E-E-A-T: what Google checks and where it is decided

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trust. The term comes from the Google Search Quality Rater Guidelines. It describes the grid that trained evaluators apply when they judge the quality of a page.

Google employs people for this job. They are called Search Quality Raters. Their verdicts do not feed into rankings. They check whether Search is working.

The four letters split into two halves. Experience and Expertise are shown on your own website. Authoritativeness and Trust are largely decided elsewhere.

This article walks through both halves using the sentences from the primary source. It is written for anyone responsible for a website or a topic cluster who wants to know which part of the work sits on their own domain.

It closes with seven widespread statements about E-E-A-T checked against Google's documentation. Three of them do not survive the check.

What the four letters stand for

The acronym has four parts. Google describes them in section 3.4 of the Search Quality Rater Guidelines, in the version dated 11 September 2025.

Each letter carries a question about the page and about the person or organisation behind it.

Google explains each letter with a comparison. For Experience it asks which product review you would believe, the one from someone who used the product or the one from someone who did not. For Expertise it contrasts a skilled electrician with an enthusiast of antique homes who knows nothing about wiring.

For Authoritativeness Google gives two examples of very different size. A local business profile can be the reliable source for what is on sale today. The official government page is the single authoritative source for renewing a passport.

The order inside the acronym says nothing about weight. Google ranks the four differently.

Trust sits at the centre of the four letters

Google puts this plainly. Trust is the most important member of the E-E-A-T family. The other three support it.

The reason follows immediately. An untrustworthy page has low E-E-A-T no matter how experienced, expert or authoritative the creator appears. Google illustrates it with a financial scam whose author is a recognised expert at running scams.

One practical consequence follows. Working on expertise while leaving the trust question open only moves the effort around.

Two further details from the primary source belong here. Content does not have to demonstrate all four aspects. A piece can be helpful purely through the experience it shows.

Where the second E in E-E-A-T came from

Until late 2022 the grid was called E-A-T. It covered Expertise, Authoritativeness and Trustworthiness.

On 15 December 2022 Google published a post titled E-A-T gets an extra E for Experience. Experience has been part of it since.

Google justifies the addition with an everyday example. To fill in a tax return you want the accountant. To judge a tax filing product you want people who have used it.

The same post carries a sentence that most summaries drop. The guidelines do not directly influence ranking.

E-E-A-T, EEAT or E-A-T, which spelling applies

Three spellings circulate for the same concept. Google writes E-E-A-T with hyphens. EEAT without them fills articles and search queries. E-A-T is the version before December 2022.

Google offered a fourth form in its 2022 post. Anyone who prefers it may say Double-E-A-T. All of them name the same section. The spelling changes nothing about what the acronym means.

The last letter changed along the way. E-A-T ended in Trustworthiness. Section 3.4 now reads Trust.

The grid also gets quoted as a standard for search engines at large. It comes from one Google document and describes Google's own quality assessment.

The Search Quality Rater Guidelines as the origin of the term

The document is titled "General Guidelines". The current version carries the date 11 September 2025 on page 1 and runs to 182 pages. Google publishes it openly as a PDF.

It is written for the people who evaluate search results. The wording never addresses site owners.

Who the Search Quality Raters are

Search Quality Raters are contracted evaluators. They receive search queries with results. They award two kinds of marks.

The first is called Page Quality and judges the page. The second is called Needs Met and judges how well a result matches the query.

E-E-A-T belongs to the first mark. The scale runs from Lowest through Low and Medium up to High and Highest.

Raters work inside their own rating locale. Google explicitly requires them to represent the people who live there.

What Google says about rating and ranking

Google leaves little room in its documentation on helpful content. Raters have no control over how pages rank. Their data is not used directly in the ranking algorithms.

Google compares the arrangement to feedback cards in a restaurant. The cards show whether the kitchen works. They do not change the dish on the plate.

The same document carries the second half. E-E-A-T itself is not a specific ranking factor. What is useful is a mix of factors that identify content with good E-E-A-T.

Read together, the two sentences give the workable interpretation. There is no E-E-A-T dial inside Search. There are signals meant to measure the same thing a rater judges by hand.

That is why the document still pays off. It describes the result Google considers good.

What separates E-E-A-T from SEO in the narrow sense

SEO in the narrow sense works on findability. Crawling and indexing come first. A crawler fetches the page. The index takes it in. Then the content has to match the query the page should appear for.

Those steps sit before the assessment. A page missing from the index appears in no search results. No rater ever sees it. Its E-E-A-T never enters a rating.

The quality assessment builds on top of that. Its basis is the guidelines. They ask about the creator, about first-hand experience and about what other people report.

The two levels have different arbiters. An algorithm decides the position. A person awards the Page Quality rating.

The two halves of E-E-A-T

Section 3.4 of the guidelines names three sources of evidence for the assessment. That short list is the core of the whole topic.

• What the website or content creators say about themselves.
• What others say about the website or content creators.
• What is visible on the page, including the main content, reviews and comments.

Two of those three sit on your own domain. The third sits elsewhere. Only the third one is trusted by Google to overturn a self-description.

The four letters split along the same line. Experience and Expertise become visible on the page. Google defines Authoritativeness as being known as a go-to source, and that label by definition comes from other people. Trust draws on both sides.

The guidelines make the research outside the domain compulsory. Reputation research is required for every Page Quality rating task. Not for YMYL pages, not for borderline cases, for every task.

The search query Google uses to check the second half

Section 3.3.3 gets very specific. Google hands its raters the queries they should run to research the reputation of a website.

The example in the document reads [ibm -site:ibm.com]. The operator removes every page on ibm.com. A second variant reads [ibm reviews -site:ibm.com].

That instruction is the clearest confirmation of the split. Google has its evaluators switch off the provider's own domain and look at what remains.

One clarification in the same section surprises many people. A company's official social media profiles do not count as independent sources either.

Experience and Expertise on your own website

This half can be planned. It consists of statements and evidence that live on the page.

It is also the half that most guides cover in full. Author box, publisher details, source citations. All of that belongs there, and it does not finish the job.

How experience becomes visible on a page

Experience means first-hand exposure. Google asks whether the creator has the necessary first-hand or life experience for the topic.

It shows up in material that can only exist if somebody actually did the thing.

• photographs taken on site rather than stock images, tied to the case being described
• numbers from your own operation, with a period and a base population
• intermediate steps, failed attempts and cancellations that no summary of other sources would contain
• an honest note about who the thing suits less well

Google draws the same distinction in its guide to generative search. A first-hand review offers a unique perspective. A summary of existing content restates what is already available elsewhere.

Weak version: "Our team brings years of hands-on expertise across a wide range of industries."

Better version: "Since January 2024 we have run eighteen campaigns in seven languages. We stopped two of them after the first quarter because the landing pages could not carry the traffic."

How expertise becomes visible on a page

Expertise means knowledge or skill for the topic. Google asks for no diploma. The bar moves with the subject.

The guidelines list three counter examples that make the point fast. A restaurant review written by somebody who has never eaten there. An article about skydiving written by somebody with no expertise in it. Tax form downloads offered on a cooking website.

Expertise shows up as precision. A technical term is used correctly. An exception gets named. A figure carries its date.

One warning appears in the document word for word. The assessment should rest on the main content, on reputation research and on verifiable credentials, and not on a creator claiming "I'm an expert!".

Publisher, author and source details

Google recommends author information outright. The documentation strongly encourages adding accurate authorship information, such as bylines to content where readers expect them.

The guidelines draw the line harder in the other direction. For YMYL pages and other pages that require a high level of user trust, an unsatisfying amount of information about who is responsible leads to the Low rating.

In practice that means three things:

• An article names a person as the author rather than the brand.
• The name leads to a page carrying the role, the topics and the credentials.
• The details about the operator live at a fixed address reachable from every page.

A grounding page covers the third point well. It gathers legal name, services, prices and evidence in verifiable form at a single address.

Overstating backfires here. Information that looks exaggerated or mildly misleading also earns the Low rating.

Where the first half runs out

Everything listed so far sits on your own domain. That puts it inside the first of the three evidence sources, which is what a website says about itself.

Google tells its evaluators to be skeptical of such claims. Particularly where a conflict of interest is visible.

A customer quote on your own homepage falls into the same bucket. It is there because the provider selected it.

In our experience the bottleneck for most websites sits here. The author box exists. A search that excludes the brand's own domain still returns nothing usable.

Authoritativeness and Trust away from your own website

This half cannot be written. It has to be reported.

For Google, Authoritativeness is not a property of a text. It is a label. A creator or a website counts as the go-to source for a topic when other people treat it that way.

Google describes the top rung accordingly. A website or creator who is the uniquely authoritative, go-to source for a topic has very high E-E-A-T.

The self-test in four queries

The test takes ten minutes and uses the queries Google hands its raters. Replace brand.com with your own domain.

• [brand -site:brand.com]
• ["brand.com" -site:brand.com]
• [brand reviews -site:brand.com]
• [brand topic -site:brand.com], with the topic the site is meant to rank for

The fourth query tells you the most. The guidelines require reputation research to match the topic of the page. A website can be the reliable source on one subject and not on another.

Look at the first page of results and answer three questions:

• Did a person or an organisation write the text?
• Does it sit on a page the company does not own?
• Does it deal with the topic the brand wants to be known for?

What counts as evidence and what does not

Section 3.3.1 gets unusually concrete about what carries weight in reputation research.

Two rows deserve a closer look.

Google asks for information written by a person or an organisation, not statistics and not machine-compiled data. Traffic figures from services such as Similarweb are named in the document, with the instruction to ignore them.

Customer reviews count, with reservations. Raters are told to stay skeptical of positive and negative reviews alike, because anyone can write them. The content of the review matters more than the star rating.

Google adds a sense of scale. A handful of negative reviews is normal for any business. Credible reports of fraud or financial wrongdoing are treated as evidence of an extremely negative reputation.

When the search comes back empty

An empty result is not yet a bad sign. Google states plainly that small websites often have little or no reputation information and that this indicates neither high nor low quality.

A page can earn the High rating without any reputation information at all. That sits in section 7.2.

The instruction for that case is to pay extra attention to the remaining criteria. For YMYL topics, explicitly more.

The reverse direction bites harder. A lack of E-E-A-T cannot be offset. Neither a positive reputation nor the type of website overrides it.

How the missing half shows up day to day

In daily work the gap rarely appears as a ranking drop. It appears as a series of observations that look harmless one at a time.

• The brand ranks for its own name and for very little else.
• Competitors turn up in trade articles where the brand is absent.
• An AI system describes the offering wrongly, drawing on an old directory entry.
• Two companies with similar names get merged into one.
• Good pages sit at position seven for months.

Those observations share one root. Too few independent pages say the same thing about the brand.

In our experience the first three arrive together. Teams usually notice the third one first, because a customer quotes it back to them.

How evidence on other websites comes about

Up to here the article has described how to recognise a good result. Now to how it comes about.

The starting point stays the instruction given to raters. They look for independent reviews, references, recommendations by experts and news articles about the provider. Every one of those findings is a page on somebody else's domain.

What makes a mention usable

Not every namecheck works as evidence. The criteria follow directly from the guidelines.

• The page does not belong to the company and is not maintained by it.
• A human wrote the text, visible in the byline and the editorial context.
• The topic of the page matches the topic the brand wants to be known for.
• The mention says something verifiable rather than only the name.
• For YMYL topics the source comes from the field.

The third point gets overlooked most often. A mention in a subject area unrelated to the business does not improve the reputation for that business.

Weak version: "Providers in this space include A, B, C and D." The company name sits in a list of twenty, and the page says nothing about any of them.

Better version: "Provider D placed forty articles in trade media for a Bavarian manufacturer in the first quarter of 2026 and raised its referring domains from 180 to 260." A rater can check that sentence. It belongs to a topic. The difference between a linked placement and a plain namecheck sits in the article on brand mentions and link building.

Which formats produce that evidence

The guidelines name no formats. They name characteristics. The formats that work in practice follow from those characteristics.

Original data lasts longest. An analysis that exists nowhere else keeps getting cited for years, because nothing replaces it. Suitable material includes operational figures, benchmarks, price comparisons and evaluations of customer data, each with a period and a base population.

Spread matters more than volume. A single mention creates no reputation for a topic. Ten fitting sources beat a hundred arbitrary ones from the same corner of the web.

How long the build takes

Google documents no duration for this. Anyone quoting a number of weeks has estimated it.

The sequence can still be described. First the piece is published. Then it gets indexed. Then the result of a search excluding your own domain starts to change.

In our experience the last step is the slowest. Several weeks between checks makes more sense than days.

How to get hold of it

Two routes exist, and both are legitimate.

The first is your own editorial work. You build a study, an analysis or a comparison and wait for trade media to pick it up. That costs little money and a lot of time, and the timing stays out of your hands.

The second is the deliberate placement. You select publishers that match your topic, check their figures and book the publication. That costs money and delivers a date you can plan around.

Paid placements often get labelled as risky across the board. That reading falls short. The problem lies with link practices built to manipulate rankings. A publication embedded editorially in a topically relevant outlet meets exactly the characteristic Google hands its raters to check.

GetMentioned exists for that second route. The marketplace shows the SEO data, the price and the topics for every publisher, and the booking runs through it directly. If you would rather hand the execution over entirely, we take it on as a link building agency in full service. That covers link building end to end and excludes work on the client's own website, so no full-service SEO.

YMYL, and where Google raises the bar

YMYL stands for "Your Money or Your Life". The label covers topics where wrong content can harm people or society.

Google weights those topics more heavily. The documentation on helpful content states that its systems give even more weight to content aligned with strong E-E-A-T on YMYL topics.

The four YMYL categories

The guidelines separate four areas.

One sentence from the same section rarely travels with the rest. Many or most topics are not YMYL and do not require a high level of accuracy or trust to prevent harm.

YMYL is also a spectrum. Google talks about clear YMYL, about definitely not YMYL and about everything in between. The sharpest scrutiny goes to the clear cases.

Reputation research shifts for YMYL topics as well. There the reputation of a website should be judged by what experts in the field have to say. Recommendations from professional societies are named as strong evidence.

For the top rating the same standard applies in a stricter form. Google records that reputation research matters especially when awarding Highest. A very positive reputation on YMYL topics often rests on recommendations from known experts or professional societies.

For news sites Google names two further proofs. Prestigious awards count, and so does a history of high-quality original reporting.

When a personal account carries a YMYL topic

A common misreading holds that only formal qualification counts on YMYL topics. Section 3.4.1 of the guidelines says otherwise.

Pages sharing first-hand life experience on a clear YMYL topic can be considered to have high E-E-A-T. The condition is that the content is trustworthy, safe and consistent with well-established expert consensus.

Google supplies a table of pairs. On sleeping during pregnancy the personal account carries; on sleep medication during pregnancy the expert piece does. On coping with cancer treatment the forum discussion carries; on treatment options the expert piece does.

The dividing line runs between experience and advice. Describing what you went through needs no licence. Telling somebody what to do needs one.

E-E-A-T in AI search

At Google the same systems run underneath AI search and classic search. The number of claims about E-E-A-T in that context is large, the number of proofs is small. This section separates the two.

What Google documents about generative search

Google publishes a dedicated guide, last updated on 10 July 2026. It is called Optimizing your website for generative AI features on Google Search.

The load-bearing sentence says that the generative features are rooted in the core Search ranking and quality systems. Google describes two techniques. Retrieval-augmented generation pulls current pages from the Search index. Query fan-out issues several related searches for one question.

The guide also lists advice you can ignore. Google Search does not use files such as llms.txt. Structured data is not required for generative search. Chunking content into small pieces is not required either.

What the guide never contains is the term E-E-A-T. Not once.

The documented connection

The mechanism still argues for the second half, and it can be traced in three steps.

First, the generative features run on the same ranking and quality systems, by Google's own account. Second, those systems use a mix of factors that identify content with good E-E-A-T. Third, the grid itself requires the assessment to draw on what other people say.

Those three sentences produce a connection. It runs indirectly, and it can be evidenced from Google's own documentation.

The way the systems work adds to it. The AI models write the answer. The core ranking systems choose the pages. An answer assembled from several sources finds more corroboration for a brand the more independent pages report the same thing. Digital PR builds exactly those pages.

What stays conjecture

No study with disclosed methodology measures a causal link between E-E-A-T and citations in AI Overviews.

Two negative findings belong here. The Quality Rater Guidelines of 11 September 2025 do not mention AI Overviews once in 182 pages. Google's guide to generative search does not mention E-E-A-T.

So anyone calling E-E-A-T the key to citations in AI Overviews lacks the proof. Which characteristics have actually been measured in generative answers sits in the article on generative engine optimization.

Google's own sentence about mentions

The guide contains a paragraph that reads at first glance like an argument against building mentions. The exact wording repays attention.

Google writes that its generative features can show what is being said about products and services across the web, including in blogs, videos and forum discussions. The qualification follows. Seeking inauthentic mentions across the web is not as helpful as it might seem.

The word Google chose is inauthentic. It points at fabricated namechecks with no editorial substance.

A publication that actually happens, in an outlet that actually exists, carrying a statement that is actually true, falls outside that. It falls under what Google describes in the first half of the same paragraph.

Seven widespread statements about E-E-A-T against Google's documentation

Few SEO terms attract as many unchecked sentences. These seven appear in article after article, sometimes word for word.

Two of these deserve an addition.

On the ranking factor, the full formulation matters. Google writes that E-E-A-T itself is not a specific ranking factor and that a mix of factors identifying content with good E-E-A-T is useful. Quoting only the first half turns a clarification into an all-clear.

On expertise, section 3.4.1 settles it. A personal account with no formal qualification at all can carry high E-E-A-T there. In return the guidelines warn against self-description, because a claim is not a credential.

An eighth statement circulates less often and still earns a mention. Many articles hold that AI-written text automatically has poor E-E-A-T. The guidelines say something different. The use of generative tools alone determines neither the level of effort nor the rating, and such tools appear in high quality and low quality production alike.

What does determine the rating sits in the same section. A page drops to Lowest when almost all of its main content is copied, paraphrased or embedded with little effort, little originality and little added value.

How to check E-E-A-T without a metric

There is no E-E-A-T score

Google has never published a number for it. No tool reads it out, because it does not exist.

Vendors displaying an E-E-A-T score calculate something of their own. That can be useful. It is not a Google figure.

Which also means no increase can be measured. Claims about E-E-A-T gaining importance have no baseline to grow from.

What can be counted instead

What is checkable are states, not point scores. Four measures give a picture.

• the number of independent domains mentioning the brand inside its own subject area, counted through the four queries from the self-test
• the share of articles carrying a byline linked to an author page
• whether the operator details are complete and dated
• the generative AI performance report in Google Search Console, which Google names for this purpose in its guide

Visibility in systems outside Google follows its own measurement logic. That sits in the article on AI visibility.

In our experience the first of those four moves slowest and says most. The other three are done within a few weeks.

Conclusion: the second half is the work

E-E-A-T is no dial inside Search. It describes a result Google considers good, together with the manual that people use to judge it.

The first half of it stays manageable. Byline, author page, operator details, original photographs, original figures. That is a few weeks of work followed by maintenance.

The second half sits outside your own domain. Google has its evaluators exclude the site's own pages and look at what remains. Whatever is there was written by somebody else.

In our experience the difference rarely sits in the first half. It sits in the second, because a brand cannot issue its own confirmation.