The Search Quality Rater Guidelines are the instruction manual Google gives to the outside contractors who judge its search results. A rater is shown a query and the pages returned for it, then scores how well each result serves the person searching and how trustworthy the page is. The manual exists because Google needs a consistent human yardstick. Without one, engineers would have no stable way to tell whether a change to ranking systems made results better or merely different. The current edition, dated September 11, 2025, runs to 182 pages and is public.

Individual ratings do not move pages. Section 0.1 states that no single rating can directly change how a page, site or result appears in Google Search. In aggregate, though, they grade Google's own systems and, as court records have since shown, supply some of the data those systems learn from.

How a rating task works

Google put the workforce at about 16,000 people in an overview it published in November 2023: roughly 7,000 in North America, 4,000 each in Europe, the Middle East and Africa and in Asia-Pacific, and 1,000 in Latin America. All are employed by external vendors. Before rating, they are expected to read the guidelines, "understand it, internalize it, and take a test on it," Pandu Nayak, then Google's vice president of search, told a federal court on October 18, 2023.

A task pairs a query with a locale, since raters are told to represent people in their assigned language and location. The rating then has two parts.

Page Quality asks whether a page achieves its purpose. The rater first checks whether that purpose is deceptive or harmful, which earns the bottom grade outright. Otherwise the rater weighs the main content, the reputation of the site and its creators, and E-E-A-T: experience, expertise, authoritativeness and trust. The scale runs Lowest, Low, Medium, High and Highest, with half-steps between each.

Needs Met asks how well a result serves the searcher, from Fails to Meet through Slightly, Moderately and Highly Meets to Fully Meets. Flags mark pornography, foreign-language results and pages that do not load. Intent shapes the score: Part 2 of the document sorts queries into know and know-simple, do, website, visit-in-person and multiple-intent types, so an excellent page can still fail by answering the wrong question.

Subjects carry different burdens. Pages on Your Money or Your Life (YMYL) topics, which the guidelines divide into health or safety, financial security, government, civics and society, and a residual category, receive "the most scrutiny" for page quality.

Judgements are then pooled. Nayak testified that ratings on query-result pairs are rolled up to the query level and then across a sample query set, producing a top-line measure called information satisfaction (IS) on a scale of 0 to 100. To convey the unit, he estimated that losing Wikipedia entirely would cost roughly half a point.

One standard format is the side-by-side experiment, in which raters compare results with and without a proposed change. Google ran 124,942 of them in 2023, alongside 719,326 search quality tests and 16,871 live traffic experiments, which together yielded 4,781 launches, according to its How Search Works site.

Origin and evolution

Copies circulated before Google acknowledged them, leaking in 2008, 2011 and 2012, according to Search Engine Land. In 2013 Google published a 43-page abridged edition, which it called a "Cliff's Notes" version of the 161-page text raters used. The full document followed on November 19, 2015: 160 pages, rewritten for mobile search. "The guidelines reflect what Google thinks search users want," Mimi Underwood of Google said at the time.

The E-A-T framework entered the guidelines in 2013 or 2014, with accounts differing on the revision. A July 2018 rewrite built much of the quality language around it, weeks before an August 1, 2018 core update moved health and medical sites sharply.

Later editions narrowed the focus. The July 28, 2022 version, at 167 pages, redefined YMYL around topics needing a high level of accuracy to prevent significant harm. On December 15, 2022, Google added a second E, for experience, creating E-E-A-T with trust at its centre. The November 16, 2023 revision simplified Needs Met definitions and added guidance on forums and short-form video. On March 5, 2024, significant factual inaccuracies became a marker of untrustworthy pages, taking the length to 170 pages.

Generative artificial intelligence (AI) drove the next expansion. The January 23, 2025 edition grew from 170 to 181 pages, defining expired domain abuse, site reputation abuse and scaled content abuse, and giving the document its first definition of generative AI. Main content produced with automated tools and little added value can now be rated Lowest, a change John Mueller of Google discussed at Search Central Live in Madrid on April 9, 2025. Raters were also told to disable ad blockers. The September 11, 2025 edition added examples for rating AI Overviews and widened the civic YMYL category to cover election and voting information.

Why marketers read it

Few public texts set out in such detail what Google's ranking engineers are aiming at, so search engine optimisation (SEO) teams, publishers and content marketers parse every revision. Marie Haynes, an SEO consultant, described the two feedback loops behind Google's systems: "Two things fine-tune the systems. One is the quality rater rankings, and then the other is the actions of users."

YMYL status carries commercial weight. Health publishers, whose subjects sit in the most scrutinised category, have absorbed some of the steepest declines, with WebMD's visibility down about 43% from its peak as of January 2026. Format follows the text too. Google's Discussions and forums feature appeared in 77% of search results by December 2024, a prominence tied to the guidelines' stated value on forum discussions from people with experience. Effort is another lens: the word appears 120 times in the document, a count cited in July 2026 coverage of Marie Haynes's analysis of pages Google crawls but declines to index.

For media buyers the connection is indirect: where publisher traffic comes from search, standards that reshape organic rankings also reshape the supply of impressions.

Limits, criticisms and disputes

Influence is the oldest argument. Google has said since 2015 that ratings do not determine individual rankings. Its November 2023 overview, however, also says ratings improve its systems "by giving them positive and negative examples". The antitrust record made that role concrete. In his September 2, 2025 remedies opinion, Judge Amit Mehta found that RankEmbed and its successor RankEmbedBERT rely on two main sources of data: a share of 70 days of search logs, plus scores generated by human raters. The court ordered disclosure of user-side data behind those models, although plaintiffs conceded the scoring data could be withheld. A presentation from the rebuttal testimony of Douglas Oard, a University of Maryland professor who testified for the government, listed RankBrain, DeepRank, term weighting and QBST (query-based salient terms) among components trained or fine-tuned on rating data. "Not directly" is a narrow claim.

Rater judgement has acknowledged limits. Internal Google documents cited in the same presentation warned that raters may not understand technical queries, cannot accurately judge popularity and do not always weigh freshness. Nayak argued the alternative was worse: following live click data alone would promote clickbait, and page quality, he told the court, is "a little anticorrelated with clicks."

Labour is a third dispute. Raters supplied by Appen received a pay rise on January 1, 2023, to between $14 and $14.50 an hour from as little as $10, covering 3,000 to 5,000 workers, according to the Alphabet Workers Union. A year later Google ended the Appen contract, effective March 19, 2024, a relationship worth $82.8 million of the vendor's 2023 revenue. The programme still rests on contractors, leaving Google's quality benchmark dependent on outsourced, revocable arrangements.

Misreading completes the list. Parts of the industry treat the document as a ranking checklist. The guidelines are "not a guide for search ranking," Mueller wrote on Bluesky in May 2026.

Not the same as

E-E-A-T is one framework inside the document, applied by raters within the Page Quality scale. The guidelines are the whole rulebook.

Manual actions are penalties applied to specific sites after review by Google staff. External raters cannot penalise a site, and a rater's visit has no effect on it.

Spam policies are the published rules for site owners, enforced by automated systems such as SpamBrain and by manual review. The guidelines borrow their spam definitions but address evaluators, not publishers.

AI model raters score chatbot answers to train large language models under separate instructions. In September 2026, 404 Media reported that contractors among OpenAI's roughly 10,000 quality raters and AI trainers had been dismissed for using AI to do the rating.

Recent developments

As of October 2026, the edition Google hosts remains the one dated September 11, 2025. Claims of a June 2026 revision circulating on marketing blogs are not reflected in the published document.

Rater standards and public policy increasingly move in step. On May 15, 2026, Google stated that its spam policies, scaled content abuse included, apply to content surfaced in AI Overviews and AI Mode, categories the rater guidelines had absorbed in January 2025. On October 1, 2026, Google revised its guide to using generative AI content, citing the rater guidelines in its changelog and adding that all AI-generated content should be manually fact-checked for accuracy and trustworthiness before publication. The changelog presented the edit as bringing documentation into line with developer-event presentations, though the paragraph pointing to rater sections 4.6.5 and 4.6.6 was already in place beforehand.

Timeline

  • 2008: A version of Google's rater guidelines leaks publicly, one of several leaks reported in 2008, 2011 and 2012
  • 2011: A 125-page edition leaks
  • 2013: Google publishes a 43-page abridged edition while raters work from a 161-page document
  • 2013 to 2014: E-A-T enters the guidelines, with accounts differing on the exact revision
  • November 19, 2015: Google publishes the full 160-page guidelines, revised for mobile search
  • July 2018: A revision rebuilds much of the quality language around E-A-T
  • August 1, 2018: A core update produces heavy movement among health and medical sites
  • July 28, 2022: A 167-page edition narrows YMYL to topics requiring high accuracy to prevent significant harm
  • December 15, 2022: Experience is added, creating E-E-A-T
  • January 1, 2023: Pay for Appen-supplied raters rises to $14 to $14.50 an hour
  • October 18, 2023: Pandu Nayak testifies to roughly 16,000 raters and the 0 to 100 IS metric
  • November 2023: Google's overview document puts the workforce at about 16,000 across four regions
  • November 16, 2023: Needs Met definitions are simplified and guidance on forums and short-form video is added
  • March 5, 2024: Factual inaccuracies become a marker of untrustworthy pages; length reaches 170 pages
  • March 19, 2024: Google's contract with Appen ends
  • January 23, 2025: The 181-page edition adds spam abuse definitions and a definition of generative AI
  • April 9, 2025: John Mueller discusses the AI content changes at Search Central Live in Madrid
  • September 2, 2025: Judge Amit Mehta's remedies opinion describes rater scores as training data for RankEmbed
  • September 11, 2025: The 182-page edition adds AI Overview examples and expands the civic YMYL category
  • May 15, 2026: Google applies its spam policies explicitly to AI Overviews and AI Mode
  • May 2026: John Mueller states the guidelines are not a guide for search ranking
  • October 1, 2026: Google updates its generative AI content guidance with material drawn from the rater guidelines

Summary

Who: Google writes the guidelines; roughly 16,000 contractors employed by outside vendors apply them; Google's search engineers use the resulting ratings; SEO teams, publishers and content marketers study the document for signs of what the systems reward.

What: A public manual, 182 pages in its current edition, instructing raters to score search results on two scales, Page Quality and Needs Met, with extra scrutiny for YMYL topics and E-E-A-T as a core quality lens.

When: Leaked by 2008, published in abridged form in 2013 and in full on November 19, 2015, revised repeatedly since, most recently on September 11, 2025, with generative AI reshaping the January 2025 edition.

Where: In Google's rating programme across North America, Europe, the Middle East and Africa, Asia-Pacific and Latin America, with ratings feeding side-by-side experiments, the IS quality metric and, according to court findings, ranking model training data.

Why: Google needs a consistent human benchmark to judge whether changes improve search, because live click data alone can reward clickbait. The ratings do not adjust individual sites, but they shape what Google's systems are built to prefer.