Content SEO and E-E-A-T
E-E-A-T stands for experience, expertise, authoritativeness and trust, and on the page where Google introduces them it writes that of the four, trust is most important and the others feed into it. The same page says E-E-A-T itself is not a ranking factor, which means what actually gets measured is the evidence you put on the page.
- Lesson 6 of 15
- Intermediate
- Free, no signup
Three pillars and a roof, not four pillars
The shape comes from Google's own sentence: of the four aspects, trust is most important and the others contribute to it. So trust is the roof and the other three hold it up.
-
Experience
Have you done the thing yourself? The scarcest of the three.
-
Expertise
You know the subject and you know where it usually goes wrong.
-
Authoritativeness
Others in the same field point at you.
None of these three is ever finished and none of them carries a score. Google states plainly that E-E-A-T itself is not a ranking factor.
Last checked: Facts and tool names in this lesson are re-checked against their sources on this date.
What is E-E-A-T, and why does trust outrank the rest?
Four letters, four words: experience, expertise, authoritativeness and trust. Content SEO is the work you do so those four are visible on the page itself, rather than claimed somewhere else.
On the page where it introduces the four, Google has one sentence that most summaries drop: of these aspects, trust is the most important, and the others contribute to trust. So they are not four separate jobs to tick off one by one. Three of them are evidence and one is the result.
The second sentence matters even more. Google writes that E-E-A-T itself is not a specific ranking factor; what its systems use is a mix of factors that can identify content with good E-E-A-T. The difference between those two sentences is everything. There is no E-E-A-T score that moves up and down, which is also why no plugin and no tool can hand you one.
And a point that gets said less often: a page does not have to show all four. Google gives the example itself, that some content may be helpful because of the experience it demonstrates and other content because of the expertise it shares. A review of a pair of headphones by someone who used them for six months needs no technical expertise. It has experience, and that is enough.
The three questions Google asks about your content
Instead of vague criteria, Google proposes three plain questions, and those three cover the whole subject: who wrote it, how it was made, why it was written at all.
The "who" question wants something specific: a reader should be able to tell who created this and why their word is worth relying on. That means a real name, not "our content team". A name someone can find somewhere else too.
The "how" question is about method. If the piece is a comparison, the reader has a right to know how many things were examined and on what basis. Google says plainly right here that this question also covers automated and AI-generated content, and that describing the process helps readers understand what role the automation played.
The "why" question is, in Google's own words, perhaps the most important one. The right answer is that the content was made primarily to help people, not to attract clicks from a search engine. The wrong answer is visible from outside as well: a page with three paragraphs of preamble that then answers in one line was written for a ranking, and a reader works that out in ten seconds.
These three questions have one good property: they cannot be ticked off. You cannot solve "why" by adding a tag.
The same three questions, once answered and once only dressed as answered
Actually answers
- An author name you can also find somewhere else
- A sentence saying how you measured or checked this
- Every number with a source you can open
- A boundary: where this advice stops working
Only looks like an answer
- Author: the content team
- "According to research" with no research named
- A date nudged forward every month with no change to the text
- A page that never says where it can go wrong
The right-hand column is not penalised. It simply gives the reader nothing, and a page that gives nothing loses its place to one that does.
What does evidence on the page look like?
This is where most articles about E-E-A-T stop, because from here on it takes work.
Evidence means something a reader can check. A number with its source next to it. A sentence saying where and when you measured that. A name attached to a real profile. A date saying when the information was last reviewed. And more than any of those, a boundary: the place where you say yourself that this advice does not apply.
The checklist version of the same things looks like this: an author box with a stock photo and a ten word bio, a date that gets nudged forward every month, and the phrase "according to research" with no research named. All three have the appearance of evidence and none of them is evidence.
One simple exercise that genuinely works: open your own article and mark every sentence that makes a claim. Now ask, for each one, how the reader is supposed to know it is true. The sentences with no answer are exactly the ones that make your page weaker than any competitor whose sentences have one.
Naming what you do not know is evidence too. Writing "we have not measured this" takes nothing away from your standing; on the contrary, it is the only thing that makes the rest of your numbers believable.
Experience is not the same thing as expertise
The first letter of E-E-A-T was added later than the others, and it is exactly the one with the least presence in Persian content.
In the list of self-assessment questions it publishes, Google asks this one separately: does your content clearly demonstrate first-hand expertise and a depth of knowledge, for example expertise that comes from having actually used a product or service, or from visiting a place. The load-bearing words are "actually used".
Now look at the market. A hundred articles about one product, all built from the same datasheet the manufacturer published. None of them lies and none of them adds anything to the world. The first person to write what broke after three months of use owns the only page that is distinguishable from the rest.
This is good news for small businesses. Experience is not something you can buy, and a larger competitor does not automatically have it. A repair technician who has opened one particular model for ten years knows things about that model no large site knows. The problem is that they do not write it, because they think an "article" means something that looks like every other article.
And a warning: fabricated experience is the worst case of all. The sentence "we have seen this in our own projects" with no such project behind it is the kind of lie that, once caught, does not reduce trust but zeroes it.
Helpful content in Google's terms: what not to do
On the same page, Google lists questions where a yes should make you reconsider how you produce content. Four of them matter most.
- Is the content primarily made to attract visits from search engines?
- Are you producing a lot of content on many different topics in the hope that some of it performs?
- Are you using extensive automation to produce content on many topics?
- Are you mainly summarising what others say without adding much value?
The third and fourth are the ones that matter here, because today they are glued together. Language models have made summarising other people endlessly cheap, and the result is sites publishing two hundred pages a month, none of which holds anything that was not already in the first ten results.
Google's spam policy has a specific name for this and writes its own definition: scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users. There is a clause in that same definition that rarely gets quoted and settles the whole argument: no matter how it is created.
And one point that applies directly to daily work: summarising is not forbidden. Summarising plus one thing only you have is exactly what a good page does.
Where is YMYL, and why is the bar higher there?
Google writes that its systems give more weight to strong E-E-A-T for topics that could significantly affect the health, financial stability or safety of people, or the welfare of society. It calls that group YMYL, short for "your money or your life".
So a piece on medication, investment advice, law and safety topics sit in one group, and a headphone review sits in another. Which means that if you work in a YMYL area, the things that are merely nice to have elsewhere are entry conditions here: an author with a real name and real standing, a source for every numeric claim, and a review date.
A common mistake is the site outside healthcare that publishes a few health articles to pull traffic. Those articles get judged by the YMYL bar, and the site meets none of its conditions. The result is that the pages stay ineffective and the content team's time is gone.
The boundary of this is worth stating too: YMYL is not an official label attached to your domain. It is the description of a topic group, and the one page you write inside that group is judged by that bar, not your whole site.
Trust is built on the site, not inside the article
A claim sits in a sentence, the sentence sits on a page, and the page sits on a site. Trust is built at all three layers, and most content teams work only on the middle one.
The site layer asks for simple things and each of them is done once: a page saying who you are, a page naming the team, a real way to make contact, and an editorial policy that says how content is produced and checked. None of that needs schema or a plugin.
We have done exactly this on this site and you can open it and look: an editorial policy is published, and it says how content gets reviewed and where AI sits in the process. The fact that the page can be opened at all is itself part of the evidence this lesson is about.
And the last thing, which is the hardest for most sites: if content is part of your work and you do not have time for it, handing it over is fine. What we do under content production is exactly that. But the one thing that cannot be handed over is your own first-hand experience; you have to say it yourself, even if somebody else writes it down.
Trust is built in three layers
From the inside out. Most content teams build only the middle layer, then wonder why their good article does nothing.
-
1
The sentence
The claim with its source or its measurement, in the same sentence.
-
2
The page
A named author, a review date, the method, and a stated boundary.
-
3
The site
An about page, a team with names, a real contact route, an editorial policy.
The outer layer is built once and works for every page. The inner one is rewritten every time.
The fast path, with AI
What a language model is genuinely good at here is not writing, it is finding: catching the sentences that make a claim and carry no evidence. That is the job you do badly on your own text, because you know what you meant and your mind fills the gap. A cheap fast model of the Flash class is enough, since this is extraction rather than judgement; our current pick is in the <a class="text-link" href="/en/ai/">AI section</a>.
- Take the raw text of the article, without the headings and without the site menu. Just the paragraphs.
- Run the recipe below. The output is one table: each claim, the evidence present in the text, and if there is none, what would be needed.
- Answer the rows with no evidence yourself. Either find the source, or measure it, or shrink the claim until it matches what you actually know.
- Rows with neither a source nor a measurement get deleted. That is the hardest step, and it is the one that separates the page from the rest.
Copy-ready recipe
Role: claim reviewer. Your job is not to write, only to find.
Text:
{article text}
What you do:
1. Isolate every sentence that makes a checkable claim. A checkable claim means a number, a date, a comparison, a cause-and-effect relation, or a quotation from a source.
2. Build one table row for each, with these columns: the sentence, the type of claim, the evidence present in this same text, the verdict.
3. The verdict is only one of three: has evidence, has no evidence, has vague evidence. "Vague evidence" means it points at something unnamed, such as "research" or "experts".
4. Under the table, for every row whose verdict is "has no evidence", write one line naming the kind of evidence that would close that claim: an external source, a measurement by the author, or shrinking the claim.
Rules:
- Do not propose any source and do not write any URL. If you do not know the source, only say what kind of source is needed.
- Do not rewrite the text and do not add sentences.
- Leave out sentences that are the author's personal opinion and carry no checkable claim.
- If one claim is repeated twice in the text, make one row and say how many times it appears.
Before you trust the output: Two things you have to accept yourself. First, the output of this recipe is usually longer than you expect, and the first reaction is that it is not that bad. That reaction is the reason the exercise exists. Second and more important: the model does not know whether your claim is true or false; it only says whether evidence sits next to it. Filling the evidence column is your job and no part of it should be handed to the model, because a model asked for a source will invent one. And where a claim has neither a source nor a measurement, the right answer is deleting the sentence, not softening it.
AI in this kind of work
The position of this lesson is simple: assistance yes, publishing at scale without editing no. A language model can draft, propose a structure and find the sentences with no evidence. What it cannot supply is exactly what E-E-A-T is made of: what you did, what you measured and what broke in your hands. The full version of this argument, with sources and mechanism, is the AI content and SEO lesson further along this path.
Tools that actually help
- NotebookLM For when you already have the sources and want the text built only from them: you upload the sources yourself and the answers come from that set. Google's own guide says it works in the same regions where the Gemini app works, and Iran is not on that list.
- Claude For running the claim-review recipe and for getting a structure, because it takes constraints and writes down why it ruled the way it did. Iran is not on Anthropic's supported-countries list and there is no official payment route from Iran.
- Gemini For Persian drafting and rewriting, since Persian is on its web app's language list. Google's own page says the app runs in over 230 countries and territories, and Iran is not on that list.
- RGB The access and payment layer for Iran, kept separate from the tools themselves.
Where it backfires
The risk here is not being caught, it is being manufactured. A model asked to make the text "more authoritative" will produce exactly the things this lesson calls evidence: a percentage, a study, a quote from an expert who does not exist. The result reads tidier and every new claim in it is fabricated. The only defence is to never ask a model for evidence and to ask it only to point at where evidence is missing. Google has put the question elsewhere and said it plainly: its focus is the quality of content rather than how it was produced, and what its spam policy targets is worthless content produced at scale. So the tool is not the issue. Publishing something nobody checked is.
Sources: Google Search Central: creating helpful, reliable, people-first content Google Search Central: how Google Search views AI-generated content Google Search Central: spam policies (scaled content abuse)
Where this advice stops
E-E-A-T does not rank a page. It describes what Google's systems try to identify, not a lever you pull. On a phrase with no demand, or where the top ten results are all forums and social networks, no author box and no about page changes anything. And the quality rater guidelines are, in Google's own words, feedback used to check whether the algorithm is working; raters do not rank pages and their scores do not go straight into the results.
From our own work
On this site, the portfolio page carries three client quotes and each one is attributed to a real, openable site rather than to a first name with a photo. What you will not find in that page's source is Review or AggregateRating schema: neither is printed, deliberately, and the reason is written in the comment above that function in our own code. A review you collected yourself from your own customer is not eligible for stars in the results on a service or about page, and marking it up to get stars is exactly the move that damages E-E-A-T rather than building it. A star in a Google result with no vetted review behind it is precisely what this lesson called the appearance of evidence.
Real follow-up questions
Is E-E-A-T a ranking factor?
No, and Google says so itself. What exists is a mix of factors its systems use to identify content with good E-E-A-T. So there is no score to raise; there is only evidence you either put on the page or do not.
Should I put an author box under every article?
The box is not the point, the name is. What works is a real name a reader can also find elsewhere and use to see why this person is qualified to say this. A box holding only a photo and two lines of praise is what this lesson called the appearance of evidence.
Does content written with AI get penalised?
Google says its focus is the quality of the content and not how it was produced. What its spam policy targets is worthless content made at scale, whatever tool made it. So the right question is not what wrote it; the question is what is in it that was not already in the first ten results.