E-E-A-T as an editing checklist, not a score

E-E-A-T is usually discussed as something a site has. In practice it is more useful as something a draft either survives or does not. What follows is the checklist we run on our own pieces before publication — four questions, one per letter.

Experience: has the author been in the room?

The question is not whether the topic is covered. It is whether the writer saw the thing happen.

Signals that a piece has been written from the room:

• specific numbers with a time window, rather than round industry averages
• description of what went wrong, not only what worked
• details that only someone doing the work would notice — the two-week waiting period, the field that nobody fills in, the client objection that always comes up third

Signals that it has not: every paragraph could have been assembled from other articles, and nothing in the piece could be contradicted by a practitioner.

Expertise: is the claim scoped to what we actually know?

Expertise fails most often through overreach. A team that ran 11 projects in consumer electronics writes "how brands win overseas" — and the sentence is wrong for 90% of readers.

The fix is small and mechanical: attach the scope to the claim.

• Not "brands should localize pricing" but "for mid-priced consumer goods entering Southeast Asia, price localization mattered more than channel choice in 9 of our 11 projects"
• Not "AI search rewards structure" but "structured, claim-first paragraphs were quoted three times more often in our four-week comparison"

Scoped claims read as less confident and perform better.

Authoritativeness: who else would vouch for this?

This is the one teams cannot manufacture internally. What can be done:

• cite the primary source, not the article that cited it
• link out to the standard or the dataset, with a date
• name the methodology when there is one, including what it does not measure
• accept corrections publicly and timestamp revisions

A piece that links to its sources is easier for both reviewers and models to place. That placement is the practical form of authority.

Trust: what does this page ask the reader to believe?

Trust is mostly about the absence of a few specific things:

• outcome promises (rankings, returns, inclusion)
• unverifiable superlatives
• statistics without origin
• a commercial ask that appears before the information does

None of these kill a page on their own. Together they form a pattern that readers and classifiers both pick up quickly.

Running it

Print the four questions and read the draft once per question. Four passes, not one. The first pass catches missing specifics, the second catches overreach, the third catches missing sources, the fourth catches the sales language that crept in during editing.

It takes about as long as one careful read. It catches most of what later costs a rewrite.