How Google May ‘Understand’ Unique Content

Jul 15, 2026 08:30 PM - 3 hours ago 157

Thanks to Rand’s fantabulous research and Barry’s expletive-laden ranting, we cognize that Google processes over 5 trillion searches each year. Trillion. Per day, that’s 13.7 billion. Per second, 158,000.

There are immoderate sizeable and increasing caveats here:

Only 32% of these extremity pinch a click.
Only 66.61% of the consequent 32% spell to the unfastened web.
This fig is rising each year.

That still intends Google processes 2.92 cardinal clicks to the unfastened web each day. It’s still a fig worthy fighting for – peculiarly for publishers whose business models heavy trust connected a click.

So let’s not wholly suffer show of what matters successful the present and now. And unsocial contented surely fits that mould.

I person reviewed a fewer erstwhile patents (Google’s in-depth article patent explained and how Google ranks news sites), and it is not a thoroughly enjoyable experience. A granted patent protects an idea; it doesn’t beryllium deployment aliases real-world usage cases – and it’s surely not dissimilar large tech to declare ownership of thing conscionable truthful it can’t beryllium utilized elsewhere.

Generally, if:

The patent is cited regularly and recently? This patent (Contextual estimation of nexus accusation gain) has been cited 24 times and arsenic precocious arsenic past year.
Whether it has world filings? Yes, but pinch immoderate caveats. US, China, ceased successful Europe and worldwide, but extended successful the US to 2039 very recently.
Whether Google has protected the ranking exertion astir the world? Yes, again pinch immoderate caveats.
Does it broadly align pinch your knowing of the conception (in this lawsuit non-commodity content)? Very overmuch so. As the rasping breaths of SEO-first, commodity contented make moreover robust lungs activity hard, it would beryllium inconceivable for Google to not measurement aliases measure characteristic successful immoderate manner.

It is much apt to beryllium utilized successful immoderate capacity.

TL;DR

Google has aggregate nationalist and leaked systems that look to measure originality, effort, and unsocial publication – spot OriginalContentScore and ContentEffort.
The patent describes an information summation people (potentially successful a 0 – 1 framing) that is assigned to a archive based connected really overmuch new accusation it adds beyond documents a personification has already seen connected the aforesaid topic.
In my – and galore others’ – opinion, Google’s systems reward originality successful immoderate way. Whether that’s straight done an accusation summation people and re-ranking system, a Bayesian predictive score, aliases indirectly done affirmative engagement signals, I couldn’t show you.
Originality doesn’t mean an entirely different document. As small arsenic a 10% quality could beryllium the delineator betwixt trading occurrence aliases failure.

How Does It Work In Practice?

This patent is not astir the accusation summation applied to the existent group of hunt results. It’s astir the consequent group of results – ranking the adjacent group of hunt results based connected wider personification hunt behavior, personalization, and added archive value.

It highlights that documents:

May beryllium reranked.
May beryllium excluded.
May beryllium significantly demoted.
May no longer appear successful results.

Based connected the magnitude of novel, applicable accusation provided erstwhile compared to different akin documents.

For immoderate tech SEO geeks retired there, you’ll beryllium good alert of the conception of preloading. In nerd circles, preloading tells browsers which resources should beryllium prioritized to amended the page load velocity and above-the-fold rendering.

I deliberation this patent useful successful a akin manner, but pinch bloody unreliable people instead of machines. Maybe bfcache is simply a much apt comparison, but I haven’t really sewage stuck into method SEO for a while, truthful forgive maine for my appalling analogies.

Step-By-Step

A personification sounds a archive astir a definite topic, let’s say, increasing an pome tree.
Google understands that the mostly of users don’t extremity astatine 1 page here. It’s a rich | topic. When should I works one? Where? What do I provender it?
With 13 months of click and engagement data to hand, Google knows – with, I imagine, an unerring level of accuracy – what portion of contented each personification should beryllium shown and erstwhile based connected extremity fulfillment.
But caller contented is written each day. Pages are updated. So this isn’t a fixed corpus to activity with. And possibly personification has a caller measurement of increasing pome trees?
So pages are compared. A personification sounds a archive (d1). Google past compares a caller aliases updated article (d2) to the original.
If d2 generates a favorable accusation summation score, it will apt beryllium shown to the personification arsenic portion of their journey. If it doesn’t, it’s doomed.

“An accusation summation people for a fixed archive is suggestive of further accusation that is included successful the fixed archive beyond accusation contained successful different documents that were already presented to the user.”

Let’s opportunity 2 documents are chosen based connected a user’s hunt and hunt history. They’re represented arsenic d1 aliases d2. D1 is an already-consumed document, and d2 is marque spanking new. Well, to the personification astatine least. These documents tin beryllium represented arsenic a vector (or immoderate different semantic representation) to thief the exemplary clone knowing of the archive and its position against akin documents.

A sketch showing really documents are scored against each different successful the vector space

The strategy provides a quantitative people to measure whether the personification should besides position d2 aft having viewed d1. If the instrumentality learning exemplary generates an accusation summation people of archive d2 over document d1, past d2 is apt to beryllium shown – for early usage cases, perchance astatine the disbursal of d1.

There are immoderate incredibly applicable implications here.

If a taxable has been done to death, you person a much constricted chance to rank and make worth without providing thing extra. In a script wherever your article scores 0, the strategy has assessed it provides thing extra, and a personification who has seen d1 is little apt to spot d2 – your article.

If thing else, make judge you guidelines retired supra your closest competitors successful immoderate manner.

A batch of this describes the foundations of creating superb content. Being different and opinionated out.

As pinch truthful galore of these Google-led ideas aliases initiatives location are flaws. You don’t person to travel it to the letter. But E-E-A-T and “information gain” are sound principles. You person to beryllium memorable. There is nary alternative.

How Important Is It?

I deliberation characteristic and opinionated retired are much important than ever. Strip the patent retired of the conversation. People aliases brands who people contented won’t past if they aren’t memorable to group and – by proxy – hunt engines.

So you’ve sewage to do thing differently.

In Google’s case, I deliberation it’s much astir ratio than thing else. If they cognize the accusation summation scores of 2 documents are virtually identical, past a personification isn’t going to beryllium shown some versions of the document. The 2nd archive will beryllium deprioritized successful favour of richer, much unsocial content.

Google has capable engagement information to spell on pinch these proxy scores to understand what archive should beryllium shown and when. They tin get a personification person to their extremity by removing overly akin pages from a user’s SERP aliases AI response.

Which whitethorn beryllium precisely why they’re thinning their index – the removal of non-value-add content. Well, that and each the AI slop you’re creating.

It is rather virtually down to a) computational resources (money) and b) getting the personification to the constituent of completion quicker. In the DOJ Antitrust trial, Pandu Nayak’s sworn grounds called Navboost “one of the important signals that we have.”

“…a shorter query convention aliases less speech turns tin supply a corresponding simplification successful the assets demands of the strategy e.g. pinch respect to representation and/or powerfulness usage of the system.”

And the Quality Rater Guidelines make galore references to effort, originality and talent. Frameworks for illustration E-E-A-T and the merchandise reviews update really item the value of really utilizing products and showcasing the effort you person gone to. The magnitude of “effort” you put successful is rather virtually quantified (highly urge Sean’s breakdown here). It is portion of the Helpful Content update (booooooo) and the much difficult your page is to replicate, the amended chance it has of success, each things being equal.

These are not stupid principles. They’re very bully ones. The problem is, effort is expensive. The less clicks contented produces, the little each article will generate.

In an attributable mode astatine least.

Google Is Building An Audience Loyalty Ecosystem

Don’t return my connection for it, return Barry’s. Google has wanted to get free of click-chasing churnalism for years. Now it can. And it is – successful astir cases, I think, a positive.

They are trying to build thing astir engaged users – like each patient retired there. Your most engaged users are your astir valuable. Google’s softly building a subscriber ecosystem that could 1 time rival their advertisement business. No logic to deliberation that

Publishers that tin show they person an assemblage extracurricular of SEO are being “rewarded.” Although I fishy you could switch rewarded pinch crushed a small much slowly.

You tin travel your favourite patient via Preferred Sources and arsenic a Search Profile via the Discover feed (U.S.-only astatine the clip of penning this), and badges for illustration “highly cited” person been successful play for immoderate time. It doesn’t activity very well, but they are trying to beforehand unsocial reporting.

You tin now spot how contented from societal and video platforms performs connected Google Search if you meet the requirements. Your integer footprint and effect wrong the manufacture you’re successful really matters. Particularly erstwhile you see really prevalent societal and creator accounts are successful Discover.

I interest that this is wholly intolerable to explicate what is happening to users. What is Preferred Sources vs. a Search Profile?

It’s reliable to unit group to travel you connected platforms – possibly that’s the point. Which I benignant of understand – but I deliberation 1 of these would’ve sufficed.

If you want to cognize a small much astir wherever Discover is heading, I made a short video astir it:

Does Information Density Matter?

Yes and no. Long articles are not needfully much effective astatine satisfying the user.

Google has methods to normalize the magnitude of an article to forestall further keywords and semantically applicable phrases from ranking the archive excessively highly. Factors like TF-IDF normalization prevent agelong documents pinch precocious connection counts from artificially inflating their relevance scores conscionable because they’re quote-unquote richer.

More item whitethorn beryllium the incorrect phrasing here. Detail and rigor are typically positives. But it’s little important than answering the mobility and getting the personification person to their extremity goal.

User restitution is quantified done extremity completions and Navboost data – it trumps everything else.

How Does It Affect AI Systems?

Well, accepted hunt ranking is still important successful AI systems – whether that’s really efficaciously you rank for the superior search, your inclusion successful the training data, RAG, aliases suite of fan-out searches tally concurrently. And AI searches are highly personalized – thing that’s apt to only summation complete time.

When Claude starts knowing what toilet insubstantial I bargain aliases selects a poorly chosen “Happy Mother’s Day” paper for my mum’s day that showcases my deficiency of effort and empathy, it’s clip to telephone it a day.

According to Kevin Indig’s latest fantabulous research, first-party investigation is uncommon successful AI citations, but it earns 3.3x more. And original information is the strongest azygous predictor of page originality. Good for accepted SEO, bully for AI search. Who knew?

The ideas described successful this patent representation almost excessively neatly onto really modern AI hunt systems retrieve applicable information. Of the SGE. It helps expect the user’s adjacent liking successful an assistant-like context. Personalized, “helpful” and pinch utmost memory.

As Roger Montti pointed out, this whitethorn springiness a clearer denotation of how AIOs usage pages that the personification successful mobility whitethorn beryllium willing in. Their full occupation is to synthesize answers from aggregate sources and searches to supply the cleanable jumping-off point. I fishy this scoring strategy is an fantabulous measurement to debar computationally expensive, unnecessary utilization of documents.

contentEffort – described arsenic a ‘Large Language Model (LLM)-based effort estimation for article pages’ – estimates the magnitude of effort invested successful creating an article. As slop makes up much than 50% of the internet, this is seemingly 1 of Google’s measurement of dealing pinch it.

How Can I Use This Effectively?

Make differentiated, non-commodity content. It’s really simple. Apply what we telephone accusation summation successful this discourse to your ain contented – if you cannot adhd thing of worth to the existing index, past don’t bother.

You tin usage this with:

Original data.
First-hand experience.
Interviews.
Real reporting.
Being first connected the segment and processing the communicative arsenic it happens.
Proprietary analysis.

You don’t request a large budget. You tin do astonishing things pinch a fewer free information sources, immoderate creativity, and a spot of rope. Just make judge the article has an constituent of uniqueness.

I deliberation this really helps framework whether contented is still worthy creating. If you’re doing thing conscionable for SEO reasons and you can’t adhd thing other to the existing suite of information, termination it. If a archive contributes very small caller information, the patent suggests it’s a beardown campaigner to beryllium deprioritized erstwhile selecting consequent documents.

Still costs clip and money to make, but is little and little apt to thrust immoderate existent value. Stay successful your lane, but thrust a nicer car.

I person a emotion your indexation study successful GSC is invaluable here. Beige contented has a support life truthful debased it’s successful the moving for the caller UK Prime Minister. So cheque for immoderate pages dropping retired of the index astatine standard for much superior issues.

More Resources: