On any website, any landing page, any sales brochure, the word appears without a second thought: quality of content, quality of service, quality of production. It is a reassuring word for the buyer, and one that the person uttering it almost never has to defend, because there is virtually no one on the other side with the tools to challenge it.
The result is the state of the web today: everyone produces ‘quality’, and the gap between the quantity of the worst and the quantity of the best is only widening. Especially on the web, where a practical definition of quality has been replaced, without us really noticing, by traffic metrics.
From usage to metrics
Historically, the quality of a product was based on two things: its utility and its durability. A quality chair is one that serves its purpose and lasts, because it is made from durable material; whether it is considered to be of quality by the user is another matter (that is the subject of a future article). A quality tool is one that performs the intended task for a long time. So, of course, one could say that maximising utility—and therefore traffic—is quality if we follow this definition. And from that perspective, it makes sense, but perhaps it isn’t quite that simple.
A high-quality product also commanded a higher price. Producing a higher quality product was more expensive, and so it sold for more. This logic had a filtering effect: those producing at a lower standard knew they could not, in good faith, charge the same price as competitors who had put more effort into their products (the lure of profit proved too strong).
On the web, this mechanism has broken down, because we have accepted that quality is measured using traffic. More specifically: the traffic captured by a competitor, which we must now catch up with (estimated by a third-party tool that invents metrics based on other, false metrics).
The human reasoning follows three steps:
(And if I can’t do it myself, I outsource it to someone who guarantees me ‘quality’ without going into the details.)
The detail, precisely, is that we have confused two very different things: the metric Google uses to rank a page once it has already been included in the race, and the quality of that page’s content.
The misunderstanding stems from Navboost
The US antitrust case against Google, the documents for which were made public in 2024, confirmed what the SEO community had suspected for years: Navboost is one of Google’s main ranking engines, powered by thirteen months of aggregated click data combined with metrics such as good, bad and ‘longest last’ clicks. The Vice President of Search, Pandu Nayak, described it under oath as “one of the important signals” of the engine, putting an end to years of public denials regarding the direct use of clicks in ranking.
“Navboost is one of the important signals that we have.”
Pandu Nayak, VP of Search at Google, sworn testimony, U.S. v. Google antitrust trial, 18 October 2023
Navboost is essentially a ‘learning system’ that analyses users’ past behaviour. If a page receives a lot of high-quality clicks (long clicks, with no ‘skip’ clicks – i.e. no immediate return to the search results) – and, above all, actual clicks – it will rise in the rankings. If it receives poor-quality clicks, pogo-sticking or immediate returns, it may drop down the rankings.
What is Navboost?
Navboost is a Google re-ranking system that has reportedly been in use since around 2005 and remained secret until it was revealed during the 2023 antitrust trial.
Its core principle: observe how users actually click on search results, then use those signals to reorder pages.
In practice, Google stores 13 months (and no more) of click data — queries, clicked pages, and click quality — and uses it to promote results that satisfy users while demoting those that disappoint them.
Marketing translation: your CTR and the quality of the experience on your page are not just reporting KPIs — they are signals directly used in Google’s ranking systems.
What stands out when skimming through this, particularly from a marketing perspective, is that Google measures quality by clicks, and that generating clicks equates to quality, so traffic is supposedly proof of quality… That’s where the logic breaks down.
The missing word: re-ranking
Navboost is a re-ranking system; this subtle distinction makes a difference, one that is all too easily overlooked.
Before a page is even eligible for Navboost, it must pass through a series of upstream filters: the part of the stack that Google didn’t need to defend in court because it was never contested.
Indexing, deduplication, anti-spam classifiers, Query-Based Salient Terms (QBST? Hey, just like on Yourtext Guru! What a coincidence!), Helpful Content System, E-E-A-T signals thanks to evaluations by Quality Raters (or by an LLM replacing them) etc… A page that doesn’t pass these filters won’t see Navboost arbitrate it: it simply won’t be in the SERP. No clicks, no signals, no chance of ranking higher.
And these filters have become more demanding, not less so. Helpful Content (2022), site reputation abuse classifiers (2024), dismantling entire sections of low-value affiliate ecosystems… Google has spent the last three years tightening the selection process upstream. The paradox to grasp if we want to understand Google today can be summed up in one sentence: Google has become more demanding on production quality, whilst rewarding traffic once the page is accepted. When Google tightens the screws, isn’t it actually editorial quality that truly ensures you stay in the race in the long term?
The media narrative following the leak has simplified this mechanism in two steps, aided by a measurement method and a business model dependent on traffic (advertising): we see “Google uses clicks” and interpret this as “so clicks are absolutely essential”. That’s not quite right: clicks do play a part in the selection process, within the pool of pages already deemed worthy of being in the running, but if you don’t have clicks, it’s true that you hardly appear at all.
It’s like being in a Marvel film…
A Marvel film is a technically superb production, with massive resources and a guaranteed audience; it’s entertaining, but then again, is it quality? Huge filters have forced the film to be produced to a strict standard, so it can be distributed globally through massive distribution networks, and the ‘hype’ is maintained with a clever marketing mix: post-credits scenes in other films, teasers, fan service, trailers, and clumsy leaks from likeable actors.
Through film communication and marketing, audiences are drawn into cinemas, which boosts ticket sales; the film enters the box-office charts, and the promotion feeds itself. Yet it is entirely possible to say a film is ‘the most-watched’ without necessarily claiming it is ‘the best film’.
On Google, we’ve come to accept this confusion. A page that ranks has indeed passed the filters: it’s indexable, clean, thematically relevant and hasn’t been penalised. But what pushes it above that threshold is Navboost – in other words, optimisation for clicks from a specific user demographic on a given query over a rolling 13-month period. Just as Marvel optimises for the 18–34 age group. What emerges are the most viewed pages, not necessarily the best ones.
Serious SEO has drawn a useful lesson from this: producing content that genuinely satisfies the click is profitable, and that is true. The trap is believing that just because it pays off, that’s what quality is. Navboost doesn’t measure the quality of content: it measures user satisfaction via their click, for a given query, over a thirteen-month window, across a population of Google users. That doesn’t mean the content highlighted is high-quality content, but it’s useful to know. Generating traffic isn’t necessarily linked to conversion, but fake traffic has a temporary positive effect, enough to justify certain practices. From a metrics perspective, traffic is therefore also somewhat manipulable.
Why LLMs are reshuffling the deck
And this is where the shift to LLMs really matters, beyond the simple drop in traffic everyone is talking about.
An LLM does not reward clicks; it does not have Navboost (even if it uses Google Search). When a model is trained or its responses are fed by search, what causes it to cite a source is specificity, verifiability and substantial authority (this isn’t synonymous with PageRank, I can see where you’re going with that), but not CTR (and besides, it doesn’t really encourage clicks, so…). A page that ranked very highly because it satisfied a click but doesn’t provide any original content misses out on the new filter. Conversely, a page that generated little traffic but says something that can’t be found anywhere else gains visibility in the LLM context.
What does this mean for SEO, which still works exclusively for Google? Given that Google has added an LLM* (*This isn’t official, but between BERT, MUM, etc., there’s a series of clues that point to this logical conclusion) to its filters to assess content quality, this is essential.
In other words: the layer that Navboost added on top of production quality is still important, but the quality filters have significantly narrowed down the pool of candidates for indexing.
The disconnection that no metric captures
Today, we can see this more clearly than at any time in the last fifteen years: traffic and revenue do not automatically grow at the same rate. Before LLMs, we often saw a steady rise in organic traffic, year after year, accompanied by an increase in revenue, but rarely in the same proportions.
This distinction is important: when it comes to qualified traffic, aligned with a genuine intention to purchase or make a decision, this increase can obviously have an impact on conversions. On the other hand, a broader increase in traffic—whether informational or poorly qualified—does not automatically guarantee an equivalent rise in revenue.
Since the rise of large language models (LLMs), overall website traffic has tended to decline, whilst, according to various customer feedback reports, turnover often remains relatively stable. This therefore calls for a more nuanced distinction to be made between traffic volume and its actual value.
Traffic is not, in itself, a measure of customer engagement. What matters is a visit’s ability to translate an intention strong enough to lead, directly or indirectly, to a purchase.
This is because traffic is in fact an aggregate of very different intentions: information seeking, comparison, purchase decision, curiosity, accidental clicks – many of which can now take place directly within LLMs. A page may well attract a lot of traffic without capturing any of the moments that matter in a buyer’s decision. Conversely, a page may generate little traffic yet play a decisive role in the argumentation leading to the sale. This is particularly true if it serves to enhance the brand’s value for LLMs.
It is also worth noting that this decisive page has never necessarily been hosted on your site, even though verifying an argument via the site remains a reassuring factor for the prospect.
No metric captures this accurately. The revenue attributed to a particular factor comes close, but remains constrained by the last-click model. Behavioural tracking is hampered by the GDPR, the fragmentation of user journeys, and the reality that many decisions are made offline, off-screen, and outside the scope of tracking. User surveys are lengthy, intrusive, and biased towards the response. What we’d really like to measure is the influence of content on a reader’s decision, but there’s no specific metric for that. Traffic is merely a distant, partial, approximate, and misleading estimate when taken in isolation.
The system effect
The problem is that everyone has ended up using the same metric. Google because it runs a search engine covering billions of pages and needs an aggregated measure. SEO agencies because that’s what Google rewards via Navboost, and it’s what the client can understand. Advertisers because that’s what’s being sold to them.
The result, after a decade, is an ecosystem that produces content tailored to the metric, but which overlooks the reader’s need for the best information. Articles that grab the click but disappoint once opened. Landing pages optimised for time spent that tell you nothing. ‘Ultimate guides’ that haven’t been written by someone who knows, but by someone who has looked at what others have written before them.
This content gets past Google’s filters because it is technically well-crafted: formatted, superficially sourced, and structured. It then benefits from Navboost because it is designed to attract clicks. But it doesn’t inform, it doesn’t convince, and it doesn’t leave a lasting impression. It’s exactly the web equivalent of a blockbuster that fills the cinema but is forgotten as soon as you walk out.
The problem isn’t that this content exists, but that we’ve called it ‘quality content’, and over time, we’ve come to believe it.
Moving beyond the proxy
Redefining the quality of web content isn’t done by replacing one metric with another—which is obviously imperfect—since we must inevitably take into account that any attempt to impose one will, within a few years, generate the content or techniques that circumvent it.
The question is therefore not ‘how to generate more traffic’. It is not even ‘how to generate the right traffic’; that is still thinking within the framework of the proxy. The question is: what standard do I accept for the content I publish, and am I able to defend it when no one is measuring it?
E-E-A-T, scientific peer review and web quality assurance models are attempts to answer this question, though they cannot do so on their own. But all are better than the metric we have allowed to become the default.
Can we offer something scalable, that meets ever-stricter standards, and recognises that quality isn’t just about getting your pages indexed by a search engine and generating clicks? I’ll explore this further in a future article.
Sources used:
Hobo Web (summary of documents from the DOJ v. Google trial, October 2025),
analysis of the May 2024 Content Warehouse leak by Mike King (among others)
Amélie Lachat, (PhD in Management Sciences – Marketing (Neuroscience), Paris-Dauphine PSL University), case study article on ‘content quality’ (in Hypertexte, October 2025).
Pandu Nayak’s testimony and the May 2024 API leak are the two primary sources (all those cited here are paraphrases).
Exact quote from Nayak: “So I would say that navboost is one of the important signals that we have.” (Nayak’s testimony, transcribed by the DOJ, October 2023).
