# Google's 'nearest trusted seed' ranking patent expires this year. The leak shows the idea did not.

> A 2006 Google patent ranks pages by distance from a few trusted seed sites. It expires in 2026, but the leak shows the idea survived as PageRank-NearestSeeds.

Canonical: https://brandonlazovic.dev/articles/google-nearest-seed-patent/  
Author: Brandon Lazovic  
Published: 2026-07-17

## The short version

- Google's 2006 patent for ranking pages by their distance from a small set of hand-picked trusted seed sites expires in October 2026, but the 2024 Google leak shows the same idea lived on inside Search as a field named PageRank-NearestSeeds.
- The real mechanism is narrower and sharper than most coverage says: a page's rank decays exponentially with its distance to the k-th nearest trusted seed, computed as a single shortest path, not the sum-over-all-paths math of classic PageRank.
- Across the patent family and twelve write-ups from 2015 to 2025, none stated the actual formula and only three mentioned the load-bearing k-th-seed rule; the confirmed link from patent to live system is the leaked field name, not courtroom testimony.
- Labeled inference: the useful takeaway for AI citation is proximity, not authority scores. Reduce your distance in the citation graph to the sources an answer engine already trusts, which matters most in YMYL and medical topics where the trusted set is smallest.

I read Google's "nearest trusted seed" ranking patent and twelve SEO write-ups about it spanning 2015 to 2025, and not one stated the mechanism's actual formula; only three named its single load-bearing detail. That patent, US9953049B1, is one people periodically rediscover: it ranks pages by how far they sit, in the web-link graph, from a small set of hand-picked trusted "seed" sites, and its priority date is 2006, so its 20-year term runs out on October 12, 2026. [1] The recurring take is that it "proves" Google ranks by trusted sources. The more defensible and more interesting reading comes from the primary text plus one confirmed data point: the 2024 Google documentation leak revealed an internal attribute named `pagerank_ns`, expanded by the analyst who documented it as PageRank-NearestSeeds. [5] The patent describes the math and the leak shows the vocabulary survived inside Google; the coverage in between, as my tally shows, mostly gets both wrong. That gap is why this piece exists.

> Across the patent family and twelve SEO write-ups from 2015 to 2025, zero stated the patent's actual ranking formula, and only three mentioned the load-bearing detail that the score uses the k-th nearest seed, not the closest one.

<aside class="guardrail"><span class="lab">Accuracy guardrail</span> This is a patent as described, not a confirmed live ranking system. Three caveats travel with every claim below. The leaked <code>pagerank_ns</code> field is a real, confirmed attribute name, but the analyst who documented it flagged it as deprecated, so "Google uses this today" is not established. The phrase "distance from a known good source" that circulates as DOJ trial testimony traces to a single secondary outlet's paraphrase; I could not verify it against a primary transcript, so I label it suspected. And a patent expiring in 2026 does not mean the idea leaves the algorithm; expiry ends the monopoly on the method, not the use of it.</aside>

## What did Google actually patent?

Google patented a way to rank a page by the exponentially-decayed distance from a small set of trusted seed pages to that page, using the k-th nearest seed rather than the closest. The patent is US9953049B1, "Producing a ranking for pages using distances in a web-link graph," sole inventor Nissan Hajaj, assigned to Google. [1] It is a continuation of an earlier patent, US9165040B1, filed on the October 12, 2006 priority date and granted in 2015; the continuation was granted April 24, 2018. [2] Both share the title, the inventor, and the mechanism.

The mechanism has three moving parts. First, a small set of trusted seed pages is selected, and the patent is explicit that this is human work: the seeds "need to be reliable, diverse to cover a wide range of fields of public interests, as well as well-connected." The two examples the patent gives are the Google Directory and The New York Times. [1] Second, every link is assigned a length that grows with the number of outgoing links on the page it comes from, so a link from a page that points to five thousand others is worth far less than a link from a page that points to five. Third, the system finds the shortest total distance from the seeds to a page and scores the page as a decreasing function of that distance.

In plain terms: trust radiates outward from a few hand-picked authoritative sites, and the farther a page sits from them along the cheapest path, the more its ranking value decays. Hajaj's patent uses the distance to the k-th nearest seed, where k is greater than one, and gives its own reason: using a seed further down the list "facilitates suppressing unfairly high scores due to lack of proportionality at the vicinity of the seeds." [1] My labeled inference is that this also blunts manipulation, because earning a single link from a single trusted site cannot teleport a page to the top when the score depends on the k-th seed agreeing. The patent frames it as proportionality; the anti-gaming benefit is my read, not its words.

## Is this just PageRank, or Yahoo's TrustRank?

It is neither, and the confusion between the three is the single most common error in the coverage. Two clarifications matter. TrustRank is not a Google invention: it is a 2004 paper by Gyongyi and Garcia-Molina of Stanford and Pedersen of Yahoo, and it propagates trust from a human-vetted seed set using biased PageRank, an iterative computation over the whole graph. [3] Separately, Google does hold a patent with "trust" in its name, Ramanathan Guha's "Search result ranking based on trust" (US7603350B1), but that one ranks results by trust in the annotators of documents, not by link-graph distance, so it is a different machine entirely. [4] Neither is the Hajaj patent.

The real mathematical break is worth stating precisely, because it is where the loose paraphrases fail. Classic PageRank and TrustRank both sum contributions over every path through the graph; that is what the iterative eigenvector computation does. The Hajaj patent instead computes a single shortest path from the seed to the page, the way a routing algorithm finds one cheapest route. [1] Because the final score is the exponential of the negative distance, and exponentiating a sum turns it into a product, the score is effectively the product of per-hop "damping divided by out-degree" ratios along that one best path. So "links lose value the farther they are from trusted sites" is directionally right but imprecise. The value decays over cumulative, out-degree-weighted distance along a single best path to the k-th nearest seed, which is a much more specific claim than the phrase most write-ups reach for.

## Did any of it ever go live?

The honest answer is that one confirmed data point connects Google's 2018 nearest-seed patent (US9953049B1) to a real internal system, and everything else is either adjacent or unverified. The confirmed point is the 2024 leak. Google confirmed the authenticity of roughly 2,500 pages of internal Content Warehouse API documentation, and analysts catalogued an attribute named `pagerank_ns`, read as PageRank-NearestSeeds, a PageRank variant computed by proximity to seed nodes. [5] [6] [7] That is the patent's own idea, carried by name into Google's internal vocabulary. The same analyst flagged the field as deprecated, so its current use is not established, and I keep that caveat attached rather than dropping it for a cleaner headline.

Everything else is weaker, and I grade it honestly below.

| Signal | Status | Evidence and caveat |
|---|---|---|
| `pagerank_ns` / PageRank-NearestSeeds field | CONFIRMED (name), deprecation flagged | Named in the 2024 leak, whose authenticity Google confirmed; the documenting analyst marks it deprecated, so present-day use is unproven. [5] |
| `siteAuthority` site-level score | DIVERGENCE | A `siteAuthority` field appears in the leak, yet John Mueller spent 2016 to 2020 publicly and repeatedly denying Google uses any domain or site authority metric. The leak and the public statements point opposite ways. [8] [10] |
| PageRank as an input to the Q* quality system | LIVE | DOJ v. Google testimony, via secondary coverage, describes a Quality (Q*) system with PageRank among its inputs. [9] |
| "Distance from a known good source" as DOJ testimony language | SUSPECTED | This exact framing traces to one secondary outlet paraphrasing the testimony; I could not confirm it against a primary transcript. [8] |
| Follow-on Google patents extending the mechanism | PATENT-ONLY | The patent's forward citations are three later patents, none assigned to Google and none extending seed distance; no visible sign Google kept patenting the idea. [1] |
| E-E-A-T and authoritativeness | SPIRIT-LIVE | E-E-A-T is live as public rater guidance, but it is a human-rating rubric, not the seed-distance algorithm; the shared goal is not the same mechanism. [12] |

The pattern is that the direction is well supported and the specific mechanism is not confirmed as live. Gary Illyes of Google has said for years that a patent existing does not mean it is used in ranking, and that disclaimer applies here in full.

## What does a decade of coverage get wrong?

Most of the coverage flattens the mechanism into a slogan. I read the patent family and twelve SEO write-ups published between 2015 and 2025, all named and linked below, and classified how each described the mechanism: zero stated the actual ranking formula, only three named the load-bearing k-th-nearest-seed rule, and nearly all conflated the patent with PageRank or Yahoo's TrustRank. The specifics:

- Zero of the twelve stated the actual ranking formula, the exponential of the negative distance. Most give no math; the rest reach for "links lose value with distance" or, in one case, "most backlinks pass zero value." [20]
- Only three mentioned the load-bearing k-th-nearest-seed rule. Bill Slawski quoted the claim language directly in 2018, [15] a WebmasterWorld thread quoted it the same month, [17] and Lubble quoted it in 2022. [20] The other nine reduce it to "nearest seed" or "distance from seeds," which drops the entire proportionality safeguard. [14] [16] [19] [21] [22] [24]
- Nearly all conflate the patent with PageRank or Yahoo's TrustRank. Only the WebmasterWorld thread noted the tell that settles it: the patent "doesn't even use the word trust." [17]
- Only one named both of the patent's own seed examples, the Google Directory and The New York Times. [15] Several named The New York Times alone; the rest named neither.
- Only one connected the mechanism to the 2024 leak at all, and even that piece referenced the leak in general rather than naming `pagerank_ns` as the specific bridge. [23]

To be fair to the coverage, most of it hedged the right way on the question of whether the patent is live: piece after piece repeated some version of "a patent is not a confirmed ranking system." [13] [18] The field is more disciplined about "is it live" than about "what does it actually say." The formula and the k-th-seed rule are where the understanding thins out, and those are exactly the parts that change what you would do about it.

## What is the patent not saying?

It is not saying that a link from The New York Times ranks your page, and it is not saying the system is live. The seed examples are illustrations of the kind of page that qualifies, reliable and well-connected, not a shopping list of sites to get links from. The scoring depends on the whole shortest-path structure to the k-th nearest seed, not on a single prized link. Reading it as "get a link from a seed site" is the same category error as reading a recipe as a grocery receipt.

The expiration is also being over-read in advance. When the patent lapses in October 2026, the method becomes public prior art, which means anyone can implement it without licensing from Google. That is the opposite of the idea disappearing. If the leak is any guide, Google already moved the concept into named internal machinery years ago; the patent lapsing changes Google's legal exclusivity, not the presence of seed-distance thinking in modern ranking.

## Why this matters now: the 2026 expiry meets AI citation

The reason to care in 2026 is not the patent itself; it is that its core assumption has quietly become the operating logic of AI answer engines. There is a real 2026 appetite for reading old patents as a map of current behavior; Search Engine Land made the case this year that practitioners should "be an archaeologist" rather than a futurist. [11] Seed distance is a good artifact to dig up, because the assumption underneath it, that trust is something you inherit by proximity to an already-trusted core, is exactly how large models and answer engines behave when they lean on a narrow set of authoritative sources to ground an answer.

I call this citation-graph distance, and I label the framing as my own inference, not a confirmed Google mechanism. Where the 2006 patent measured hops in the web-link graph from a few hand-picked seed sites, an answer engine's practical trust signal is proximity in the graph of who cites and grounds whom. That is a reframing of the target, not a ranking factor I can prove; the evidence in this piece is the patent text and the leaked field name, not a measured citation outcome.

## Where the seed-distance logic bites hardest: YMYL and medical

The seed-distance model has the sharpest consequences in Your Money or Your Life topics, and especially in medical content, because that is where the trusted set is smallest and most conservative. In a low-stakes topic, the trusted core is broad and forgiving, and many paths lead back to a seed. In medicine, the sources an answer engine will ground on visibly narrow to a short list of institutional authorities, and everything else is measured by its distance from that list. This is my inference from how AI answers behave in health queries, not a documented seed set; I have no access to Google's or any engine's actual seed list.

The practical implication follows directly. In a YMYL niche, a new site is not competing on content quality in isolation; it is starting far from the trusted core with few short paths back to it, and closing that distance is slow, credential-driven work: being referenced by the institutions already close to the seeds, earning author-level recognition, and accumulating the kind of citations that shorten the path. It is the least gameable part of the whole picture, by design, and the k-th-seed rule is what makes it so.

## What should a practitioner actually do?

Less than the headlines imply, and the little that holds up comes straight from the mechanism. The patent does not support the two readings it keeps attracting: that a link from a named seed site ranks you, which the k-th-seed rule rules out because no single link teleports a page, and that the 2026 expiry retires the idea, which the leaked field name argues against. What it does support is a shift in target. If proximity to an already-trusted core is the signal, being cited and referenced alongside sources that are themselves close to that core matters more than any single trophy link, because the k-th-seed rule rewards broad, well-connected proximity over one strong connection. That is the least gameable part of the picture, and in YMYL and medical topics, where the trusted set is smallest, it is close to the only durable lever. This is not a new optimization checklist; it is the proximity logic the 2006 patent already described, read correctly for once.

## Sources

1. Patent US9953049B1 - 'Producing a ranking for pages using distances in a web-link graph' (Nissan Hajaj, Google), Google Patents — https://patents.google.com/patent/US9953049B1/en
2. Patent US9165040B1 - the 2006-priority parent in the same family (Nissan Hajaj, Google), Google Patents — https://patents.google.com/patent/US9165040B1/en
3. Gyongyi, Garcia-Molina (Stanford) & Pedersen (Yahoo) - 'Combating Web Spam with TrustRank', VLDB 2004 — https://www.vldb.org/conf/2004/RS15P3.PDF
4. Patent US7603350B1 - 'Search result ranking based on trust' (Ramanathan Guha, Google), a separate and mechanistically unrelated Google 'trust' patent, Google Patents — https://patents.google.com/patent/US7603350B1/en
5. iPullRank - Mike King, 'Secrets from the Algorithm: Google Search Documentation Leak' (documents the pagerank_ns / PageRank-NearestSeeds attribute; notes it as deprecated) — https://ipullrank.com/google-algo-leak
6. SparkToro - Rand Fishkin, 'An Anonymous Source Shared Thousands of Leaked Google Search API Documents' — https://sparktoro.com/blog/an-anonymous-source-shared-thousands-of-leaked-google-search-api-documents-with-me-everyone-in-seo-should-see-them/
7. Search Engine Land - coverage of the 2024 Google Search document leak — https://searchengineland.com/google-search-document-leak-ranking-442617
8. Hobo-Web - Shaun Anderson, 'Is Google PageRank Still Relevant in 2026?' (secondary analysis connecting pagerank_ns and DOJ testimony to seed distance) — https://www.hobo-web.co.uk/is-google-pagerank-still-relevant-in-2026/
9. Search Engine Land - 'The ABCs of Google ranking signals' (DOJ testimony on the Q* quality system and PageRank as an input) — https://searchengineland.com/google-abc-ranking-signals-455360
10. Search Engine Journal - John Mueller's repeated denials that Google uses a domain or site authority metric — https://www.searchenginejournal.com/domain-authority/246515/
11. Search Engine Land - Donna Rougeau, 'What patents reveal about the foundations of AI search' (the 2026 patent-archaeology framing) — https://searchengineland.com/patents-foundation-ai-search-471945
12. Google Search Central - E-E-A-T and the Search Quality Rater Guidelines — https://developers.google.com/search/blog/2022/12/google-raters-guidelines-e-e-a-t
13. SEO by the Sea - Bill Slawski, 'Recalculating PageRank' (2015, covers the parent patent US9165040B1) — https://www.seobythesea.com/2015/11/recalculating-pagerank/
14. Beyond Search / ArnoldIT - 'New Patent for a Google PageRank Methodology' (2015) — https://arnoldit.com/wordpress/2015/12/18/new-patent-for-a-google-pagerank-methodology/
15. SEO by the Sea - Bill Slawski, 'PageRank Updated' (2018, quotes the k-th-closest-seed claim language and names both seed examples) — https://www.seobythesea.com/2018/04/pagerank-updated/
16. Search Engine Journal - Roger Montti, 'PageRank Patent Update - How it Impacts SEO' (2018) — https://www.searchenginejournal.com/pagerank-patent/250603/
17. WebmasterWorld - 'Google Updates PageRank Patent' forum thread (2018; the one discussion that quotes the k-th rule and notes the patent never uses the word trust) — https://www.webmasterworld.com/google/4897980.htm
18. Search Engine Journal - Roger Montti, 'Link Distance Ranking Algorithms' (2019) — https://www.searchenginejournal.com/link-distance-ranking-algorithms/308517/
19. SEO Werkz - 'Basics On Reduced Link Graphs, Part 2' (2021) — https://www.seowerkz.com/resource/uncategorized/basics-on-reduced-link-graphs-part-2/
20. Lubble - 'Google's Web-Link Distance Model: Why Most Backlinks Pass Zero Value' (2022; quotes the k-th language) — https://lubble.com/blog/breaking-down-backlink-clusters/
21. Shopify - 'Google's PageRank Algorithm: What You Need To Know' (2023) — https://www.shopify.com/uk/blog/pagerank
22. KeywordsPeopleUse - Edd Dawson, 'PageRank - What it is & How it works' (2024) — https://keywordspeopleuse.com/seo/guides/pagerank
23. Semrush - Erika Varangouli, 'Google PageRank in 2025: What Google Search Leak Reveals' (the one write-up that ties the mechanism to the 2024 leak) — https://www.semrush.com/blog/pagerank/
24. Kopp Online Marketing - Olaf Kopp, patent summary of 'Producing a ranking for pages using distances in a web-link graph' — https://www.kopp-online-marketing.com/patents-papers/producing-a-ranking-for-pages-using-distances-in-a-web-link-graph