantAnchorCount	🔶 API EXTENDS
Name server profile analysis	No direct match found	📜 PATENT ONLY

The most telling attribute is phraseAnchorSpamDays — it measures the number of days for 80% of a page's anchor text to accumulate. Combined with phraseAnchorSpamRate, this strongly corresponds to the velocity detection concept described in the patent. And the API reveals something the patent didn't anticipate: demotedStart and demotedEnd dates — velocity-based demotions have explicit start and end dates. They're temporary, bounded events. They're not permanent penalties.

---

Citation Network

Patent Family Chain

US20050071741A1 (application, 2003) → US7346839B2 (this patent, 2008) → 6 continuation patents covering document scoring (US7797316B2), content freshness (US7840572B2, US8112426B2), anchor text analysis (US8051071B2), domain trust (US8316029B2), and historical content signals (US8082244B2).

Forward Citations (Key Patents Citing This One)

Patent	Relevance
US9953049B1	Seed Distance PageRank — combines temporal signals with trust distance calculation
US8577893B1	Reference Contexts — adds surrounding text fingerprinting on top of temporal monitoring
US7716225B1	Behavioral Link Weighting — uses ML to weight links by position, font size, and click probability

Related Articles on This Site

• US11409748B1 (Passage Ranking) — where this patent evaluates the link graph over time, Passage Ranking evaluates the on-page heading structure. Together they represent the off-page + on-page scoring foundations.
• US8661029B1 (NavBoost) — NavBoost adds behavioral user signals on top of the temporal link signals from this patent. A link that users actually engage with (NavBoost) that is also temporally consistent (this patent) carries maximum weight. See also: How NavBoost Really Works.
• US10235423B2 (Entity Scoring) — Entity Scoring uses knowledge graph entities to validate the relationship between linking and linked pages. This patent watches when links change; Entity Scoring watches who those links connect.
• US9767157B2 (Panda) — Panda evaluates on-page content quality. This patent evaluates off-page link quality over time. A page can survive Panda and still fail this patent's velocity checks — and vice versa.
• Quality Scoring Ensemble — the ensemble system that combines signals from all quality scoring components. This patent's temporal link data feeds into the same quality aggregation pipeline that Panda, NavBoost, and Entity Scoring contribute to.
• US7603350B1 (Entity Trust) — This patent's trust decay mechanism has an entity-level counterpart in the Entity Trust patent. Trust relationships between people also decay over time — and sudden appearances in the entity trust graph may trigger the same temporal flags that Historical Data applies to links.

---

Historical Data: What Doesn't Matter as Much as SEOs Think

The nature of this patent is a philosophical bet that Matt Cutts and Jeffrey Dean placed in 2003: things that happen naturally leave different patterns than things that are engineered. Legitimate documents attract links slowly. Legitimate domains keep their registrar for years. Legitimate links survive page updates. The nature of this insight hasn't changed and won't change — because it describes how the web, as a human-created information ecosystem, actually behaves.

The flavor — specific velocity thresholds, the exact definition of a "doorway domain," the name server profiling criteria — was the 2003 approach. And honestly, so many new systems have been put on top that the effects this patent originally aimed to achieve have been greatly nullified by more sophisticated mechanisms. SpamBrain is a neural network. This patent is if-then logic. The philosophy survived; the implementation has been absorbed into something far more complex.

That's where I genuinely push back on the SEO community's treatment of this patent. It gets cited constantly — usually the "legitimate documents attract backlinks slowly" line — as if it's the final word on link velocity. It isn't. I've bought expired auction domains. I've seen domain changes that the patent says should reset link equity but didn't. I've watched velocity that should trigger flags pass without any visible effect. The patent says a lot of things that reality doesn't exactly follow.

That's the point. Not that the patent is wrong — it describes a real system with real API confirmation. But it's a 2003 foundation with 22 years of construction on top. Don't just take patent interpretations — from LinkedIn gurus, from SEO bloggers, from me — at face value. Test it. Acquire an expired domain and rebuild the same site. Buy an auction domain and change the topic. Observe what happens in Search Console. Nothing beats lived experience, and 2003 is dinosaur-era Google.

---

Frequently Asked Questions