Why it matters who signs, not just what is written

The question usually comes around the fourth month of working together, phrased almost identically every time: “we have good content, why aren't we moving up?”. Almost always, the answer has nothing to do with the text. It has to do with the fact that, from where Google is looking, there is no way to see who writes it.

E-E-A-T is the abbreviation Google uses for four signals: experience, expertise, authoritativeness and trust. This chapter explains what can be concretely demonstrated for each, what cannot, and why even a well-built checking tool draws wrong conclusions if you take it at its word.

What it is and, above all, what it is not

The first correction, the one that saves you from spending money for nothing: there is no E-E-A-T score at Google. The official documentation says explicitly that “E-E-A-T itself isn't a specific ranking factor”, meaning it is not a value the system calculates for your site and that you could push up directly.

What exists instead are people. Google pays raters who receive search results and grade them according to a public guide of a few hundred pages. Their grades do not change the position of any site, but they serve to train and check the automated systems. That guide describes what a good result means, and E-E-A-T is the vocabulary in it.

The practical consequence: you do not optimize a score, you make verifiable the things a rater, an algorithm or an AI assistant looks for on the page. The difference seems subtle and it is not. The first path leads to tricks, the second to a site that can be trusted.

Trust cancels out the other three

The four letters do not carry equal weight. The Rater Guidelines are unusually blunt:

Trust is the most important member of the E-E-A-T family because untrustworthy pages have low E-E-A-T no matter how Experienced, Expert, or Authoritative they may seem.

Google Search Quality Rater Guidelines, version of 11 September 2025

Translated into consequences: a site with excellent articles, signed by a real specialist, which however does not say who operates it, is not “a good site with a small gap”. It is a site with low E-E-A-T, no matter how good the articles are. Trust does not add to the others, it caps them.

The bar rises further for the sites the guide calls YMYL, that is, those that touch the reader's money, health or safety. For them, the requirement is worded without nuance:

Any site that handles personal, private or sensitive data must provide extensive contact information.

The Rater Guidelines, section 4.5.1

A shop, a clinic, an insurance broker or a law firm without complete identification details do not get a lower grade. They get the lowest grade.

A detail that matters for Romanian sites and that international tools miss: the same guide states that “different locales... may have their own conventions and standards”. A unique registration code and a registration number in the Trade Register are exactly the local equivalent of the information the guide asks for. Tools that look for American patterns do not recognize them and report something missing where everything is in order.

What can be shown and what remains empty talk

This is the line that separates a credible site from one that merely boasts. The guide draws it itself:

E-E-A-T assessments should be based on the MC itself... verifiable credentials, etc, not just website or content creator claims of “I'm an expert!”.

The Rater Guidelines, section 5.6

In other words, a claim about yourself is worth nothing without something to back it up. The table below is the direct translation of this rule into elements that either exist on the site or do not.

The last row deserves insistence, because it is the most widespread mistake and the easiest to make in good faith. Many site themes offer a box with stars and an average rating, marked in the code as structured data. The guide is explicit about its value:

“Reviews” by the product manufacturer (“Our product is great!”) or “reviews” from an influencer who is paid to promote the product are not as trustworthy due to the conflict of interest.

The Rater Guidelines, section on trust

A rating you award yourself does not add trust. At best it is ignored, at worst it is read as an attempt to mislead, which hits exactly the letter that caps everything else. We removed such a box from our own site for this reason.

Three sites in the same field, three results

Because none of the existing tools checked what needed checking for Romanian sites, we built one. It analyzes between ten and sixteen pages of a site, not just the home page, because some contradictions only show up by comparison: the company is called one thing on “about us”, another in the terms and conditions, and the declared experience is eighteen years in one place and twenty-eight in another.

The clearest example comes from insurance, where we have three sites in the same regulated field, with the same legal identification obligations. Measured on 18 August 2026:

Same rules, same field, thirty-one points of difference. It is worth unpacking, because each of the three illustrates something different.

What one hundred means

It does not mean “perfect E-E-A-T”, as we explain below. It means that all the signals that can be checked by reading the pages are in order: the company is named with its legal form, the tax code passes the check digit validation, the Trade Register number exists, the six content pages are signed, and the articles cite sources from outside the site.

How you lose nine points without doing anything wrong

The second site has an identity just as good. It loses exclusively on content, for two reasons that are often heard as a reproach: the pages are short, and the text never says “we did”, “we came across”, “this is what happens with us”. It is written impersonally, correctly, and interchangeably with anyone else's.

A design detail shows up here too. The site has no articles, so the “author” dimension appears as zero out of zero and drops out of the calculation, instead of being scored zero. A brochure site without a blog is not punished for not having a blog. The rule, which we apply everywhere: no points are deducted for a test that had nothing to check.

How you lose thirty-one

The third site fails on something very concrete. It has a single article, and that one has no author name, neither in the text nor in the structured data, and it does not link to anyone's profile page. All twenty points of the dimension are lost at once. Plus two more, because the contact page cannot be found starting from the home page.

This is the practical form of this chapter's title. The content is no weaker than elsewhere. It is just that nobody knows who writes it, and from where Google is looking, that difference weighs a fifth of the grade.

The figures, however, are not the interesting part of the chapter. A tool built by us, which gives us high grades, proves nothing. The part worth reading is the other one: how many times it got things wrong before getting here.

Seven times the tool said something untrue

Each of the errors below was discovered by running the tool on real sites, not by reading the code. Each looked, on screen, exactly like a real problem.

To which two errors of the same nature are added: the registration numbers of sole proprietorships, written with a letter in front, were not read at all, so a correctly registered business appeared unregistered. And one check had ended up contradicting itself, asking us to specify something that was already specified in the sentence it had just read.

And one that is still there

The ones above are fixed. The next one is not, and we write it down precisely because it is more instructive that way.

The tool tries to recognize on its own whether a site touches money, health or safety, in order to raise the identification bar. Of the three insurance sites in the table above, the first two are classified correctly. The third, which sells exactly the same type of product, is not, because it has too little text for the signals to cross the recognition threshold.

The practical effect is small here, because the site has complete identity anyway. But if it had not, its grade would have been gentler than it deserves, and we would have read a reassuring report about a site with a real problem. A tool that errs in your favor is more dangerous than one that errs against you, because nobody disputes good news.

The conclusion that matters for you is not the list of errors, but what is learned from it. An automated tool tells you where to look, not what is true. Any audit report, from anyone, including from us, is a list of places to check. The verdict comes from a person who looks at the page.

A minor case from the same measurement is worth adding here. We asked the tool to analyze one of our domains and it reported results for another. It was not an error: the domain redirects, and the tool reports the host it actually lands on, not the one requested. Without this behavior, you would receive findings about one site and apply them to another.

What cannot be measured remotely

The Rater Guidelines require reputation to be established from three directions: what the site says about itself, what is visible on the page and, essentially, what others say. The last is the hardest and it is completely outside your site: independent reviews, press mentions, references from other professionals, discussions in communities.

No tool that reads the code of a page can evaluate that. One that claims otherwise is selling you an invented figure. For this reason, our tool explicitly declares reputation as not measurable remotely and does not score it at all, instead of recording it as zero. The rule we apply everywhere: no points are awarded and none are deducted for a test that did not run.

Which also means that a score of one hundred does not translate as “perfect E-E-A-T”. It translates as “all the signals that can be checked automatically are in order”. The hard part, what others say about you, remains to be built over time and is not solved from code.

Sources

• Google Search Quality Rater Guidelines, version of 11 September 2025, sections 2.5.3, 4.5.1, 5.5 and 5.6.
• Google's documentation for creators, “Creating helpful, reliable, people-first content”.
• The scores in the table were measured on 18 August 2026 with the audit tool on goai.ro.