The quality score and E-E-A-T

What the document says

Section 3.0 reveals the rater’s mental algorithm, in three steps in exact order: first, assess the page’s true purpose — if it’s harmful or deceptive, the score is Lowest and you’re done. Second, assess the potential for harm — pages that hurt people or society, untrustworthy pages and spam are also Lowest. Only in the third step does the classic question appear: how well does the page achieve its purpose? The score lands on a slider from Lowest to Highest, with in-between ratings. And an important warning: harmful pages exist on “official,” respected sites too — government, academia, big stores. Everyone goes through the same filter.

Quality of the main content

MC is judged through four lenses: effort (did a human actively work to create something satisfying?), originality (is there something here that other sites don’t have? if it exists elsewhere, who’s the original source?), talent or skill (is the result good enough to satisfy?) and — for informational and YMYL pages — accuracy and consistency with expert consensus. The document offers a surgical example on effort: building a machine-translation service for users involves real effort; running other people’s content through translation software to generate thousands of pages does not. The rater is told to actually use the page: read the article, watch the video, put a product in the cart, play the game.

Reputation

Reputation research is mandatory in every evaluation — and it must happen outside the site being evaluated. The technique is the exclusion search ([brand -site:brand.com]), looking for what independent sources say: news, Wikipedia, articles, awards, industry bodies. Self-description doesn’t count as evidence; when what the site says about itself conflicts with what independent sources say, the independent sources win. Customer reviews count — but by their content, not their star rating: credible reports of fraud are extremely negative reputation; one late package is not. For YMYL, reputation is measured by what experts in the field say. And absence of reputation is not negative: a small local business living on word of mouth is normal.

E-E-A-T

Experience, Expertise, Authoritativeness and Trust — and the document is explicit about the hierarchy: Trust is the center of the family. The other three exist to support the assessment of trust:

• Experience: does the creator have first-hand experience with the topic? (A review from someone who used the product vs. someone who never touched it.)
• Expertise: do they have the necessary knowledge or skill? (Electrical rewiring explained by an electrician vs. an enthusiast.)
• Authoritativeness: are they recognized as a go-to source? (The official government page for passport renewal.)

The inversion matters: an untrustworthy page has low E-E-A-T no matter how experienced, expert or authoritative it seems. The document’s example is perfect — a financial scam is untrustworthy even if the scammer is a highly experienced expert considered the go-to on running scams. Conflict of interest also breaks Trust: a “review” from the manufacturer, or from a paid influencer, doesn’t count as a review.

Section 3.4.1 delivers the most underrated distinction in the document: on YMYL topics, lived experience can be worth as much as formal expertise — depending on the page’s purpose. A respectful forum thread of people describing how they cope with cancer treatment can have high E-E-A-T. Treatment options and life expectancy cannot — that’s expert territory. Pillow tricks for sleeping while pregnant: experience qualifies. Sleep medication during pregnancy: experts only.

In practice

The order of the three steps is a thesis about risk. Harm and deception are eliminatory before any merit is weighed. An excellent site with one deceptive layer — purpose cloaking, forged authorship, false information — isn’t “good with caveats”; it’s Lowest. When auditing, look first for what can eliminate you, then for what can improve you.

Effort, originality, talent, accuracy — this is the anti-AI-content checklist. The document doesn’t ban generative AI (it defines the term neutrally in section 2.1). What it punishes is the production pattern: scale without oversight, without curation, without anything that doesn’t already exist elsewhere. AI-assisted content with real human editing, proprietary data and an original angle passes all four lenses. An unreviewed automatic content pipeline fails all four.

Run the reputation search against your own site — today. [your-brand -site:yourdomain.com] is literally what the rater will do. What comes up? Complaint boards? Nothing? News? Your external reputation is part of your SEO, and it can’t be fixed with on-site content — it’s fixed with digital PR, customer service and presence in the sources your industry respects.

Treat E-E-A-T as displayable evidence, not a concept. Trust isn’t declared, it’s demonstrated: author bios with verifiable credentials, a corrections policy, cited sources, trustworthy HTTPS and checkout, real support channels. For each template on your site, ask: what evidence of trust does this page show a skeptical stranger?

First-hand experience is the competitive advantage of small sites in YMYL. You won’t outrank a hospital on “treatment options,” but genuine, tested, safe first-person accounts — from patients, users, people who lived it — have their own legitimate place on Google’s scale. The trap is the disguise: personal experience packaged as expert advice lands in the worst of both worlds.