Two things happened in the back half of 2024-2025 that reshape how you should think about E-E-A-T:
- September 2025 Quality Rater Guidelines update — Trust is now framed as a hard cap on overall E-E-A-T quality. You can't out-experience or out-expertise a trust deficit.
- May 2024 Content Warehouse API leak — confirmed
siteAuthorityexists as an explicit field inCompressedQualitySignals. Google denied "Domain Authority" for years. The denial was wrong.
This post is the operating manual on what changed and how to score yourself.
The Trust cap — what it actually means
E-E-A-T has four components: Experience, Expertise, Authoritativeness, Trust. The Sept 2025 QRG update changed how they combine.
Pre-2025: E-E-A-T was effectively a weighted average. A high-expertise medical article from a respected doctor on a sketchy site could still rank.
Post-Sept-2025: Trust caps the entire rating. Specifically (paraphrased from QRG §4.5.4): a page can have demonstrated experience and expertise, but if the website lacks trust (poor reputation, no clear ownership, security issues, scam complaints), the page cannot earn high quality, regardless of the content's quality.
Our E-E-A-T analyzer applies this server-side:
const rawOverall = weightedAverage(experience, expertise, authoritativeness, trust);
const cappedOverall = trust.score < 50
? Math.min(rawOverall, trust.score)
: rawOverall;
If your trust score is 40, your overall E-E-A-T is capped at 40, no matter how brilliant your authors are. We flag this explicitly as cappedByTrust: true so you know which lever to pull.
YMYL broadening — it's not just health and finance anymore
YMYL ("Your Money or Your Life") was historically a narrow category — medical advice, financial advice, legal guidance. The Sept 2025 QRG broadened it implicitly. Per Lily Ray's March 2026 core update analysis and the DollarPocket correlation study:
- E-E-A-T-aligned signals correlate with ~8% of ranking variance across all queries in the cited correlation studies. (Correlation, not causation — Google does not publish ranking weights.)
- For YMYL queries that correlation jumps to ~24% — roughly 3× the impact.
- YMYL now extends to any topic where bad advice can cause real-world harm: career advice, parenting, technical security recommendations, vehicle maintenance, product safety — not only the historical med/legal/fin.
In practice this means: if your category has any "wrong answer = real-world consequence" angle, you're effectively in YMYL territory and Trust caps matter 3× more than the headline number suggests.
Our audit auto-classifies YMYL via LLM judgment over the homepage + category, surfacing ymyl: true | false and adjusting the recommendation weights accordingly.
The Content Warehouse leak — what was confirmed
On 27 March 2024, a bot named yoshi-code-bot committed 2,596 module definitions to a public GitHub repo. The docs weren't pulled until 7 May 2024. The contents were analyzed by Rand Fishkin and Mike King and published on 27 May 2024 (iPullRank; Search Engine Land).
Google spokesperson Davis Thompson confirmed the leak was legitimate. What the leak — corroborated by Pandu Nayak's DOJ testimony in US v. Google — proved:
| Signal | Where | What it does |
|---|---|---|
siteAuthority | CompressedQualitySignals | Site-wide authority score. Google denied "Domain Authority" for a decade. |
goodClicks / badClicks | NavBoost | Pogo-sticking is a ranking signal. |
lastLongestClicks | NavBoost | Final, longest dwell in a session — counts more than first click. |
chrome_trans_clicks | ChromeInTotal | Click telemetry from Chrome powers sitelinks. |
| Author entity recognition | Authors | Bylines are stored as entities, not text strings. |
hostAge | Per-domain | Sandbox-like signal for new domains. |
Pandu Nayak confirmed under oath that NavBoost uses a rolling 13-month window of aggregated click data. User clicks are a ranking signal, contrary to a decade of public Google denial.
The implication for your E-E-A-T strategy: site-wide authority is real and measurable. Patrick Stox of Ahrefs urged caution (blog) — presence of an attribute doesn't prove live use — but the corroborating DOJ testimony makes the click-and-authority signals as confirmed as anything outside Google's own public docs.
Page-level E-E-A-T signals you can actually fix
Here's the checklist our audit runs:
- Author byline with
Personschema —PersonJSON-LD withsameAslinking to LinkedIn + ORCID + a credible directory. datePublished+dateModified— both inArticleschema and visible to readers. StaledateModified(>12 months) reads as abandoned.- Author bio block on the page — credentials, relevant experience, links to other work.
- About page with named team + photos — Trust signal #1.
- Contact page with real address + phone — required for trust on YMYL.
- HTTPS + valid cert — non-negotiable.
- No deceptive overlays / interstitials — explicit QRG §4.5 demerit.
- Wikipedia / Wikidata entity — see our schema-for-AI post.
- Outbound links to primary sources — citing is a trust signal, not a leak.
- No scaled / auto-generated content — Google's March 2024 spam policy update made this an explicit demerit.
We score each of these and surface the gap. The trust cap is the most common one we flag — sites with 70+ on experience and expertise but 35 on trust, capped to 35 overall.
Sources
- Quality Rater Guidelines (Sept 2025) — https://guidelines.raterhub.com/searchqualityevaluatorguidelines.pdf
- iPullRank Content Warehouse analysis — https://ipullrank.com/google-algo-leak
- Search Engine Land leak coverage — https://searchengineland.com/google-search-document-leak-ranking-442617
- Lily Ray, March 2026 core update analysis — https://www.amsive.com/insights/seo/google-march-2026-core-update-winners-losers-analysis/
- Patrick Stox cautionary read — https://ahrefs.com/blog/google-documents-leaked-seos-are-making-some-wild-assumptions/
Curious if your E-E-A-T is trust-capped? Run our free audit — we score all four E-E-A-T pillars individually, flag YMYL classification, and call out the cap explicitly when it bites.