Patent US9558233B1: The Click Test — How Google Decides Whether Your Backlinks Count

The Click-Validated Quality Patent

If you've read my breakdown of the Reasonable Surfer patent, that system predicts which links should get clicked — from font size, position, where they sit on the page. This one works from the other direction: it measures whether links actually get clicked, and uses that to judge whether a linking page can be trusted at all. Predicted clicks and measured clicks — the a-priori guess and the a-posteriori audit.

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

Two names on the inventor list matter for how much weight to give this. Hyung-Jin Kim is a Google VP of Search — the witness whose sworn testimony put Google's use of click data on the record. Paul Haahr, another inventor here, is the ranking engineer behind Google's widely-cited 2016 "How Google Works" talk and a co-author of the Reference Contexts patent. This is not a speculative filing from a corner of the org; it's from the people who work on ranking.

---

What This Patent Does (Plain English)

The patent names the problem in its own opening: "some resources may be linked to by a number of resources that is disproportionate compared to the amount of traffic the resource receives from theses [sic] links." The point hiding in that sentence: when a page collects far more links than the traffic those links actually generate, that gap is itself the manipulation signal the system is built to catch.

Its answer is three scores per page — a seed score, a source score, and a resource score — computed in a chain:

1. Measure the destinations — For pages with enough data, compute a seed score: the clicks on the links pointing at that page, divided by the number of pages linking to it, adjusted for click quality.
2. Judge the linking pages — Score each source page by the seed scores of the destinations it links to. A page that links to healthy, clicked destinations is a good recommender; a page that links to behaviorally dead ones is not.
3. Classify sources — Sources below a threshold are labeled unqualified.
4. Propagate to everyone else — Pages that don't have their own click data yet inherit a resource score from the source scores of the pages linking to them.
5. Subtract the dead weight — In ranking, the effective count of linking sources is reduced by the number of unqualified ones.

Here's what that looks like in the patent itself. This is FIG. 2A — the three roles any page can play in the link graph:

---

Seed → Source → Resource: The Three Scores on Every Page

Before we touch a single ratio, get this distinction clean — everything downstream depends on it, and it's where the confusion usually starts. Ranking factors don't exist. There are systems. What a page has isn't a "link factor" you switch on. It's three different scores that pass value down a chain — I call it the Vouch Chain — and they don't all help the page that owns them: the score that ranks you is set by other pages, not by you. Your rank is borrowed.

FIG. 2A above showed the three roles a page can play. FIG. 2B shows how the scores are computed — the same seed, source, and target, now with the math flowing through them:

Walk it through in order:

• Your seed score is the proportionality and quality of the clicks your inbound links earn — the ratio we unpack next. It does nothing for your own page's ranking; it's the raw material other pages are judged by.
• Your source score is a function of the seed scores of the pages you link out to. It mostly helps the pages you link to — including your own internal pages.
• Your resource score is a function of the source scores of the pages linking to you. This is the one the patent routes into ranking (Claim 7: adjust ranking "based on the resource score"). This is the one that moves you.

---

What Sets Your Seed Score

Two inputs set a page's seed score: how often the links pointing at it get clicked, and how good those clicks are. Take the ratio first — the patent puts it in claim language (Claims 3, 12, 21): the seed score is based on "a ratio between the total number of selections of the links to the seed resource and the total number of resources that each include a link to the seed resource."

Read the formula carefully, because the "at the destination" part is what everyone gets wrong. It's measured from the links pointing at a page — any page. This is how a page earns (or fails to earn) a seed score, the first box in the diagram above. And keep the priority straight: the seed score is the secondary one. It doesn't rank the page that holds it — it's the raw material source and resource scores are built from. A "selection" counts only when a real user clicks through; a raw impression doesn't move it. Drop a perfect link into the footer of a busy page and let nobody click it, and it adds nothing to that destination's seed score — the host page's traffic is beside the point; only whether the link itself gets selected.

Then the clicks are graded: short, medium, long

That's the ratio. The second input is click quality: the patent doesn't just count clicks — it grades them. Claims 5 and 6 classify each selection by click duration: "a length of time the user making the selection dwelled on the resource after making the selection," sorted into short, medium, and long clicks. A short click (a bounce straight back) counts for little; a long click — someone who arrives and stays — counts for real. The description adds a second quality test: whether the visitor then navigates further on the destination site.

Click type	What it looks like	Quality contribution
Short click	Click, then bounce back quickly	Low — reads as accident, bot, or disappointment
Medium click	Some engagement, moderate dwell	Moderate
Long click + onward navigation	Arrives, stays, clicks deeper into the site	High — the pattern of a real, satisfied human

Here's the honest limit, though: the patent describes a heuristic that makes crude automation less useful. It does not lay out the full anti-fraud stack — the identity, browser, and IP defenses — so it would overstate the evidence to say bots categorically can't clear this bar. What we can say is that the design specifically devalues the shallow, bounce-back click that cheap manipulation produces.

And a reminder of what the seed score does not do: lift your own ranking. A high seed score makes a page worth linking to — it raises the source score of whoever links to it, and inside your own site it's what your internal links recycle into new pages. The seed score is the raw material. The next two scores are where it gets spent.

---

What Sets Your Source Score

Your source score is set by the company you keep — specifically, the seed scores of the pages you link out to. Link to seeds that carry high seed scores (destinations whose inbound links get genuinely clicked) and you read as a credible recommender; link to behaviorally dead ones and you don't.

Now, to become a strong source you have to link to strong seeds — and that sounds like a catch-22, because a page only becomes a seed through who links to it. But the good seeds already exist. I tell people to go find the government-level resource, the position-one authority for the ancillary topics your page leans on. Say you sell a niche insurance that only applies if you already hold a specific policy in certain states. Link to the government page that defines that rule. That's a seed sitting on a seed site — and linking out to it is what makes you a credible source. You don't get to be trusted; you earn it by who you're willing to vouch for.

The other half of the story is what happens when a source fails — the patent's teeth, in claim language (Claims 8, 17, 24). A source page whose score falls below a threshold is classified unqualified. Then, in ranking:

Read that carefully, because the precise unit matters. The thing removed is the unqualified source page — not the individual link, judged in isolation. A source fails on its pattern across the destinations it links to. One quiet link doesn't condemn a page; a page that systematically hands out links into behaviorally dead destinations condemns all the links it emits. Guilt by association, at the level of the linking page.

So a source's score doesn't just describe it — it decides which of its links survive to be counted at all. Which brings us to the only score that ranks you.

---

What Sets Your Resource Score — the One That Ranks You

Your resource score is built from the source scores of the pages linking to you — Claim 7 is explicit that ranking is adjusted "based on the resource score." After the subtraction has done its work, only the qualified sources remain in the count. So your resource score is, in plain terms, the quality of the pages still vouching for you once the unqualified ones have dropped out.

This is the score that moves your rank, and the uncomfortable part is that you don't set it. You can influence your seed score by earning clicked links, and you fully control your source score by choosing who you link to — but the score that actually ranks your page is handed to you by the pages that link to you. Your rank is borrowed. Read through this patent, the whole job of link building is to be vouched for by pages that are themselves credible sources — pages that link to good seeds and get their own links clicked.

---

Click-Validated Quality SEO Implications: What This Means for Your Link Building

Now that the three scores are straight, the most-repeated piece of link-building advice can be corrected without it misleading you.

"Build links on pages with traffic" — right instinct, wrong mechanism

You've heard it a hundred times, and it's almost right. The patent shows why the usual reasoning is wrong. The traffic of the host page isn't the signal. What matters is whether the links on that page get selected — clicked through to destinations people actually want — because that's what makes the page a qualified source, and a qualified source is what passes you a real resource score. A high-traffic page whose links nobody clicks is not a good source. A modest page whose links genuinely get used can be. So chase link selection and the quality of what the page links to, not the host's traffic number. And to kill the dangerous misread head-on: nothing here says the link to you has to rack up clicks for your page to rank. Your ranking rides on your resource score — the quality of the sources pointing at you — never on your own seed score.

1. Vet the whole outbound neighborhood, not just the DR

When someone brings me a stalled campaign — "I've been buying links from these places for a long time and nothing moves" — the container pages almost always share two traits: no internal links to the rest of their own site, and no external links to anything better than the client. Under this patent, that page can't make anyone a strong resource, because it isn't a credible source of anything. Before you value a placement by its domain metrics, ask what else is on the page and where it points.

2. The guest post that links only to you is the weak one

This is the most common way the industry wastes link budget. A guest post that links only to you derives its entire source score from you — a page that doesn't rank yet. So it has a poor source score, and you inherit a poor resource score from it. You still get the anchor relevance and some topical context, but you've left the biggest lever untouched.

So the rule I give: a guest post should link not only to you but to two or three highly authoritative, high-traffic pages relevant to the container's topic — and the site should carry internal links to its own best content. That's also a fast editorial-standards check. A publisher whose guidelines require internal links is a publisher who controls what it publishes, which lowers the odds the page gets neutralized as spam.

3. A link from a household-name brand isn't automatically a good link

The upgrade most people reach for is the "link insert" — paying to slot a line into an existing, well-established article, usually on a big SaaS or media brand. It feels like a win: you got a link from a DR90 name everyone recognizes. Whether it's a good link is a harder question, and this patent is one reason why.

These pages accumulate outbound links. A resource section or a "best tools" roundup on a strong brand routinely carries 200 or more. Under the source-score mechanic that works against you: a page that links to everyone is vouching for a lot of destinations, and most of them are low-traffic pages with weak seed scores. The more of those it points to, the more its own source score gets dragged down — and a weak source hands you a weak resource score, brand name or not. The keyword-rich anchor text these lists lean on doesn't help either (that's Reference Contexts territory).

But here's the relativity that makes link building genuinely hard: it's never one system in isolation. A link insert on a DR90 brand with millions of monthly visitors is probably sitting on a site that is itself a seed, or one hop from the seed list, carrying real entity trust — a separate mechanism this patent doesn't cover, closer to Seed Distance and brand signals. So even with 300 outbound links on the page, that link can still carry merit. The same insert on a DR30 nobody visits, twenty hops from any seed, with no entity behind it, is not the same link — even though a marketplace will price and label the two almost identically.

So the honest answer to "is this a good link?" is usually "it depends, and you have to look at all of it" — the container's own inbound profile, its entity standing, its distance from trusted seeds, and what else it links to. Perfect link building doesn't exist. What exists is being more right about these variables than your competitors are, and that is enough.

4. Recycle seed value through your own architecture

Your seed page gives you no direct ranking bonus — but linking to it does work for you. When an internal page links to a strong internal seed, that page's source score rises, which lifts the resource score of every other page it links to, including brand-new content. This is the real mechanism under the old "link new posts from your highest-traffic pages" advice.

5. Or manufacture the seed: the calculator play

Everything so far assumes the strong seeds live on other people's sites. You can also build one. The play I document in Linkable Assets: The Calculated Play is a working example: an interactive calculator that targets a high-search-volume keyword. Done right, two things happen at once. The page pulls steady traffic on its own, and it accumulates links from people referencing the tool — and links to a calculator are exactly the kind that get clicked, because the link exists so the reader can go use the thing. High click-through on its inbound links is the definition of a strong seed score.

Then the cascade starts, and every step of it runs on this patent's chain. The calculator's own traffic flows through its outbound links, which pushes up the seed scores of the pages it links to. Every internal page that links to the calculator is now linking to a strong seed, so its source score rises. And a page with a stronger source score hands a better resource score to everything else it links to — including the money pages that would never attract a link on their own. One genuinely useful page, and the scores improve in three directions around it. That's what a power page actually is, mechanically — not "a page with high DR," but a page whose seed score is strong enough to be worth recycling through your whole architecture.

6. Spam-prone niches play by different rules

The patent's third layer scores queries by their propensity to surface spam, using anchor n-grams that inherit source quality. For a query it reads as spam-prone, the description says the system may "trust" link counts less and click evidence more. If you work finance, health, gambling, or legal, assume link quantity is discounted harder and genuine engagement matters more. (This layer lives in the description, not the claims — hold it as design intent, not proven production behavior.)

7. It runs per page, and partly at query time

The patent says the scores can be computed "in terms of domains, web sites, URLs, or partial URLs." That kills "every link from a high-DR domain is equal." A strong domain can contain an unqualified page or section; the relevant unit is the page emitting the link, not the domain's average.

---

Google API Leak Cross-Reference: The Click Machinery

The 2024 Google API leak — first reported by Rand Fishkin and investigated by Mike King at iPullRank — contains fields that match this patent's machinery. Read the alignment column carefully: none of these is documented as this patent's output. They show Google has the constituent parts, not that this exact design is a named production system.

Patent Mechanism	API Attribute	Alignment
Short / medium / long click classification	goodClicks, badClicks, lastLongestClicks	🔶 STRONG MATCH — same click-quality taxonomy, but in the NavBoost/SERP-click surface, not proven for within-page link clicks
Qualified / unqualified source classification	sourceType (HIGH_QUALITY → LOW_QUALITY)	🔶 STRONG MATCH — a compatible source-quality tier; can encode several systems, not uniquely this one
The subtraction / per-link discount	pagerankWeight (per anchor)	🔸 WEAKLY COMPATIBLE — an available lever, cleaner match for Reasonable Surfer
Per-page source quality	nsr (per-page quality score)	🔸 WEAKLY COMPATIBLE — occupies a similar role; no documentation ties it to the seed ratio
Anchor n-gram spam scoring + temporal smoothing	AnchorPhraseSpamInfo	🔶 STRONG MATCH — anchor-phrase spam + spike detection; multiple anchor systems could produce it
Seed → source → resource propagation	—	📜 PATENT ONLY — no leaked counterpart

🔶 STRONG MATCH = the field clearly corresponds to the mechanism's class. 🔸 WEAKLY COMPATIBLE = plausibly related, but the field's role points elsewhere. 📜 PATENT ONLY = described in the patent, no leaked field found.

---

DOJ Trial Corroboration

The U.S. v. Google antitrust trial is the one venue where Google's use of click data was extracted under oath rather than volunteered in a blog post. It corroborates the fuel this patent runs on, with one boundary worth keeping honest:

• Classified clicks are real and used: the concepts behind goodClicks ("long click") and badClicks ("short click") appear in the plaintiffs' proposed findings, drawn from trial evidence — the same short/long taxonomy this patent claims.
• Click data is a major ranking input: testimony described NavBoost as one of the most impactful ranking systems, with a multi-month memory of click behavior.
• Why Google defends its click data at any cost: the trial's core showed Google paying billions to be the default search engine across the browsers and phones people already use — and running Chrome as its own front door to search — which is exactly how it preserves the behavioral signal this kind of system depends on, even as privacy layers erode it elsewhere.

---

Citation Network

Related Articles on This Site

• US7716225B1 (Reasonable Surfer) — The predictive counterpart. Reasonable Surfer estimates which links should be clicked from their features; this patent measures which were. Note Reasonable Surfer expired in March 2025 — cite it as the historical a-priori model, not current production.
• US8577893B1 (Reference Contexts) — Paul Haahr is on both. Reference Contexts judges the text around a link; this patent judges the behavior the link produces. Two independent anti-manipulation layers.
• US9953049B1 (Seed Distance) — Measures how far you are from trusted seeds through the link graph. This patent asks whether the links along that path are actually clicked. Position plus confirmation.
• NavBoost Deep Dive — The SERP-click sibling. Same click-quality taxonomy, different surface: NavBoost re-ranks results by result clicks; this patent evaluates links by link clicks.
• US7346839B2 (Historical Data) — Both handle temporal manipulation. Historical Data watches link velocity; this patent smooths sudden score spikes and can trigger human review.
• US9002832B1 (Low-Quality Site Classifier) — The consumer, and a near-sibling (it shares inventor Chung Tin Kwok). It buckets sources and applies a distribution demotion, feeding on exactly the per-page quality scores the Vouch Chain produces.
• SpamBrain (Link Spam) — The modern neural successor to the classify-and-subtract step. Where this patent removes a link from the count, SpamBrain more often neutralizes it silently.
• U.S. v. Google (DOJ Trial) — Where Google's use of classified click data was established under oath.

---

Click-Validated Quality: What Doesn't Matter as Much as SEOs Think

Strip the arithmetic away and the nature of this patent is a very human idea: a link is a claim about human interest, and the claim can be checked. That's all Google is doing here — emulating how trust works in the real world. A recommendation is only worth something if the people receiving it actually act on it. You can count the links all you like. If nobody walks through the door, the endorsement was never real in the first place.

The flavor — the explicit ratio, the seed/source/resource arithmetic, the classify-and-subtract step — that was the 2012 way of saying it. The math has very likely been folded into learned models since; I'd be surprised if the production code still divides two integers. But the architecture — judging a page by whether its endorsements produce real behavior — isn't going anywhere. It's too close to how authority actually moves through a network to ever throw out. The formula is the flavor; the trust is the nature.

This is why "ranking factors" is the wrong model, and why I keep pulling clients off it. A factor is a thing you toggle. This is a system — the Vouch Chain: conditions, inputs, outputs, and three scores that serve three different parties. External links were never a checkbox you tick for a point. They're how a page proves it's a credible recommender, which is how everything it touches earns a little trust. Miss that and you'll keep paying for links that got subtracted before ranking ever saw them — and you'll never know why, because nothing tells you.

So the question to walk out of here with isn't "how many links do I have?" It's the one the patent's own math is asking: when someone links to me, does anyone actually click — and does the page doing the linking send its readers somewhere worth going? Answer that honestly and the rest follows. Your seed score is a gift to everyone else in your graph. Your resource score is the only one that's really yours.

---

---

Frequently Asked Questions