What Google’s Quality Raters Actually Do With AI Content
Google revised its Search Quality Rater Guidelines twice in 2025 — and most SEO advice online blends the two into one event. The January revision is the one that actually gave 16,000+ human raters a worked example of “no effort” and told them to rate it Lowest. The September revision, by Google’s own description, changed almost nothing. Here’s each one, correctly dated, a debunk of the update rumors circulating in 2026, and a self-audit tool at the end.
The guidelines don’t rank your page. They train the thing that does.
Start with the correction most SEO advice skips: the Search Quality Rater Guidelines (SQRG) are not an algorithm, and the more than 16,000 external raters who use them don’t touch your rankings directly.1 They read live search results, apply the guidelines, and submit quality and needs-met scores. Those scores become training and evaluation data for Google’s ranking systems — closer to a labeled dataset than a lever you can pull.
That distinction matters because it changes what “optimizing for the guidelines” should mean. You’re not trying to satisfy an algorithm’s parsing logic. You’re trying to satisfy the judgment of a trained human reader who has been given an explicit rubric — and who, since January 2025, has also been given explicit instructions on how to spot AI-generated pages built for volume instead of value.2
The document itself has grown in two distinct steps, not one. The January 23, 2025 revision expanded it from 170 to 181 pages and rewrote the spam and AI-content sections substantially.3 The September 11, 2025 revision added a single further page — 182 in total — and Google described it plainly, through a spokesperson: “a minor update… with small clarifications and a handful of new examples,” making no change to rating guidance itself.4 Treating both as one event, as a lot of recycled SEO content does, gets the timeline and the stakes wrong. The breakdown below keeps them separate — and addresses the newer confusion around whether a third update landed in 2026.
Two confirmed updates — and what each one actually changed
170 → 181 pages
Formally defined generative AI, restructured the spam section, added the AI-content worked examples and the exaggerated-claims rule.
181 → 182 pages
Expanded the YMYL definition and added AI Overview rating examples. Google: no change to rating guidance.
No new revision found
No Google announcement, change log entry, or updated PDF has surfaced since September 2025. See below.
January 23, 2025: the substantial rewrite
This is the update that actually matters for how AI-assisted content gets judged. Google added a formal definition of generative AI to the guidelines for the first time and rebuilt the “Spammy Webpages” section (4.6) around it, introducing named subsections including Expired Domain Abuse, Site Reputation Abuse, and Scaled Content Abuse (section 4.6.5).5 That subsection is explicit that using automated tools — generative AI or otherwise — to produce many pages that add little value counts as spam, regardless of the tool.6 The following subsection, 4.6.6, states plainly that a Lowest rating applies when main content is created with little to no effort, originality, or added value, whether it was copied, paraphrased, or AI-generated.7
The same revision added a worked failure example that later coverage sometimes misattributes to the September update: a page discussing a medical topic that opens with a generic AI disclaimer about its knowledge cutoff and trails off mid-sentence near the end.8 The lesson isn’t “don’t use AI.” It’s that raters were given, for the first time, a concrete tell to look for — leaked scaffolding and content that stops rather than concludes — rather than an abstract instruction to “watch for AI.”
January 2025 also added a rule for the gray zone between honest and dishonest self-presentation. Outright false claims about who wrote a page were already enough for a Lowest rating. The new rule covers claims of experience or expertise that are exaggerated or mildly misleading without being outright false — a Low rating applies once a creator’s claimed qualifications read as marketing rather than substance.9
September 11, 2025: the clarification, not the overhaul
Seven months later, Google published a second revision — and was unusually direct about its scope: a minor update with small clarifications and a handful of new examples, changing nothing about rating guidance.4 Two things in it are worth knowing. First, the YMYL (“Your Money or Your Life”) category that previously covered “Society” was renamed and expanded to “Government, Civics & Society,” explicitly reaching content about elections, public institutions, and civic trust.10 Second, Google added rating examples for AI Overviews, mirroring the examples already provided for featured snippets and knowledge panels.11 Beyond that, it was typo fixes and wording tightened for clarity.
The practical upshot: if your niche touches policy, elections, regulation, or anything civic-adjacent, the YMYL bar is higher as of September 2025. If your question is “what changed about how AI content gets judged,” the honest answer is: nothing did in September — that work was already done in January.
Is there a 2026 Quality Rater Guidelines update? Here’s what we actually found.
As of this check on August 26, 2026, no. Google’s own public change log for the guidelines still lists September 11, 2025 as the most recent revision, and no post on the Search Central blog, no spokesperson statement, and no updated PDF has surfaced since.4 That’s worth stating explicitly, because a cluster of SEO and AEO-focused blog posts published in mid-2026 describe a “June 2026 update” in specific detail — including new section numbers (5.2, 5.4, 6.1), a supposed “Synthetic Authority” flag, and rewritten rules for unattributed AI summaries.
None of this means the underlying advice in those posts is bad — writing with real, attributed human experience and avoiding unattributed AI summarization is exactly what the confirmed January 2025 update already rewards. It means you shouldn’t cite a specific 2026 section number to a client or in your own content until Google’s actual document reflects it. If a real update does land, the two markers that will confirm it are a dated post on the Google Search Central blog and a new “What changed” entry inside the PDF itself — we’ll update this page when either appears.
The five-tier rating scale, in plain language
Every Page Quality rating raters assign lands on this scale. It’s the single most useful mental model for editing your own drafts — read your page and ask which row it would land in, honestly.
| Tier | What it takes to land here | Where AI content typically fails or lands |
|---|---|---|
| Highest | Exceptional, original main content; deep, demonstrable expertise; very high trust; the page is often the definitive resource on its topic. | Achievable with AI as a drafting tool plus real reporting, testing, or lived experience layered on top — not with AI alone. |
| High | Satisfying main content, clear expertise or experience, good trust signals, minor gaps only. | Reachable when AI drafts are heavily edited, fact-checked, and given a specific point of view. |
| Medium | Adequate, safe, meets the basic need but not distinctive; common for competent-but-generic content. | Where most lightly-edited AI content actually lands — correct, unobjectionable, forgettable. |
| Low | Unsatisfying MC, unmet purpose, or E-E-A-T claims that read as exaggerated relative to what’s demonstrated. | Thin AI rewrites of existing top-ranking pages with no new information or verification. |
| Lowest | Little to no effort, originality, or added value; deceptive authorship claims; scaled, templated, or unedited generative output. | Bulk-published AI content with no human review — the explicit target of the January 2025 update. |
Paraphrased and condensed from the published rating-tier structure of the Search Quality Rater Guidelines; not a verbatim reproduction.12
E-E-A-T, reread specifically for AI-assisted content
E-E-A-T isn’t a score and it isn’t a ranking factor — Google has said this repeatedly, and it’s worth taking at face value.13 It’s the mental checklist raters are trained to apply. For AI-assisted content specifically, each letter has one failure mode worth naming.
Experience — proof you did the thing
This is the pillar AI can’t fake, and the one raters are trained hardest to look for: original photos instead of stock images, your own measurements instead of paraphrased ones, dated results instead of evergreen claims.14 If your process is “research the topic, then write about it,” you’re demonstrating expertise at best — never experience. One real, specific, first-hand detail per section is the highest-leverage edit you can make to an AI draft.
Expertise — command, not vocabulary
AI models are fluent in the vocabulary of almost every field, which is exactly why fluency alone no longer reads as expertise. Raters are trained to weigh whether content anticipates the follow-up question or corrects a common misconception — the things a fluent-but-shallow summary skips.
Authoritativeness — granted externally, not claimed
No amount of on-page copy substitutes for this one. Raters do outside reputation research — what other sites, forums, and reference sources say about the author and the domain.15 Link-building, citations, and being quoted elsewhere serve this pillar; it’s structurally impossible to write your way to authority inside the article itself.
Trust — the foundation the other three sit on
Google is explicit that trust is the most important member of the group: a page can show real expertise and still rate low if it isn’t trustworthy.13 For AI-assisted pages, trust erodes fastest through small things — unlinked claims, stats without a source, a byline with no verifiable identity, or the January-2025-specific one: experience claims that read as manufactured.
The practical checklist
None of this requires abandoning AI as a drafting tool — Google’s own language treats generative AI as a legitimate tool, not a violation.16 It requires making sure a human’s judgment is visible in the finished page. In rough order of how much they move a page across the rating scale:
- 01Add one thing the model couldn’t have generated. A tested result, an original screenshot, a specific number from your own work, a disagreement with the consensus view.
- 02Kill unverifiable superlatives. “Industry-leading,” “cutting-edge,” “the ultimate guide” read as filler to both readers and raters — cut them or replace with a specific claim you can back up.
- 03Source every number. A stat with no attribution is a liability under the current rules, not a strength — link to where it came from or remove it.
- 04Write (or link) a real author bio. Name, a verifiable credential or track record, a way to check it. Generic “our team of experts” bylines are exactly what raters are trained to discount.
- 05Read your own ending. The 2025 worked failure example is literally a page that trails off mid-sentence — a five-second final read catches what a human editor would never ship.
- 06Date it, and mean it. A “Last updated” stamp only helps if something in the piece actually changed — cosmetic date bumps are a trust signal that backfires under scrutiny.
- 07Say where you’re not sure. Explicit uncertainty (“this figure is Google’s own estimate, not independently verified”) reads as more trustworthy than false confidence — to readers and, per the guidelines’ own logic, to raters.
Self-audit: which tier would your page land in?
This is a directional self-check built from the factors above — not a Google tool, not a scoring formula from the actual guidelines, and not a prediction of your real rating. Check what’s genuinely true of the page you’re editing; the result is a rough band, not a number to chase.
Quality Rater Self-Audit
7 signals, each drawn from the checklist above. There’s no official weighting behind these — check honestly and read the band, not the count.
This tool counts signals, it doesn’t grade them — one missing byline can matter more than three present superlative-fixes. Use it to spot gaps, not to certify a page.
Myth vs. fact
“AI-written content gets penalized by Google.”
The guidelines explicitly say tool use alone doesn’t determine the rating — effort, originality, and value do, for AI and human writers alike.16
“E-E-A-T is a ranking factor you can optimize like a meta tag.”
It’s a rater framework that correlates with what the algorithm rewards — there’s no E-E-A-T score to target directly.13
“The September 2025 update was the big AI-content shake-up.”
The AI-content overhaul happened in January 2025. Google called September’s revision minor and said it made no change to rating guidance.4
“A June 2026 update added new sections 5.2, 5.4 and 6.1.”
Unconfirmed. No Google source corroborates this; the real sections on this topic are 4.6.5 and 4.6.6, from January 2025.15
“Quality raters directly control where my page ranks.”
Raters produce training and evaluation signal for the ranking systems; an individual score isn’t a live ranking input.1
FAQ
Is the Search Quality Rater Guidelines PDF public?
Yes. Google publishes it openly; it’s typically found by searching “Google Search Quality Evaluator Guidelines PDF,” and it’s updated roughly once or twice a year.17
How many quality raters does Google actually use?
Widely-cited figures put the number of external raters at around 16,000, though Google has not published an exact, current count.1
What is “scaled content abuse” exactly?
It’s section 4.6.5 of the guidelines, added in January 2025: content, including AI-generated content, produced at volume with little original value, primarily to manipulate search rankings rather than help users.6
Does paraphrasing an existing top-ranking article count as original content?
Not under the current guidance. Content that mainly repackages existing information without new insight doesn’t clear the originality bar the guidelines describe, regardless of how it was produced.7
Which update actually added the AI-content rules — January or September 2025?
January 2025. That revision defined generative AI for the first time, restructured the spam section, and added the worked failure examples. September 2025 only expanded the YMYL definition and added AI Overview examples — Google described it as a minor update that changed no rating guidance.4
Was there a new Quality Rater Guidelines update in 2026?
Not as of this check on August 26, 2026. Google’s public change log still shows September 11, 2025 as the latest revision. Posts describing a “June 2026” update with new section numbers haven’t been corroborated by any Google source and don’t match the document’s actual numbering.4,15
Does this apply to every page, or mainly YMYL content?
The scrutiny is sharpest for YMYL — Your Money or Your Life — pages, and the September 2025 revision specifically widened the Government, Civics & Society category. But the core “little to no effort or value” test for the Lowest tier applies across all content types.10
Can a heavily AI-assisted page still reach a High or Highest rating?
Yes — the guidelines rate the finished page, not the tool used to produce the first draft. What separates a Highest-rated page from a Lowest-rated one is the human judgment, verification, and original value layered on top, not authorship method.16
What’s the fastest way to check if my content reads as “exaggerated” E-E-A-T?
Ask whether every claim in your bio or intro is independently checkable — a linked credential, a named publication, a verifiable result. If a claim only exists on your own page, a rater is instructed to weigh it against outside reputation research rather than take it at face value.9
Do quality rater scores show up anywhere I can check, like Search Console?
No. Ratings feed into Google’s internal evaluation and training pipeline; they’re not surfaced per-page to site owners anywhere.
Glossary
- SQRG
- Search Quality Rater Guidelines — the public document raters use to score search results.
- MC (Main Content)
- The part of a page that fulfills its purpose; what raters weigh most heavily for quality and E-E-A-T.
- YMYL
- “Your Money or Your Life” — topics that can affect health, financial stability, safety, or (as of September 2025) civic trust.
- Scaled Content Abuse
- Section 4.6.5, added January 2025 — volume-produced, low-value content, AI-generated or otherwise, aimed at manipulating rankings.
- Needs Met
- A separate rating dimension from Page Quality — how well a result satisfies the specific intent behind a query, independent of how well-made the page is in isolation.
Editorial verification note
This piece was checked against reporting on both the January 23, 2025 and September 11, 2025 revisions of Google’s Search Quality Rater Guidelines, cross-referenced across Search Engine Land, Search Engine Journal, PPC Land, Search Engine Roundtable, Originality.ai, and Google’s own developer documentation. Page-count figures (170→181, then 181→182) and section numbers (4.6.5, 4.6.6) are attributed to their original reporting or to Google’s own docs rather than restated as SEOHack.info’s own research. Before publishing this revision, we specifically searched for any Google Quality Rater Guidelines update dated in 2026 and found none confirmed by Google — only a small number of SEO blog posts describing an unverified “June 2026” revision with section numbers that don’t match the actual document. We’ve flagged that explicitly above rather than repeating it as fact. Where a specific detail — such as the exact wording of the worked AI-content failure example — could not be independently confirmed against the primary PDF, it is described in paraphrase rather than presented as a verbatim quotation.
Sources: 1) Originality.ai, “Google Search Quality Rater Guidelines: Key Insights About AI Use”; Search Engine Land / Google Search Central, on rater scope and process. · 2,3) MedResponsive and Search Engine Journal / PPC Land, coverage of the January 2025 update (170→181 pages). · 5,6,7) Google Search Central developer documentation, “Google Search’s Guidance on Generative AI Content on Your Website” (citing sections 4.6.5 and 4.6.6 directly). · 4,10,11) Search Engine Land and Search Engine Roundtable, coverage of the September 11, 2025 update, including Google’s spokesperson statement that it was “a minor update… [with] no change to our rating guidance.” · 8) Search Engine Journal / Dream Warrior Group, on the January 2025 worked example. · 9) PPC Land, on exaggerated-claims guidance. · 12) TheGuideX, “Google Quality Rater Guidelines Explained.” · 13) Google Search Central Blog, “Our latest update to the quality rater guidelines: E-A-T gets an extra E for Experience,” and subsequent Google statements on Trust as the foundational pillar. · 14) TheGuideX; general E-E-A-T reporting on the Experience pillar. · 15) Google Search Central developer documentation (section references); Keywords Everywhere, “Google E-E-A-T Guidelines: an Overview.” · 16) Google Search Central developer documentation on generative AI content; Semrush, “AI or Human: Google Expands Rules on Low-Value Content.” · 17) General reporting on the guidelines’ public availability. · 2026-update claims checked against and found unconfirmed by Google Search Central’s public update log as of August 26, 2026.
Related on SEOHack.info: E-E-A-T hacks for AI-generated content, humanizing AI content, optimizing for Google’s AI Overviews, and our Content Evaluator tool for a deeper page-level audit.