What is Quality for Google? Notes from SCL Deep Dive Europe

The last day of Search Central Live Deep Dive Europe 2026 was all about serving content. We heard so much about quality content how Google evaluate?

In one of the sessions, Duy Nguyen, a Search Quality Analyst who helps lead Google's anti-spam teams, gave a session on quality.

The written version of his talk followed almost immediately. Google updated their documentation on creating helpful, reliable, people-first content to include a new "Main content" section that uses the same four attributes from the talk.

In this note, I go through what Google said, what the slides added, and what I think it means for anyone running a content-heavy site.

The four pillars of quality

Google's slide defined quality as the degree to which a human put effort, originality, talent, skill and accuracy into making something. If you know the Search Quality Rater Guidelines, this is a condensed version of section 3.2 on main content quality. Here is how I read each pillar.

One point that is easy to miss: quality is not only about text. Google's documentation defines main content as anything that helps the page achieve its purpose. That includes:

• Text and media: articles, images, audio and video.
• Interactive features: calculators, tools, games and search.
• User-generated contributions: reviews, forum replies and comments, when they serve the page's purpose.
• Tabbed or expanded sections: product specs, safety notes or reviews behind a tab.
• Titles and headings: the visible title and headings that summarise what the page is about.

The four pillars apply to every one of these. A custom-built interactive map can show as much effort as a long-form article, and a thin product video can be as commodity as a rewritten blog post.

Effort

Effort is about how much human work went into the content, or into the system that produces it. Building a genuinely useful tool counts, even if the pages it outputs are automated.

What does not count is the opposite: pages generated from feeds, or large volumes of AI text published with no one reviewing or curating it. Google's updated documentation also adds a line I think a lot of sites need to hear: crediting your sources does not replace doing original work yourself.

There is a technical footnote here too. The 2024 Content Warehouse API leak included an attribute called contentEffort, described as an LLM-based estimate of effort for article pages. Nobody outside Google can confirm how it is used, but it is hard to ignore.

Originality

Originality asks whether the page offers information or a perspective you cannot already find elsewhere. When similar content exists across the web, raters are asked to consider which page is the original source.

This lines up with Google's information gain patent, which scores how much new information a document adds compared with what the user has already seen.

Talent or skill

This is the most subjective pillar: does the content show enough skill to give visitors a satisfying experience? That can mean clear writing, a well-produced video or a tool that actually works.

The nuance I like in the new documentation is that not all content needs a credentialed expert. Google's own example is someone describing a new way they found to clear snow from their driveway. Personal experience can carry a page on its own.

Accuracy

Informational pages need to be factually correct. For YMYL topics (health, money, safety), the bar is higher: content must be highly accurate and in line with established expert consensus.

Can AI content rank?

Duy was clear about this in the room: quality is judged the same way whether a person or a model wrote the content. The rater guidelines already say that using generative AI does not, by itself, determine the effort or quality rating.

The real issue is scale. AI makes mediocre content almost free to produce, and mediocre content at scale is exactly what these four pillars penalise.

The documentation also now calls out deceptive authorship directly (anyone remember the brightonSEO talk that told you to fake EEAT?). Made-up author names, AI-generated headshots and false credentials are described as deception, and as a signal of a low-quality page.

Quality is not the same as relevance

The rater guidelines separate two ratings, and I see them mixed up all the time:

• Page Quality is how good the page is in itself. It does not depend on any query.
• Needs Met is how well that page satisfies one specific search.

A page can be excellent and still useless for a given query. That is fine. The important part is the first one: because Page Quality is query-independent, it travels with the page into every query it could rank for. A weak quality grade does not hurt one keyword. It hurts all of them at once.

Commodity content: the term to remember

Commodity content is content anyone could have written, because it is built entirely on common knowledge. It has come up more and more often in Google documentation, such as in their GEO guide.

The slide in Barcelona gave some more concrete example of what commodity and non-commodity content looks like:

The non-commodity versions all share one thing: they come from something the business actually saw, did or owns. No competitor can write them without the same experience.

Example from the event: RunRepeat

The example Google showed was RunRepeat's best running shoes guide (they didn’t name the example explicitly, but it’s ranking first on the query and screenshot matches). On the surface it targets a very commodity-looking query. "Best running shoes" is the kind of keyword that thousands of affiliate pages chase with the same rewritten list.

What sets it apart is the level of effort the page discloses:

• It pays for its own shoes. RunRepeat describes itself as an independent testing lab that buys every shoe with its own money, so the reviews are not shaped by free samples.
• It runs in them. Testers vary the runs, from roads to trails and from recovery runs to sprints, before writing the review.
• It cuts them apart. The lab slices shoes into pieces and measures more than 30 parameters, including shock absorption, energy return, traction, durability, breathability, stiffness, stack height, platform width and insole thickness.

What makes it non-commodity

Mapped against the four pillars, the page scores on each one:

• Effort: buying, running and dissecting every shoe is real, visible human work. The page also explains how the work is done, which is exactly the "How" Google's documentation asks creators to show.
• Originality: the lab measurements are first-party data. No competitor can publish the same numbers without running the same tests.
• Talent or skill: a consistent testing method across 30+ parameters takes expertise to design and to interpret.
• Accuracy: recommendations rest on measured results rather than manufacturer claims or other people's reviews.

Apply the brand-swap test and it fails immediately: put a competitor's name on this page and the claims are no longer true. That is the clearest sign of non-commodity content. The topic is common, but the evidence behind it is not.

The lesson for me is that you do not have to avoid competitive, commodity-looking keywords. You have to bring something to them that only you have.

Why core updates keep coming

One slide explained why Google keeps updating its systems at all. It gave three reasons:

1. Content formats. As new formats spread, people start searching for them, and Google may launch features to meet that demand.
2. Content breadth. The more people publish, the more saturated topics become, and the harder it is to surface the best content. Google responds with core updates.
3. Content issues. Spammers find loopholes that let low-effort content rank, so Google launches targeted algorithms to level the field.

The second reason is the one that matters for quality. Another slide claimed that for every result for "naranja" (Spanish for orange) in 2000, there are roughly 20,000 in 2025. I haven't found a public source for that figure, so treat it as what Google showed in the room.

Even as a rough number, the logic holds. If the pool of candidates for a query grows by orders of magnitude, the bar for the top 10 has to rise. Ahrefs estimated that 74.2% of new pages published in April 2025 contained AI-generated content.

Google's core updates documentation uses a restaurant list analogy for this: when you update a top-10 list a year later, some places drop without getting worse, simply because better ones opened. That is what a lot of "we didn't change anything" traffic drops look like to me.

Quality is attached to every URL, and inherited from the site

There is no single "quality algorithm". Google runs many systems, listed in its ranking systems guide, and the helpful content system was folded into the core in March 2024.

What we do know, from the DOJ antitrust case, is that Google has a page-level quality signal called Q*. In a Googler's deposition notes, it is described as largely static and not tied to a specific query. That matches the Page Quality idea above.

A retrieval talk at the same event made this concrete. The slides showed candidate URLs arriving with stored attributes such as language, country and a quality level, and results being reordered using those attributes. In other words, quality is already calculated before anyone searches.

The next question is always whether Google judges the page or the site. The answer is both:

• Gary Illyes said in Barcelona that core updates work at page level, which is why some pages on a site rise while others fall.
• The 2024 leak lists several site-level signals, such as siteAuthority, siteFocusScore and siteRadius.
• Google's own core update documentation talks about systems confirming that the site as a whole has improved.

My reading: Google ranks pages, but every page carries some of its site's reputation. That is how thousands of thin pages end up dragging down the few hundred good ones.

My take

Nothing here is a brand-new idea. Effort, originality and E-E-A-T have been in the rater guidelines for years.

I think that name is the most useful thing to come out of Barcelona. "Helpful content" was always too vague to act on. "Could a competitor have written this page?" is a question a content team can answer in a meeting.

It also connects classic search with AI search more tightly than most GEO advice does. AI Overviews and AI Mode retrieve from the same index. If ten pages say the same thing, any of them can be cited. If only you have the data, the experience or the test results, the citation has a reason to be yours.

If I were auditing a site against this today, I would do it by page type, not page by page:

1. Group URLs by template or page type, and check indexation for each group. A large share of "Crawled – currently not indexed" is the first warning sign.
2. Run each page type through the four pillars. Is there real human effort? Does it add anything the current top 10 does not?
3. Apply the brand-swap test to find commodity templates.
4. Improve, consolidate or remove what cannot be made non-commodity.
5. Set expectations in months, not weeks.

One caution on the leak and patent references above. They are useful for understanding direction, but attribute names in a leaked API document do not tell us how, or whether, a signal is weighted today. I would not build a strategy around contentEffort. I would build it around making pages nobody else could make.