E-E-A-T for AI: What Actually Counts

Somewhere in your inbox is a pitch for an "E-E-A-T audit". It comes with a score out of 100 and a promise that raising it will get you cited by ChatGPT. There is no such score. Google says E-E-A-T is not a specific ranking factor, its own rater guidelines say ratings cannot position a page, and no AI engine documents using anything like it.

The ideas behind the acronym are not useless. A named author a reader can look up, a claim with its source attached and evidence that you have done what you write about are what make a page worth trusting. They also make your facts checkable, which is the point of this module.

So do them, for those reasons. Stop paying anyone to move a number that does not exist.

What E-E-A-T is, according to the people who wrote it

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness. It comes from Google's Search Quality Rater Guidelines, the handbook Google gives the people it pays to evaluate sample search results. Their ratings help Google judge whether its systems are working. They do not rank pages.

Google describes E-E-A-T as a concept its systems try to approximate through other signals, and it puts the most weight on trust for topics that can affect someone's health, money or safety. Invoicing software touches money, so trust matters for a brand like Ledgerly. That still does not create a number anyone can measure from outside.

Two honest caveats. First, these are Google's statements, not measurements. Google has reasons to make them, and nobody outside can verify what its systems do. Second, the statements cover Google only. OpenAI's entire public description of how ChatGPT ranks search results is that it uses "multiple factors intended to help users find relevant, reliable information. Placement is not guaranteed." Perplexity does not publish its ranking factors either. Nothing in the published GEO research has tested E-E-A-T on any AI engine. Every vendor that reports an E-E-A-T score made up the scale.

Authority is per topic, not per site

The rater guidelines define Authoritativeness by topic: "Consider the extent to which the content creator or the website is known as a go-to source for the topic. While most topics do not have one official, Authoritative website or content creator…" A tax firm can be a go-to source on tax and a poor source on running shoes. In Google's own vocabulary, authority belongs to a subject, not to a domain.

That has a practical consequence. There is one topic on which you are the authority by definition: your own company. You are the go-to source on your prices, your features, your founding date and who runs you. Nobody else can be. The most useful authority work you can do is make your own site the most accurate, most consistent source on those facts, which is what How to Create Content AI Can Quote sets up.

Well-known brands do get cited more often. That association is real and strong, and Building Off-Site Corroboration explains why it is still not a lever you can pull.

Before you pay for any service sold as "authority building" for AI, ask the seller three questions:

1. What exactly does your number measure? If the answer is a blend of backlinks, word counts and author bios, it measures their formula, not anything an engine computes.
2. Which engine documents using it? Ask for the page. Google says E-E-A-T is not a ranking factor, and no AI engine publishes an authority score.
3. Where is the test with a control group? A before-and-after chart for one client, with no comparison against similar pages that did nothing, shows only that time passed.

A seller who cannot answer all three is selling the acronym, not a result.

What still counts, and why

Three practices survive scrutiny. None of them has a measured effect on AI citation. Each is worth doing because readers check it and because it makes your facts consistent and verifiable.

Named, verifiable authors

A byline should name a real person, give their actual role and link to a profile that exists outside your site: a LinkedIn page, a conference talk, a published paper, a GitHub account. Verifiable means a reader can confirm the person exists and holds the role you claim.

Do not invent authors. An AI-generated persona with a stock headshot is a false statement about who works at your company, the same class of problem as a wrong price. If your CEO writes a post, the title in the byline should match the leadership facts on your About page. 99Visibility's accuracy check compares what engines say about your CEO and leadership against the facts you enter, and your own author pages should never be the source that disagrees.

Sourced claims

Every statistic should name its source and date in the same sentence, with a link. That is the rule this course follows. An unsourced number on your blog is a claim nobody can verify, attached to your brand, that an engine may repeat as fact. If you cannot find the source for a number, delete the number.

Note the difference from the risk flags in the first lesson of this module. A statistic your argument needs, with a source a reader can follow, is a sourced claim. Statistics sprinkled in to look authoritative are injected statistics, and testing on that tactic came back null to negative.

Real first-hand evidence

Experience is the easiest of the four letters to show and the hardest to fake well. Screenshots of your own product doing what the page describes. Your own data, with the method stated and its limits admitted. Customer case studies with the customer's name, their permission and numbers they agreed to publish. A page that says "we tested this" and shows the test is worth more to a reader than a page of general advice.

Never fabricate any of it. A case study with invented results is a liability waiting to be found, and once an engine repeats it, it becomes a wrong fact about your company that you wrote yourself.

Transparent company information

Trust starts with the basics a reader looks for: who you are, where you are, who runs the company and how to reach it. Your homepage should link to an About page and a Contact page, and those pages should state those facts in visible text. 99Visibility's audit checks whether your homepage links to an About page and a Contact page, and it files a gap as hygiene, never as a ranking lever. When those basics are missing from your own site, an engine that searches has only third-party pages to work from, which is where the next lesson picks up.

Audit three pages

Pick the three pages buyers read most, not the three you are proudest of. Usually that is one high-traffic guide or blog post, the pricing page and the About page. For each, answer five questions:

1. Is there a named author or owner, and does the name link to a profile that exists outside your site?
2. Does every number name its source and date in the same sentence?
3. Is there first-hand evidence: a screenshot, your own data or a named customer?
4. Do the company facts on the page match your canonical facts block?
5. Is there any claim on the page you could not prove if a customer asked?

Fix the worst page first. One fixed page is worth more than a spreadsheet of findings nobody acts on.

Common mistakes

• Buying a score. Any audit that outputs an E-E-A-T number is measuring something it invented.
• Inventing authors. A fake persona is a false fact about your company.
• Bylines that lead nowhere. A name with no role and no outside profile cannot be verified, which defeats the purpose of a byline.
• Inflated credentials. A title or qualification that does not survive a quick check is worse than no title at all.
• Sources that do not say what you claim. Read the source you link. A citation that does not support the sentence is an unsourced claim with a link attached.
• Adding author markup instead of a byline. Person schema for an author nobody can see on the page is a parity failure, covered in Schema Markup for AI.
• Mistaking Google's statements for proof. They are assertions. They are also the only first-party evidence available, and nobody has published evidence against them.