E-E-A-T in the AI era: 4 signals that decide your ranking

This article is for business owners who already know what E-E-A-T is and now want to know what to do differently now that AI answers in place of search engines. If you need the full definition of the four letters, we have a separate article: What is E-E-A-T in SEO. Here we go straight to the four signals machines can read, with fresh survey data and a list of mistakes Vietnamese businesses often make without realizing it.

How E-E-A-T has changed in the AI era compared with three years ago

Three years ago, a good article in eighth place on page one still earned readers, because users scanned all ten results before choosing. Now it's very different. When Google answers with AI Overviews, or when a customer types a question straight into ChatGPT, the machine has to pick two to five sources to build the answer. Eighth place effectively doesn't exist.

That is the whole new pressure. Ranking used to be a ladder: slip a few rungs and you lost some customers, not all of them. Now it's closer to a door: either your name is in the answer or it isn't, with no middle step. Readers see only the sources the machine picked, and most of them don't click further. We analyze this in detail in the article What is Zero-Click Search.

Google says plainly that E-E-A-T is not a ranking factor

This needs spelling out because the Vietnamese SEO market is selling it wrong. E-E-A-T lives in the Search Quality Rater Guidelines, the handbook for Google's human quality raters, with the current version released in September 2025. Google doesn't assign any page an E-E-A-T score. That number doesn't exist in the system.

Here's how it really works: raters read a page and score its quality against the E-E-A-T framework, those ratings are used to train and calibrate the algorithms, and then the algorithms look for technical signals that correlate with human judgment. So E-E-A-T describes the target, and what you can adjust are the signals underneath. Anyone selling you a package to "raise your E-E-A-T score to 90" is selling a number that doesn't exist, because Google's systems have no data field by that name. What the SEO market is selling and how the profession is changing, we cover plainly in the article the SEO profession in 2026.

We go into this because it changes how you work. Chasing a made-up score gets you nowhere: every month the report says "E-E-A-T improved" and nobody can say by how much. Chasing real signals is countable: how many articles now have a real author name, how many author pages declare schema, how many places still have inconsistent business information. The four signals are in the lower part of this article, and for each one we state how to check it.

What really changed: machines pick sources to cite instead of ranking ten links

With the old search engine, the algorithm ranked ten results and let the user choose. A small mistake didn't matter, because the reader made up for it. With AI, the machine has to commit. It puts a source's name in the answer and takes responsibility for that content. A model caught citing junk sources loses credibility immediately, so every vendor has tightened source selection far more than ranking ever was.

The consequence for businesses is a bit harsh. Accurate content published anonymously on a website nobody knows gets skipped by the machine. Equivalent content published under a person with a clear profile, on a domain that industry press has mentioned, gets picked. The quality of the writing is the same on both sides, and the result is different. We break down how machines expand one question into many sub-queries and then gather sources in the article What is Query Fan-Out.

The numbers: what actually gets you cited by AI

In this section we give the numbers first and the commentary after, so you can check them against the marketing budget you're spending. Ahrefs analyzed 75,000 brands, measured each brand's visibility on ChatGPT, AI Mode and AI Overviews, and then computed rank correlations with familiar metrics. The full version is in the study of 75,000 brands published December 12, 2025, and we pulled out the three numbers most worth remembering.

• YouTube brand mentions: about 0.737. Highest on the list, and also the one Vietnamese businesses notice least. Ahrefs counts brand names appearing in video titles, descriptions and captions.
• Brand mentions across the web in general: 0.66 to 0.71. This includes mentions that carry no link at all.
• Link metrics: very weak. Domain Rating measured on ChatGPT reached just 0.266, and the number of pages on a website was 0.194.

Brand mentions beat backlinks by a wide margin

A weak correlation doesn't mean backlinks are useless, and we're not saying that. Backlinks still help your domain clear the minimum bar to get into the pool of sources the machine draws from. Below that bar, you're not in the game. But once you're past it, piling on more backlinks doesn't do much, while each time an industry news site or a professional forum mentions your name, the lift is clearer.

This runs against how the market spends. Many businesses pay by the number of backlinks per month, measured in an Excel sheet, ignoring entirely whether anyone mentions them anywhere. An industry article that names your company, your field and your area correctly, even with no link, still teaches the machine that your company exists and what it does. That is something bulk-bought backlinks can't buy.

YouTube is the most overlooked place

The 0.737 figure for YouTube startled us a little the first time we read it, but on reflection it makes sense. A YouTube video has auto captions, a description and comments, and all of it is text a machine can read. A 15-minute talk by your technical director about exactly what your company does produces roughly 2,000 to 2,500 words of captions with context, plus the speaker's face and real name, which packs both Experience and Expertise into one piece of work.

We're not telling you to drop text SEO and make videos. The point is that if your budget is all going to one place, split it. For most service businesses, one 10 to 15 minute video a month about the craft, even shot on a phone, with the speaker's name and job title, already beats thirty hastily bought backlinks.

Four E-E-A-T signals machines can read, which you can do this month

The four items below are real technical signals, each checkable by eye in about 10 minutes. We've ordered them by what to do first, with cheap and fast work at the top.

Signal 1: The writer must be a real entity

The first E is Experience, and a machine only recognizes experience when it knows who went through it. A post under an "Admin" account has nobody behind it. This mistake is so common that when we take over an old website, the first thing we do is open the post list and count how many posts are under "Admin" or "Editorial team". The answer is usually all of them.

There are three parts to the fix. One, every article has a real author name with a job title and years in the trade, not a shared pen name from the marketing department. Two, every author has their own profile page on the website that says clearly what they've done, where and on which projects, and that page must declare Person schema with the property sameAs pointing to that person's real LinkedIn or Facebook. Three, the name and photo on the website must match the name and photo on those profiles. One character off and the machine treats them as two people.

The sameAs is what links a name to an entity in the machine's knowledge graph. We explain the entity mechanism separately in the article What is SEO entity.

Signal 2: Business information must match everywhere

The T, Trustworthiness, is measured by machines with a crude but effective test: they gather your business's name, address and phone number from every place they find it and check whether they match. A mismatch costs you trust.

The list of places that must match is usually longer than you think: the website footer, the contact page, the Organization schema, your Google Business Profile, your Facebook page, industry directories, e-invoices. The address is the most common mismatch, because each marketing push has someone writing it differently: one place says "Q.1", another says "Quận 1", another uses the new ward name after the merger. A machine reads those three spellings as three different locations. Going through and making them all match usually takes an afternoon, and it's the cheapest job in this entire article.

Alongside this group are some small things that carry real weight: an unexpired SSL certificate, privacy policy and terms pages that are actually written rather than copied from a template, a public tax code, and an About page with photos of the real team instead of stock photos. Machines spot stock photos instantly, because the same image sits on thousands of other websites.

Signal 3: Experience must come with visible evidence

This is what separates a practitioner's article from one written by someone who searched Google and rewrote it. Evidence of experience includes photos taken at your own workplace, before-and-after measurements, specific dates, software version names, click paths in order, and what real people said, kept in quotation marks.

A test we often use when reviewing an article before it goes live: remove every brand name from the piece and reread it from the top to see whether you can still tell who wrote it. If you can't, the article has no experience in it, only information. Machines have already swallowed billions of words of information. What they lack is what you did yesterday, at your place, with a specific number.

How you present it matters too. The answer should sit right at the top of a section rather than in the fourth paragraph, because machines cut paragraphs when they cite. We cover that in the article writing answer-first content, and the markup that machines read in the article what is Schema.

Signal 4: Mentions by third parties, even without a link

The A, Authoritativeness, is not something you declare on your own website. It's what other people say about you elsewhere, and machines gather those places.

Four mention sources worth pursuing, ordered by how easy they are. First, industry press and specialist sites: one interview or one expert quote from your company is enough. Second, forums and professional groups, where your staff answer real questions with real accounts. Third, profiles on industry rankings and directories, which many people skip because they think they are outdated, but machines still read them. Incidentally, the machine-readable markup work also includes file llms.txt, which takes about an hour; we file it under doing it for completeness, so don't expect much. Fourth is video, exactly as the YouTube data above shows.

What to avoid is buying fake reviews. Thirty five-star reviews in the same voice, bunched into three days, are obvious to people and machines alike, and they drag the T down faster than having no reviews at all.

Where Vietnamese businesses often break E-E-A-T without realizing it

We run into the six mistakes below again and again when taking over old websites. None needs a developer to fix.

• Posts under "Admin" or a department name. The E and the A are effectively left blank.
• An author page that exists but is hollow, with just a name and a line saying "content specialist". No years of experience, no projects, no links to outside profiles.
• Stock photos bought from overseas libraries for an article about your own factory. Meanwhile, that factory is open every day and a phone photo would give you a real image.
• An About page written in an anonymous voice, like "we are a reputable company with many years of experience", with no numbers and no names.
• A website with no specific address or just a district name. For content that affects money and health, a missing address costs a lot of points.
• Old posts left with 2023 figures and no update date, which as of August 2026 makes them three years stale. Machines read stale content, and so do readers.

Together, these six mistakes explain quite a few cases of well-written websites that AI still doesn't mention. The content is good, but the outer shell has no information, so the machine has nothing to hang its trust on. Fixing all six on a website of about 200 posts takes roughly two to three weeks, and most of it is labor rather than money.

E-E-A-T by industry: which evidence is worth the most

Every industry has one kind of evidence that pays off far more than the rest. Do that one first.

• Healthcare and clinics. The most valuable evidence is a complete doctor profile: degrees, practicing license number, where they trained, professional associations they belong to. Expert articles must name the reviewing doctor along with the review date.
• Education and training. Instructor profiles and public sample lessons. A 10-minute video of real teaching beats ten articles describing your method. The platform that goes with it: learner management software and AI-integrated LMS from Mona.Software.
• Restaurants and F&B. Photos of the real kitchen, the head chef's name, ingredient sourcing, a valid food hygiene and safety certificate. Manage it with restaurant management software.
• Spas, salons and aesthetics. Technician certificates, a detailed write-up of what a treatment session involves, before-and-after photos with dates and the customer's consent. See spa management software.
• Manufacturing and distribution. Valid quality certifications, production line videos, full technical specifications instead of generic praise. Go with warehouse management software.
• Travel and hotels. Room photos taken by the hotel itself, real reviews with the writer's name, travel license information. See travel website design.

If the website is slow or has no security certificate, everything above loses points. We handle that foundation in website design service and infrastructure Mona.Host.

What you can do this week

Four tasks, in order, about one working day in total, and no waiting on anyone to approve a budget.

Task one: open ChatGPT, type exactly the question your customers ask, and write down the names it mentions. That is your current scoreboard. Task two: go into your website, count how many posts are published under "Admin", and switch them to a real person's name with a job title. Task three: type your company name into Google and compare the address and phone number in the first five places you find; fix any that don't match. Task four: schedule a short video for this month, shot by a real practitioner in your company, talking about exactly what they do every day.

Do those four first, then think about spending money outside, and when you divide the budget, look at how to split your SEO budget sensibly. Measure again after three months using the metric set in the article measuring SEO as zero-click grows, and stop measuring by keyword ranking.