Corroboration: Google Does Not Believe You, and It Is Right Not To

A bank manager approves credit for people he has never met.

He does not know whether you are honest, and his work does not stop over it. What sits on his screen is something else. Bills paid on time. Other institutions that already bet on you. Your name showing up, or not showing up, on the bad lists. He reads the trail you left in places that do not belong to you, builds a number, and decides. ¹

Google solves the same problem, for the same reason. It has to choose whose answer to accept, billions of times a day, without knowing anyone. Trust, in a system that size, was never an impression. It was always a scoreboard built from indirect evidence. ¹

This is called corroboration: what other people confirm about you, in places you do not control.

The score the market buys and that does not exist

You have heard that you need to “have E-E-A-T”. The public Google documentation says, literally: “While E-E-A-T itself isn’t a specific ranking factor, using a mix of factors that can identify content with good E-E-A-T is useful”. ²

Read the second half again. The company chases something it does not measure directly, using measures it can calculate. And the internal documentation leaked in 2024 agrees by its silence: thousands of attributes are documented, and none of them is an E-E-A-T score. Nobody stores what nobody calculates. ³

The same official page carries the hierarchy that almost nobody quotes: “Of these aspects, trust is most important”. Of the four letters, trust is the one in charge. ²

So the useful question is never “how do I raise my E-E-A-T”. It is: which calculable measure moves when I do this. And the calculable measures of trust are, for the most part, corroboration.

Which mention counts, and which only looks like it counts

This is where the money gets thrown away.

A syndicated press release is not proof. The same text, on dozens of sites in one week, with no byline, is your own voice on another phone. The machine reads the replication, and replication is not an independent source.

A copied mention is not proof. The book records one case. A technical sentence about a brand existed in exactly two places on the internet: the blog of a third party and the site of the brand itself. Someone copied it from someone. A mention like that has the look of reinforcement and the function of noise. ⁴

What counts is reporting done by someone else. A journalist who picked the angle. An association that lists you by its own criteria. A client who describes the work in their own words. A good mention is the one that repeats your facts, correctly, in someone else’s house. ⁴

And all of that needs a destination: the sources point at your entity home, or the confirmation arrives and finds no house.

How it gets built, in practice

Your own data generates a real story. Publish a number only you have: from your operation, your sector, your client base. That is the kind of material that makes a journalist report and sign.

Correct listings before new listings. Fixing one wrong listing that already exists pays more than opening three new ones, because a divergent listing works against you. ⁵

One spelling, everywhere. Two spellings of the name are two half entities.

And declared patience. Corroboration is the slowest work in the method, and it is the only one competitors cannot copy in an afternoon.

One point of transparency, because the house rule applies to the house: SEO Genome sells corroboration work inside its consulting practice. This article exists so you understand the mechanism and decide, with whoever you want to run it.

Frequently asked questions

Does buying an article on a news portal work?

Paid text with no reporting is your own voice published somewhere else. It may carry media value. As proof of independence it is worth little, and proof is what the machine looks at.

How many mentions do I need?

Fewer than you imagine, and better than the ones you have. A handful of real sources confirming the same facts beat dozens of replicas.

Does this apply to the AI systems too?

It applies double. A generated answer cites whoever the model treats as a source, and the road to becoming a source runs through external confirmation. The AI search silo shows how to measure it.

Notes and sources

1. Daniel Sócrates, “AI Search Optimization” (“SEO de Entidade”, Portuguese edition), 2026, chapter 7, “Confiança não é opinião. É cálculo” (Trust Is Not an Opinion. It Is a Calculation). The image of the bank manager and the formulation of the scoreboard of indirect evidence are the book’s.
2. Google, public documentation “Creating helpful, reliable, people-first content”, cited in chapter 7 of the book. Both literal quotations are reproduced in the body.
3. Chapter 7 of the book, on the Content Warehouse API documentation exposed in 2024: none of the documented attributes is an E-E-A-T score.
4. Same work, chapter 12: the case of the technical sentence existing in only two places, and the rule that a good mention is the one that repeats your facts, correctly, in someone else’s house.
5. Same work, chapter 12, on active noise: a listing with divergent data works against the entity.