**TL;DR** — Across 22 client sites in April 2026 we tested whether visible author bylines, author pages and explicit E-E-A-T signals change Google AI Overviews citation likelihood. The short answer: yes, but the lever is narrower than the SEO discourse usually claims. Three signals correlated with measurably higher citation share — a real author page reachable in two clicks from any cited article, a sameAs graph that resolves rather than 404s, and a byline that names a single human (not a brand). Three signals widely promoted as E-E-A-T did not move the needle in our basket: photo cards, certification badges and credential strings beneath the byline. Treat the entity, not the decoration.
Why we re-ran the E-E-A-T audit
E-E-A-T became the dominant SEO content meme in 2023 and 2024. By 2026 the discourse has split — half the field still treats it as a magic formula, the other half has quietly given up because nothing tests cleanly. Our first attempt to measure it in 2024 was inconclusive; the signals were too noisy on the old organic SERP and the controlled experiments kept regressing to the mean. AI Overviews changed that. The composer either cites a URL or it does not, and over thirty days you get enough binary outcomes to attribute movement to specific signals if you keep the test honest.
We had a second motivation: clients keep being sold on byline overhauls — photo cards, "verified by" badges, three-paragraph credentials below every article — and we could not tell them with a straight face whether any of it moved citations. The audit was the only way to answer the procurement-meeting version of the question, which is "if I budget for one E-E-A-T thing this quarter, which one?"
How we ran the audit
22 client sites — 8 SaaS, 6 publisher, 5 DTC and 3 B2B services — each with at least four named authors and at least 30 articles published per author over the prior 12 months. Basket of 60 commercial-intent queries per client, captured weekly with Playwright, parsed for cited URL set and cited author. We then rated each cited URL on six byline signals: presence of named human byline, link to a dedicated author page, photo, credentials string, sameAs graph completeness, and last-published recency on the author page. We added a control: the same author appearing on two different domains (one client, one third-party guest post) lets us partially separate the byline signal from the domain signal.
The window again is 30 days because shorter windows were too noisy on a per-author basis. We dropped any author with fewer than 12 indexed articles in the audit window — the composer treats sparsely-published authors differently and they polluted the signal. The base rate for any one author getting cited in a given week was low (about 14% in our basket), so the analytical move is to compare cohorts rather than read individual numbers.
Which byline signals actually moved citations
The strongest predictor was a real author page reachable in two clicks from any of the cited articles. Authors whose byline linked to a page with at least 10 cross-linked articles, an author bio paragraph and a stable URL were cited 2.6× more often than authors whose byline was either plain text or linked only to a category-style "all posts by" archive with no bio. The composer is using the author page as an entity surface — the same way it has used Wikipedia entity pages for years. If the page does not exist, the author is functionally anonymous to the composer no matter how many articles they have written.
The second signal was a sameAs graph that actually resolves. We checked the JSON-LD `sameAs` arrays on every cited author page in the basket, then HTTP-checked each URL. Pages whose sameAs URLs all returned 200 (LinkedIn, X, GitHub, Mastodon, personal site — whatever the author actually uses) were cited 1.9× more often than pages with sameAs arrays containing at least one 404 or redirect chain. The 404s are not ignored — they are read as a low-quality signal and they pull down the entire entity confidence. We saw this most starkly on three publisher clients where the dev team had migrated authors a year prior and never re-pointed sameAs links; cleaning the array gained measurable share inside two weeks.
The third was naming a human, not a brand. Articles with a "by Acme Editorial" or "by Acme Team" byline were cited 0.4× as often as articles by the same domain with a named human byline, controlling for article topic and recency. The composer reads "Editorial" as opaque — there is no entity behind it, no author page worth surfacing, no sameAs graph to verify. Brand-team bylines are not a hack; they are an entity-erasure on every page they appear on.
Author pages: the underrated entity surface
The author page is doing more work in 2026 than it did in 2022. Five things matter on it, in roughly this order: a stable URL the author page lives at forever, a bio paragraph that explicitly states the author's beat and tenure, a list of at least 10 cross-linked articles with dates, valid Person schema with sameAs, and structured organisation affiliation if the author is staff. The first three are content; the last two are markup. We have stopped recommending author pages with bio cards that are empty placeholders or "stay tuned" stubs — the empty page is read as evidence of a low-trust entity, which is worse than no page at all. Either fully populate it or do not link to it from the byline.
One pattern we now consistently push on publisher clients: the author page should link to the canonical brand "About" page and to a "team" index page. The composer is reading the author-org graph as a triangle — author has a page, page belongs to organisation, organisation has its own About — and the citation weight for the author goes up when all three sides of the triangle are reachable and consistent. The opposite — author page exists but does not connect back to the organisation, or the org page has no team index — is read as a self-published-blog signal and the citation weight halves.
The sameAs trap — and what to do instead
SEO writeups in 2024 told everyone to stuff sameAs arrays with as many social URLs as possible. That was mediocre advice in 2024 and is actively harmful in 2026. Long sameAs arrays containing dormant or 404'd URLs underperformed short, well-curated ones in every cohort we measured. The composer is not counting URLs; it is verifying them, and an unverifiable URL is treated as a quality drag on the entity it claims to identify. Three to five well-maintained URLs beat ten where two are dead.
There is a related anti-pattern: sameAs arrays containing a URL to a third-party "verified profile" page that itself fails to mention the author. We saw this on five different clients using crowdsourced expert profile services — the sameAs link goes to a profile page, the profile page has the author's name in the title, but the body text is generic SEO filler that does not name the author or describe their work. The composer follows the link, fails to verify the entity, and pulls the citation weight down. Audit your sameAs arrays the way you would audit a backlink profile — verify each URL actually identifies the author it claims to.
What changed in our content checklist
Three additions. We require a real author page for every byline appearing on a commercial page; the page has a stable URL, a populated bio, at least 10 cross-linked articles, and Person schema with verified sameAs. We require all "Editorial" and "Team" bylines to be replaced with named authors on every commercial-intent page; staff writers can use pen names if needed, but the byline must point at an actual entity. And we run a quarterly sameAs verification — every URL in every author Person schema gets HTTP-checked, and the array is curated to entries that return 200 and actually identify the author.
We dropped two habits. The "verified by an expert" sidebar with a separate expert byline below the main author was reducing citation rates in our test cohort, not raising them — the composer treated it as an ambiguous attribution signal and dropped weight from both bylines. We now use single-author bylines with the expert credited in a contributor line at the bottom. We also stopped recommending photo cards as an E-E-A-T move; they have no measurable impact on citation rates and the dev cost is non-trivial. Spend the budget on the author page instead.
- 01Build a real author page for every byline on commercial pages. Stable URL, populated bio, 10+ cross-linked articles, Person schema with sameAs. Cited 2.6× more often than plain-text or stub bylines.
- 02HTTP-check your sameAs arrays quarterly. Pages with all-200 sameAs URLs were cited 1.9× more often than pages with at least one 404. Three verified URLs beat ten with two dead.
- 03Replace every "Editorial" or "Team" byline on commercial pages with a named human. Brand-team bylines were cited 0.4× as often, controlling for topic and recency.
- 04Skip photo cards and credential badges. They had no measurable citation impact in our basket. Spend the budget on the author page and the sameAs verification instead.
Where this argument breaks
For sites under about 50 indexed articles, the audit does not yet pay — the per-author binary outcome rate is too low to read signal from noise inside a 30-day window. For pseudonymous publications (some financial commentary, some security research), the named-human signal works against the editorial model and the trade-off is real. In Chinese-language search the byline conventions are different: 百家号 attribution is at the account level and the composer reads account history rather than per-article author entities — the structural advice transfers but the tactical execution does not. Outside those carve-outs, the byline is a load-bearing entity signal in 2026, and most teams are still investing in the wrong half of it — the visible decoration above the fold instead of the entity graph the composer actually verifies.