How Does EEAT Affect Your Visibility in AI Search Results?
Four factors, each treated accurately about how much is actually known. Every competitor has the same list. What almost nobody publishes is which parts are documented, which are observed and which are simply assumed, which is what block eight is for.
What EEAT Is And Is Not
EEAT stands for experience, expertise, authoritativeness and trustworthiness. It comes from the guidance Google publishes for the people who assess search quality. It is a framework for human judgement rather than a measurement inside any system.
It is not a score. The belief that matters most to correct.
There is no EEAT number attached to your site, no dial to move and nothing to check. Any tool presenting an EEAT score is presenting its own estimate of a concept that was never numeric. That is not necessarily useless as a prompt for work. It is not a measurement of anything.
It was written about search quality rating. Not about assistants.
The guidance exists so that human raters can assess whether results are useful. Applying the same framework to how an assistant selects sources is an extension. It is a reasonable extension, made by a great many practitioners including us. No vendor has confirmed it.
Why we say that before using it anyway. Because the alternative misleads you.
Almost every page on this subject applies the framework to AI search without mentioning that the connection is unconfirmed. We use it because block two gives a reason to, not because anybody has published that it applies.
What this page covers, plus what it does not. A firm boundary.
This page is about what a business can influence. How an answer actually gets assembled sits in our guide to that mechanism. What individual vendors have published sits in the four platform guides. Three different jobs, kept apart deliberately.
This page is reviewed quarterly. The reasoning holds longer than the platform positions do.
Why It Is A Useful Frame Anyway
An unconfirmed framework can still be the right one to work from, provided you know why. The argument here is about what is observable rather than about what Google intended.
The problem any assembling system has. It cannot verify anything.
Something composing an answer from several sources has no way to establish which is true. It cannot ring anybody, inspect a qualification or visit a premises. It can only work from what is observable about the sources themselves.
What is observable about a source. A short list.
Whether other independent sources say the same thing. Whether the material is specific enough to be checkable. Whether a named person with stated credentials stands behind it. Whether the information about the entity is consistent wherever it appears.
Why that list resembles the framework. Not coincidence. Not proof either.
Those four are approximately what experience, expertise, authoritativeness and trustworthiness were describing. The framework has survived because it names the things a system can actually see, which is a decent reason to work from it whether or not any vendor has adopted it.
What that licenses and what it does not. Worth being precise.
It licenses treating those four as sensible priorities. It does not license claiming they are ranking factors, weighting them against each other or telling a client that improving them produces AI visibility. Block eight sets out exactly how far each one can honourably be pushed.
Brand Authority
Being known, named and referenced across sources you do not control. This is the factor with the most practitioner agreement behind it and it is still observation rather than documented mechanism.
What it actually means here. Not brand in the marketing sense.
Not a logo, a tone of voice or brand awareness. It means the business existing as a recognisable thing described by multiple independent parties, so that an assembling system encounters you from more than one direction.
Why corroboration appears to matter more than it did. The structural argument.
Ordering ten links required no reconciliation, because a person did that themselves. Composing one answer from several sources does require it. Where sources agree, the material is simple to use. Where they conflict, something has to choose, then nothing about that choice is published.
What this looks like in practice. Unglamorous.
Being listed in the places your trade is listed. Being mentioned in local or trade press. Being described the same way by a professional body, a directory and your own site. None of that is a campaign.
The measurable proxy worth watching. Searches for your own name.
Branded search volume rising over time is the closest available indicator that a business is becoming a recognised entity. It is a proxy rather than a measurement of this factor. It is one you can actually see.
What we will not claim. That authority is a lever with a known effect.
No vendor has published that being referenced more often causes inclusion in answers. The reasoning in block two is why we act on it regardless.
Structured Data
Machine readable description of what a page is about and what a business is. This is the factor most often oversold in this subject, so it needs the most careful treatment on the page.
What it does. Removes ambiguity.
It states plainly what a page describes rather than leaving something to infer it from prose. A business, an address, a service, an author, a date. Nothing has to be interpreted.
What no vendor has confirmed. That it is used in generating answers.
Not one of the companies operating these products has published that structured data influences whether a source is drawn on. Anybody telling you it does has gone beyond the available evidence. This is the specific overclaim we see most often.
What Google does say about it. Documented, though narrower than people report.
Google's Search Central documentation on AI features, checked on 30 July 2026, names making your structured data match the visible text on the page among the continuing fundamentals worth attending to. That is a statement about accuracy rather than about generation. The difference matters.
The standing rule that applies regardless. Non negotiable in our work.
Structured data is an assertion about reality. It must never describe something untrue. Marking up reviews you do not have, a price you do not charge or an author who did not write the page is a misrepresentation whatever it does or does not achieve. That rule does not depend on any vendor's behaviour.
How we treat it. As hygiene rather than as a lever.
We implement it because it removes ambiguity, because it demonstrably supports ordinary search features and because accuracy is cheap. Not because it has a published effect here. Our technical guides cover the implementation.
Content Depth
Covering a subject completely rather than partially. The word depth misleads people into producing length, which is not the same thing and frequently works against them.
What depth is not. Word count.
A three thousand word page circling a subject is shallower than an eight hundred word page that answers one question and leaves nothing outstanding. Length is what you get when depth is pursued by somebody paid per word.
What depth actually is. Nothing left to ask.
A reader finishing the page has the question settled, including the awkward parts: what it costs, what it does not cover, when it is the wrong choice, what happens if something goes wrong. Most business content stops before the awkward parts, which is precisely where the value sits.
Why one question beats ten. The usability argument.
A page answering one question fully can be used as the source for that question. A page mentioning ten is the best answer to none of them, so it competes for each and wins nothing. This is observed consistently across every product we watch.
The structural requirement that comes with it. Passages that stand alone.
Depth is worth nothing if the answer only makes sense in the context of the whole page. Each passage should be liftable, which our guide to AI Overviews treats as the single most practical thing on the subject.
What this factor is worth if the field changes. Everything it cost.
Pages that fully answer questions were valuable before any of this and will be afterwards. That makes depth the safest of the four to invest in.
Citations And Being Referenced
Other people describing your business, which functions as corroboration rather than as a vote. That distinction changes what is worth doing about it.
Why references work differently here. Agreement rather than endorsement.
In ordinary search a link has long been treated partly as an endorsement. When an answer is assembled from several sources, an independent description doing the same job as yours is providing confirmation. What matters is that somebody unconnected says the same thing.
What that implies about quality. Relevance beats volume even more than usual.
One reference from a source that genuinely knows your trade is worth more as corroboration than twenty from places that describe everybody. Volume was always a weak strategy. As corroboration it is close to worthless, since twenty identical directory entries confirm nothing an assembling system did not already have.
The position we carry across from our backlinks work. Unchanged by any of this.
This is a reason to earn references rather than a reason to buy links. Bought links were a poor investment before and nothing about this subject improves them. Our backlinks guides hold the same line they always have, including on when link building is the wrong spend entirely.
What counts that people overlook. Unlinked mentions.
A business named in an article without a link still contributes a description from an independent source. For corroboration purposes that may do the same work, which makes press and trade coverage more valuable here than a link focused view suggests.
What we will not claim. A quantity, a threshold or a rate.
Author And Entity Signals
Named people with real credentials, consistent information about the business. The connection between the two. This sits underneath the other four factors rather than beside them.
Named authors rather than anonymous pages. A real person, verifiably.
A page attributed to somebody with stated qualifications, a professional registration where one exists and a presence beyond your own website is a different kind of source from an unattributed page. Inventing an author is worse than having none, since it is a false claim on the page.
Consistent entity information. The business described the same way everywhere.
Name, address, services, description, across the site, the profile and third party listings. Contradictions give an assembling system conflicting material about who you are.
The connection between them. The part usually missed.
The named expert should be connected to the named business in a way that is visible from outside. A person whose credentials appear only on your own site, with no independent trace, provides considerably weaker corroboration than one whose registration can be found elsewhere.
Where this matters most. Regulated and expertise led sectors.
Legal, financial, medical, care and anything where being wrong has consequences. In those categories the identity and credentials of whoever stands behind the material are doing real work. Our guides for those trades treat it accordingly.
Where it matters least. Urgent and transactional work.
Nobody choosing an emergency plumber at midnight is weighing author credentials, so a business entirely in that market should spend the effort elsewhere.
How Much Of This Is Confirmed
Every competitor has the list above. Almost none publishes this block, which is the calibration. Each factor sorted into what is documented, what is observed and what is assumed.
EEAT applying to AI search at all. Assumed.
No vendor has confirmed the framework governs source selection. It comes from search quality rating guidance and its application here is an extension practitioners have made, ourselves included.
Brand authority and corroboration. Observed, with reasoning behind it.
Practitioners report consistently that businesses described by several independent sources appear more often. Nobody has published it as mechanism. Block two explains why the reasoning is sound anyway.
Structured data. Documented as accuracy guidance, assumed as an influence.
Google names matching structured data to visible text among its continuing fundamentals, per its documentation checked on 30 July 2026. That it affects whether a source is used in generation is confirmed by nobody. It is the overclaim we encounter most.
Content depth. Observed, plus independently justified.
Pages answering one question fully appear more often across every product we watch. It is also valuable regardless of any of this, which makes it the least risky item here.
Citations and references. Observed, with the strongest practitioner agreement.
Still not documented. No threshold or quantity has ever been published by anybody.
Author and entity signals. Mixed.
Entity consistency is long established for ordinary local search. The author half is observed at best. It is strongest in regulated sectors, where it also serves a professional obligation.
What the pattern tells you. One thing is documented and it is about accuracy.
Everything else is observation or inference. Any page presenting these six as established ranking factors has told you something nobody knows.
What To Do With That
Calibration is only useful if it changes what you do. It does, in a specific way: it tells you which of these to build on and which to hold loosely.
Do the things that help regardless. Three of them.
Content depth, entity consistency and earned references. Each is justified by ordinary search independently, so none is wasted if the field moves. This is most of the work and it should be most of the budget.
Do the accuracy work because it is right. Structured data matching reality.
Not because it has a published effect here. Because it is cheap, it supports ordinary search features and misrepresentation is not something we do at any price.
Hold the rest loosely. And write down that you are.
The framework itself. The weighting between factors, are assumptions. Recording them as assumptions means they can be revised when something is published, which our strategy guide treats as the difference between a plan and a belief.
Refuse the weighted list. Whoever offers it.
If a supplier gives you these factors with percentages attached, they have invented the percentages. That is the single most useful thing to take from this page when reading somebody else's proposal.
And judge us the same way. The test is symmetrical.
Everything above is labelled. If a competitor's version of this page is not, that tells you what their recommendations rest on.
Everybody has the list.
Almost nobody calibrates it.
Of the six factors on this page, one is documented and it concerns accuracy rather than generation. Everything else is observation or inference. A supplier attaching percentages to them has invented the percentages.
What is included every month:
£350 per month, one target area. No setup fee, nothing billed separately.
Twenty guides.
One subject.
The mechanism, all four platforms, what you can influence, whether it is worth it yet, what it costs, how to measure it and how to judge a supplier selling it.
