TL;DR:
E-E-A-T (Experience, Expertise, Authoritativeness, Trust) isn't a separate checklist for AI Overviews, ChatGPT or Perplexity. It's the same trust framework Google has used for years, except AI search removes the ten blue links that used to give a weak entity somewhere to hide.
- E-E-A-T was never a ranking factor you could tick off, it's a proxy for trust that Google's Quality Raters have been trained to spot since long before "AI search" was a phrase anyone used.
- Generative answer engines inherit that same proxy, then add a new problem on top: they have to resolve who you are before they'll cite you at all.
- Get that identity resolution wrong and no amount of good content saves you.
This is just one of the core factors our brand audit consultation is designed around. If you think you have an issue with visibility, book an audit — we will show you where the problems are.
Below, I’ll explain the theory behind this (from both sides), using Google's own leaked documentation, Shaun Anderson's research, and what fifteen years in a restricted vertical has taught me about being invisible to a machine that can't quite work out who you are.
What Most People Think E-E-A-T Is (And Why That's Wrong)
Most site owners still treat E-E-A-T as a content quality checklist. Add an author bio, stick a doctor's name on the medical page, job done. But this isn’t how it works at all and it is one of the key things nearly all vape and cannabis brands we work with get completely wrong.
E-E-A-T is a framework Google's human Quality Raters use to assess pages, described in detail in the Search Quality Rater Guidelines, but it was never a direct ranking signal you could optimise for line by line.
Think of it more like basic brand hygiene that all sites should have in place as standard. I have a full podcast episode on this too, if you prefer that medium.
Google's own documentation states plainly that of the four components, trust is the most important, and that untrustworthy pages have low E-E-A-T no matter how experienced, expert or authoritative they otherwise appear.
Experience, Expertise and Authoritativeness all exist to support Trust.
They're not four equal boxes to tick. Trust is the roof, the other three are the walls holding it up.
This matters enormously once you're talking about AI search, because generative answer engines have exactly the same problem Google has always had: how do you mechanically detect something as human as credibility?
The answer, in both cases, is the same. You build a proxy, train it on what real people say counts as trustworthy, and let a machine learning system generalise from there.
Why "E-E-A-T Isn't A Ranking Factor" Is A Half-Truth
You'll see this line repeated everywhere: "E-E-A-T isn't a ranking factor, it's just guidance for raters." This is technically correct but there’s a good chance it is also misleading.
- What people believe: because Google says E-E-A-T isn't a direct ranking factor, it's fine to ignore it and focus purely on technical SEO and links.
- Why they're wrong: the Quality Rater Guidelines exist to train human raters, and rater feedback is used to evaluate and refine Google's machine learning ranking systems. Shaun Anderson's research into the 2024 Content Warehouse API leak connects this directly to a plausible scoring system he refers to as Q-Star, which culminates in a Site_Quality score. As Anderson's analysis puts it, E-E-A-T is best understood as the human-legible description of what that quality score is trying to measure, not a separate box you fill in.
Google doesn't need an "E-E-A-T score" as a discrete signal because dozens of smaller signals collectively approximate it.
Google's 2024 Content Warehouse API leak revealed attributes such as siteAuthority, pandaDemotion and unauthoritativeScore, alongside contentEffort, described in the documentation as an LLM-based effort estimation for article pages.
None of these are labelled "E-E-A-T" in the leaked code because they're the machinery underneath the concept.
- Why Google behaves this way: a framework based on keyword density or backlink counts is trivial to game at scale. A framework based on hundreds of interlocking signals, calibrated against tens of thousands of human ratings, is far harder to reverse-engineer and manipulate. That difficulty is the point.
- Real world example: I've audited vape and cannabis sites with genuinely well-written, technically clean content that still couldn't hold rankings through a core update. The content passed every on-page checklist. What was missing was any verifiable connection between the site and a real, accountable business. That gap is invisible if you're only checking word count and meta tags.
Stop asking "have I optimised for E-E-A-T" and start asking "would a sceptical stranger trust this page enough to act on it." That's the actual test, human or algorithmic.
The Disconnected Entity Problem: Why AI Search Punishes Ambiguity Harder Than Google Ever Did
This is where AI search changes the maths, and it's worth slowing down here because it's the part most SEO commentary skips entirely.
Section 2.5.2 of Google's Quality Rater Guidelines instructs raters to establish who owns and operates a website, and who is responsible for its content.
Shaun Anderson's research labels the failure state here the Disconnected Entity Hypothesis: sites that lack a strong, verifiable connection to a trusted real-world entity get treated as fundamentally untrustworthy, regardless of how good the writing is.
As Anderson frames it, and I’m in agreement with his thinking here because it correlates with my own, lived experience over the past decade or so, a lack of trust is one of the biggest single levers Google has for suppressing rankings, and a disconnected entity fails that check before content quality even enters the conversation.
Classic Google search could still surface a disconnected entity occasionally, buried on page two, generating a trickle of traffic nobody noticed.
With LLMs or even Google AI Overview, this is a much bigger problem:
- AI search doesn't have a page two.
- An answer engine either resolves your brand into a coherent entity it can cite with confidence, or it doesn't mention you at all.
- There's no consolation prize for ambiguous identity in a system built to produce one confident answer.
The Shy Brand Problem
Working in vape and cannabis SEO for fifteen years, I've watched a very specific version of this play out so often I ended up giving it a name: The Shy Brand Problem.
Out of all the sites I’ve audited and worked with, this “shy brand” problem is one of the most common issues I come across. Does your business have a Shy Brand problem?
Here’s the criteria:
- does your business use inconsistent product terminology across your site (nicotine salts here, nic salts there, salt nic somewhere else, no glossary connecting any of it),
- does it operate under three different trading names across the domain, ecommerce listings and payment processor,
- and do you bury compliance and ownership information because restricted categories have historically preferred to stay low profile.
That instinct to stay quiet, born from years of ad bans and platform bans, is precisely what breaks entity resolution for an AI system.
You can't build a coherent, citable entity while simultaneously trying to hide behind a brand name. The quickest way to more trust is to be people-first, and that means having actual people (owners, staff, writers, whatever) visibly front and center for all to see.
That’s why I have that stupid image of myself as the featured image on this article. Sure, it ain’t pretty. But it’s me, and people tend to respond better when they know they’re dealing with an actual person, and that rule works overtime when you move the setting online.
I see this constantly in vape SEO audits, and it's one of the first things worth fixing before touching a single blog post.
How AI Overviews Actually Pull From E-E-A-T Signals
Here's an analogy that's helped a lot of clients grasp this. Think of entity SEO as reputation, in the plain, everyday sense. You don't build a good reputation by shouting about yourself once. You build it through consistent, verifiable behaviour that other people independently vouch for, over years.
AI Overviews and answer engines are trying to shortcut exactly that process by scanning what already exists about you and forming a synthesised judgement.
This is where the industry's obsession with links alone starts to look dated. Anderson's research on what he calls the Mentions Economy argues that in an AI-summarised world, mentions, not just backlinks, function as a trust currency.
An AI system building a picture of your brand isn't only counting who linked to you, it's absorbing what's said about you across review sites, forums, industry publications and your own pages, then forming an inference.
Anderson also documents a genuine risk here, which he calls Mention Pollution: coordinated or low-quality mentions designed to game that same signal, which is precisely the kind of manipulation that will eventually attract the same enforcement links and paid placements did.
This is a problem Reddit is having to deal with in real-time right now. Tens of thousands of marketers are flooding the platform with obvious spam with the simple goal of getting “mentions” for their brands or their client’s brands. It’s called astroturfing and, to me, it feels spiritually closer to black hat SEO than anything else in the GEO optimization space right now.
It’s also incredibly easy to do which, in my experience, means it will soon be fragged and bagged by a core update. This is ALWAYS what happens when a hack gets out of control and starts messing with Google’s SERPs.
Trust Debt: What I Watch For In A Brand Audit
I use a working concept internally that I call Trust Debt; it sounds better than my previously mentioned Shy Brand Thesis. It's the gap between how a business actually operates, its real compliance record, real ownership, real customer experience, and how confusingly or incompletely that reality is represented online.
Every unexplained inconsistency, every unclear ownership structure, every unlinked "About" page, adds to that debt.
Eventually an algorithm, human rater or AI system calls it in, usually during a core update or when an answer engine simply declines to cite you.
- Trust Debt compounds the same way SEO itself does.
- Compound interest works because small consistent inputs multiply over time.
- Trust Debt is the same mechanism running in reverse: small unresolved ambiguities accumulate, quietly, until the correction arrives all at once and looks sudden even though it wasn't.
If you're trying to work out how much of this debt your own brand is carrying, this is precisely what a proper entity and trust audit is designed to surface, the disconnected signals, inconsistent entity data and trust gaps that neither a content checklist nor a technical crawl will catch on their own.
What This Looks Like In A Restricted Market
Publishing in a category where major ad platforms restrict you, where payment processors are cautious, and where mainstream press rarely covers you fairly, forces you to build authority the slow way.
No paid amplification safety net. You either become a source other people in the industry genuinely trust and cite, or you don't rank, and later, you don't get mentioned by AI systems either.
The vape and cannabis brands I've worked with that recover fastest after a core update, or that actually start appearing in AI-generated answers, are rarely the ones who rewrote their blog.
Rather, they're the ones who got proactive about the signals they’re sending to potential customers AND Google’s crawlers. They fixed the boring, unglamorous entity signals first: consistent NAP information, a real About page, clear authorship, terminology that matches across every page on the domain.
Sure, it's not exciting work and most optimization gurus seldom mention it but it is one of the main reasons why this blog — I literally only started it in May 2026 — is now cited by AI Overviews, did not experience any form of Google “sandbox”, and is already picking up key rankings for keywords and topics associated with my core business.
Where This Argument Gets Pushback
I'd be doing you a disservice if I pretended everything above was settled consensus because, as any critical thinker knows, consensus is for the weak-minded; there’s always outliers, counter-arguments and factors that blur the lines between dogma and official rules and guidelines.
Rather than doing the usual, bang-em-in-the-footnotes, I want to instead give each counter point to what I’ve discussed its own space to make its point and breath.
From Google Itself…
Danny Sullivan, Google's Search Liaison, has said directly and repeatedly that E-E-A-T is not a ranking factor and does not feed into other ranking factors either, explicitly rejecting the idea that hiring a credentialed writer produces any automatic ranking benefit.
That's a firmer denial than the "indirect proxy" reading I've leaned on above, and it deserves to sit there unresolved rather than be explained away.
From Rand Fishkin…
There's a similar caution around the Content Warehouse leak itself.
Rand Fishkin, who first published the leaked documentation through SparkToro, was clear from the outset that the documents show what data Google collects and stores, not the weighting any attribute carries, and not confirmation that a given attribute is actually used in live rankings.
Attributes like siteAuthority and contentEffort existing in Google's codebase is real evidence Google tracks these concepts internally but there is no concrete proof they actually determine outcomes the way Shaun Anderson's Q-Star hypothesis, or my own Trust Debt framing, assumes they do.
From Ahrefs & BrightEdge
Then there's the AI Overviews data itself, and here I'll be honest about the quality of what's circulating. A cottage industry of SEO tools has sprung up publishing suspiciously precise statistics, figures like 96% of citations showing strong E-E-A-T signals or a 0.81 correlation between E-E-A-T and AI citation, with no visible methodology behind either number.
I'm not repeating figures like that as fact here, and you should be wary of anyone who does.
What independent analysis from Ahrefs does suggest is that only somewhere between roughly 17% and 38% of AI Overview citations come from pages already ranking in the traditional top ten, down sharply from a year earlier.
That's a genuinely awkward fact for anyone arguing AI search simply extends classic ranking authority wholesale.
It points to AI Overviews selecting on something closer to passage-level relevance and freshness than on accumulated site-wide trust, at least for a meaningful share of queries.
My honest read: the link between E-E-A-T and AI citation is real but looser than either the confident "it's all E-E-A-T" camp or Google's flat denial admits.
Again, the winners here will be the pragmatists (not the dogmatists). Those who think critically about how they position their brand and how they communicate who they are to algorithms and potential customers, using the information they have available to them which, in this context, would be:
- the official the Google guidelines (now updated for AI search visibility),
- the Content Warehouse Leak (regardless of its weighting in the actual algorithm),
- and common sense when it comes to trust signals.
To exist in search and be visible online, either in traditional SERPs or LLM search, you need an entity, and the more defined and tangible your brand entity is, the better.
On top of this, the brand entity signals you build today and over the course of the coming months are the same signals that will protect when the next major Google algorithm update lands.
Practical Takeaways
Building an entity an AI system can confidently cite isn't a content task, it's closer to editorial and organisational discipline applied consistently over time.
- Audit whether a stranger, human or machine, could work out who legally operates your site within thirty seconds of landing on it.
- Standardise terminology across your entire domain before you write another article. Inconsistent product language is an entity resolution problem disguised as a copywriting one.
- Treat mentions with the same seriousness you'd treat links, and resist the temptation to buy your way into either.
- Don't wait for a core update to surface your Trust Debt. Address it proactively, ideally as part of a proper SEO framework or roadmap rather than a scattered list of fixes.
- If you're building or scaling an in-house function to manage all of this long term, this is exactly the kind of ongoing strategic ownership a fractional SEO lead role is built to provide.
Quick-Fire Questions & Answers
Does Google confirm E-E-A-T is used by AI Overviews directly?
Google hasn't published a technical breakdown confirming E-E-A-T signals feed AI Overviews specifically. What's documented is that E-E-A-T describes qualities Google's ranking systems, and its human raters, are trained to identify and reward, and AI Overviews sit on top of Google's existing search infrastructure rather than replacing it.
Is E-E-A-T more important for YMYL topics in AI search?
Google's Quality Rater Guidelines instruct raters to assign pages on Your Money or Your Life topics the lowest possible quality rating if E-E-A-T is low, regardless of other strengths. There's no reason to expect that scrutiny softens when the same content ends up summarised by an answer engine.
Can a small, restricted market business realistically build E-E-A-T?
Yes, and arguably it matters more, not less, in restricted categories where scrutiny is already higher. Consistent entity signals, clear ownership information and genuine, verifiable experience matter more than site size.
Do backlinks still matter if mentions are becoming important too?
Backlinks haven't stopped mattering. Authoritativeness has always included external validation, and links remain one form of that. Mentions are best understood as an additional, overlapping signal, not a replacement.
Hasn't Google said E-E-A-T isn't a ranking factor at all?
Yes. Google's Search Liaison Danny Sullivan has said directly that E-E-A-T is not a ranking factor and does not feed into other ranking factors, a firmer denial than the "indirect proxy" framing this piece leans on. It's a genuine point of tension worth knowing about rather than one worth dismissing.
What's the fastest way to check my own Trust Debt?
Start by looking at your own site as a stranger would: is ownership clear, is terminology consistent, does everything about who you are line up across every page. A structured brand audit will go considerably deeper than a self-check, but the self-check is the right first move.