What Google's Quality Raters Actually Do: The Rubric Behind the Rankings

There is a document Google has published, revised, and re-published for over a decade that spells out, in plain English, what the company considers a high quality page and what it considers junk. Most business owners have never heard of it. Most agencies quote three letters from it, E-A-T or now E-E-A-T, and skip the other 170-plus pages. This piece is a close reading of the Search Quality Evaluator Guidelines for people who make spending decisions based on search visibility: what the raters are, what they score, how the scale works, and why the rubric matters more in the AI-classifier era than it did when the full document first surfaced publicly in 2014.

Human raters do not rank pages. They grade the machine.

Start with the misconception that spoils most commentary on this topic. Google works with, through external vendors, approximately 16,000 Search Quality Raters across North America, Latin America, EMEA, and APAC, speaking more than 80 languages (per Google's overview of the rater program, November 2023). These people do not review your website when you publish it. They cannot demote you. Google is explicit that no single rating or single rater directly impacts how a given page or site ranks, and that ratings taken alone will not move results up or down the page.

What the ratings do is two things. First, they benchmark: when Google's engineers propose a ranking change, raters score result sets with and without the change, and the aggregated scores tell Google whether the change made results better. This evaluation pipeline runs constantly; Google reported more than 4,725 improvements to Search in 2022 alone. Second, and this is the part practitioners under-weight, the ratings feed the systems as positive and negative examples. A page rated Highest becomes a training signal for what quality looks like. A page rated Lowest becomes a training signal for what suppression should catch.

Hold that second function in mind for the rest of this article. The guidelines are not a policy document that humans enforce page by page. They are the specification for the target that Google's automated quality classifiers are tuned to hit. When the guidelines change, the classifiers follow. That is the entire reason a business owner should care what a rater on another continent is told to look for.

The two rating tasks: Page Quality and Needs Met

Raters perform two distinct jobs, and mixing them up muddles every downstream conversation about "quality."

Page Quality (PQ) evaluates how well a page achieves its purpose, independent of any particular search. A currency converter exists to convert currencies; a news homepage exists to inform readers about recent events; a product page exists to sell or describe a product. The rater first works out what the page is for, then judges how well it delivers. Google is careful to say that no page type is inherently better than another: there are Highest quality humour pages and Lowest quality encyclopedia-style pages. Purpose sets the standard; execution earns the grade.

Needs Met (NM) evaluates how useful a specific result is for a specific query from a specific location. The rater infers intent (a search for "coffee shops" from London means London coffee shops; "mercury" could mean the planet or the element), assumes the user wants current information, then rates the result on a scale from Fails to Meet, through Slightly, Moderately, and Highly Meets, up to Fully Meets. Fully Meets is reserved for queries with one clear intended result, matched exactly.

The distinction has a practical translation. Page Quality is about your page in isolation: effort, evidence, authorship, trust. Needs Met is about the match between your page and a real question. A brilliant page can still fail Needs Met for a query it does not answer, and a mediocre page can moderately meet a need it happens to address. Winning search visibility requires both, which is why keyword targeting with no quality substance behind it stopped working years ago. We map this two-sided requirement in our guide on how to build topical authority: coverage answers the Needs Met side, depth and evidence answer the Page Quality side.

How a Page Quality rating gets assigned

The PQ process runs in three steps, and the order matters.

Step one: determine the purpose. Everything else depends on this. Pages exist to inform, sell, entertain, share media, express opinion, run calculations. The rater's expectations scale to the purpose, which is why the guidelines hold a page about seasonal influenza to a very different standard than a page of memes.

Step two: assess harm and deception. Before any quality judgment happens, the rater checks whether the page has a harmful purpose or is designed to deceive people about its true purpose. If it does, the rating is Lowest immediately and the evaluation effectively ends. Deception is disqualifying, not a deduction. Note what does not trigger this step: content that is controversial, one-sided, or distasteful is explicitly not harmful by the guidelines' definition. Google draws the harm line at deception, danger, and spam, not at disagreement.

Step three: determine the rating. For pages that clear step two, the rater weighs three inputs. First, main content quality: as a general rule the guidelines treat content as high quality when it shows significant "effort, originality, and talent or skill" (Google, Search Quality Rater Guidelines overview, November 2023). Second, information about the site and creator: it should be clear who is responsible for the site and who produced the content, even under an alias. Third, reputation: what real users and recognized experts say about the site and its creator, researched off-site. That third input surprises people. Raters are told to leave your website and investigate what the rest of the web says about you. Your reputation is part of your rating, and you do not control the sources.

The five-point scale, from Lowest to Highest

PQ ratings land on a five-point scale, paraphrased here from the current guidelines:

Two features of this scale deserve attention. Medium is the default state of the honest but unremarkable internet; a competent page with no distinguishing evidence sits there, and so does most agency blog content. And the gap between High and Highest is mostly a gap in demonstrated effort and reputation, not in polish. A flawlessly designed page with generic content grades below a plain page carrying original data and a named expert with a verifiable track record.

How E-E-A-T is operationalized, not just abbreviated

E-E-A-T, Experience, Expertise, Authoritativeness, and Trust, is the lens raters apply during step three. The guidelines are specific about what each letter attaches to. Experience is the first-hand exposure of the creator: has this person actually used the product, visited the place, lived the situation. Expertise is the creator's knowledge or skill in the topic. Authoritativeness covers the creator, the main content itself, and the website: is this a known, go-to source for the topic. Trust is the extent to which the page is accurate, honest, safe, and reliable, and Google positions it as the anchor of the family; the other three exist to support a trust judgment.

The operational detail most summaries miss: E-E-A-T is not a score and has no fixed threshold. The required level scales with the topic. A forum post about a hobby needs everyday experience, not credentials. A page advising on chest pain needs recognized medical expertise, and a rater is instructed to check for it. The December 2022 revision that added the second E (Experience) formalized something practitioners had observed for years: for many queries, first-hand experience outranks formal credentials. A detailed review by someone who owns the product carries a kind of authority no committee-written buying guide can fake.

Making Experience, Expertise, and Authoritativeness legible to both raters and machines is its own discipline: consistent bylines, real author pages, evidence of first-hand work in the content, and structured data connecting the person to the organization. We keep our full treatment of the framework current in E-E-A-T explained for 2026, and the machine-readable half, marking up a real human author so the entity resolves cleanly, is demonstrated line by line in our Person schema worked example. This is Vector 6, Structure, in our methodology: the schema layer exists so that what a human rater would verify by reading, a classifier can verify by parsing.

YMYL: the real definition, and its 2025 expansion

"Your Money or Your Life" is one of the most misused terms in SEO, so here is the actual definition. YMYL topics are those that could significantly impact a person's health, financial stability, or safety, or the welfare and well-being of society. Medical information, financial products, legal questions, news about civic matters, safety information. For these, raters apply very high PQ standards, because a low quality page can hurt someone in ways a bad casserole recipe cannot.

The category is not static. The September 11, 2025 update to the guidelines explicitly extended YMYL to government, civics, and society topics, including election and voting information, and added rating examples for newer result types including AI Overviews (Google, Search Quality Evaluator Guidelines, September 2025; reported by Search Engine Roundtable, September 2025). The direction of travel is clear: topics where misinformation damages society are being pulled into the strictest tier alongside topics where it damages individuals.

The under-appreciated point for our clients: YMYL is a property of the topic, not the industry. A mattress store's page about mattress sizes is not YMYL. The same store's page about sleep and back pain edges toward health territory, and the standard rises with it. Agencies telling every local business "you're not YMYL, relax" are reading the label, not the definition. Advice that drives significant spending decisions, which is what most commercial content does, sits closer to the YMYL bar than the relaxed reading admits. We write our own content as if the strict standard applies, because the cost of doing so is effort, and effort is precisely what the rubric rewards.

What earns Lowest: deception and scaled content abuse

The Lowest rating is where the guidelines stopped being theoretical for the AI era. Two routes lead there.

The first is deceptive or harmful purpose: pages that misrepresent who they are, what they do, or who is behind them, and pages designed to harm. This has been in the document for years.

The second route is newer and directly relevant to anyone watching the content industry in 2026. The January 2025 update rebuilt the spam sections around three named abuse patterns, expired domain abuse, site reputation abuse, and scaled content abuse, and instructed raters to rate as Lowest any page where all or almost all of the main content is copied, paraphrased, embedded, auto-generated, or AI-generated with little to no effort, little to no originality, and little to no added value for visitors (reported by Search Engine Land, January 2025). Raters are now explicitly asked to consider whether main content appears to be AI-generated as part of the evaluation.

Read the criteria carefully, because they are better drafted than the panic coverage suggested. The trigger is not the tool. The guidelines state that "the use of Generative AI tools alone does not determine" the effort level or the rating (Google, Search Quality Evaluator Guidelines, January 2025). The trigger is the combination: scale, plus no effort, plus no originality, plus no added value. That combination describes an enormous share of what is being published right now, and Google has written its description into the training rubric for its quality classifiers. We documented where that pipeline of low-effort output is heading, and what it does to the sites that publish it, in our warning on AI slop and search. The rater guidelines are the enforcement specification for exactly that warning.

This is also the honest frame for our own positioning. The winning move against scaled slop is not avoiding AI tools; it is producing the effort, originality, and added value the rubric demands, at whatever speed your process allows. Every substantive page we ship carries dated citations, original observation, and a named accountable author, because those are the exact properties the Lowest criteria say are missing from suppressible content. Compliance and quality converged; the rubric made them the same job.

Why the rubric predicts the algorithm

Here is the mechanism that makes all of the above actionable rather than trivia. Google's quality classifiers, including the systems behind the helpful content signals folded into core updates, are machine-learned models. Machine-learned models are trained on labelled examples. The labels come, in meaningful part, from raters applying this rubric. So the guidelines function as a published specification of the classifier's training target: not a description of today's algorithm, but a description of what the algorithm is being continuously tuned to detect and reward.

The document's revision history supports the reading. Experience was added to E-A-T in December 2022; the helpful content system and subsequent core updates then visibly rewarded first-hand experience signals. The scaled content abuse language entered the guidelines in January 2025, ten months after Google's March 2024 spam policy update named the same abuse pattern; rubric and enforcement moved together. The September 2025 update added AI Overview rating examples just as AI Overviews became a dominant result surface. The pattern since 2014 has been consistent: the guidelines telegraph, with a lag measured in months, where the automated systems are headed.

For a business owner, the strategic consequence fits in one sentence. You cannot reverse-engineer the ranking algorithm, but Google hands you the grading rubric it trains that algorithm against, so building pages that would score High or Highest under the rubric is the most durable optimization strategy available. It is also the strategy that survives updates, because updates move the algorithm toward the rubric, not away from it.

Reading your own site the way a rater would

A working exercise we run in audits, condensed. Take your three most important pages and ask the rater's questions in the rater's order:

When we rebuilt Mattress Miracle's content library, every article was written to pass this exercise before it shipped: named authorship, sourced statistics, first-hand retail knowledge from a 40-year operator. The domain went from roughly 1,000 to 82,400 monthly organic visits (SEMrush, April 2026). One engagement, one industry; outcomes depend on competition and existing digital presence, and the guidelines themselves would insist on that qualifier. But the direction of the result is what the rubric predicts: pages built to grade High get treated like it.

The honest so-what for a business owner

Strip the jargon and the guidelines say something a Brantford shop owner already believes: be genuinely useful, show your work, be who you say you are, and do not fake it at scale. The remarkable part is not the advice. It is that the world's dominant search company published its internal definition of quality, keeps it current, trains its classifiers against it, and almost nobody in the market builds to it. That gap is an opening for any independent business willing to do the unfashionable work of effort and evidence. The document is no-cost; the full current version sits at guidelines.raterhub.com. Read the Lowest section against your own site before a classifier does it for you, and if you want a second set of eyes on that reading, our contact page is the fastest route to one. The rest of our primary-source teardowns live in the research hub.