Author nameMeasured on our own author table: 1,442 separate records are named
exactly “Y. Wang”, 1,307 exactly “Wei Zhang”, and 45,354 have “Smith” somewhere in the name — so a name match is not one
person’s output. The author directory is where identities live, and the link under this box
carries whatever you have typed — though only 7,182,821 of its 30,780,981 records (23.3%) carry an
ORCID, so it disambiguates a minority of people rather than most of them.
Matched as text against the names printed on each article — not against a person.
Topic of the paperThe subfield our source assigns to the article itself. The corpus is not
spread evenly across the 246 of them: the largest single subfield, Molecular Biology , holds 4,091,690 of
the 104,664,791 articles (3.9%). A few subfields are also used by our source as catch-alls, which is a labelling problem
rather than a subject one — pick Aerospace Engineering and the warning above the results says what it costs.
Any topic
Discipline of the journalRecorded for 61,903 of the 63,664 catalogue records (97.2%), which between them cover 62,643 journals. It needs a search term, an author,
a topic, a journal or a language beside it. On the largest disciplines it can time out on a cold cache — Medicine
alone spans 12,005 catalogue records (11,949 journals), so the database checks every matching article against its journal. Observed: a common word
plus Medicine timed out on the first attempt and completed on the second. The ceiling is eight seconds, inherited from the
role PostgREST connects as (measured 3 Sep 2026; this said “the three-second ceiling” until then, and the
measurement behind the sentence was taken when the repo believed that). A narrower search term
always completes. Making it reliably fast would mean copying the discipline onto all 104.7M article rows, and this database
does not have the room.
Any discipline
LanguageMost of the corpus has no language recorded: 97,520,411 of the
104,664,791 articles — 93.2% — carry none in the source , and none of them is returned when a language is
selected. So this filter does not narrow the corpus, it replaces it: the counts in this list sum to 7,144,380, the
6.8% that do carry one. Search without it unless the language itself is the question. The language is the source’s, not ours, and it is
sometimes wrong: the two most-cited results under Turkish are English-language papers labelled tr upstream.
We do not overwrite it — 778 of the 12,221 tr titles carry no Turkish-specific letter, which is an upper bound
on how many look wrong, not a count of errors.
Türkçe
English
Any language
Type
Articles and reviews
Articles only
Reviews only
Conference articles only (776)
AccessAccess categories are OpenAlex’s own oa_status, carried through unchanged — not a classification of ours, and not derived from the APC field.
Any access model
Open access (any route)
Diamond — free to read and publish
Gold — open in the journal
Hybrid — open article in a paywalled journal
Green — open in a repository
Bronze — free to read, no licence
Closed
ImpactCNI and percentile are computed for 100,954,420 of 104,664,791 articles
(96.5%). The remaining 3,710,371 were all published before 1950 , which is where the normalisation
baseline starts — setting either filter removes every pre-1950 article and nothing else. From 1950 on every
article carries both: a subfield-year-type cell too small to normalise against is normalised at field or domain
level instead, never dropped. This said the excluded rows "sit in a cell too small to
normalise against" until 30 Aug 2026. The two sets are identical — re-measured 19 Sep 2026 on the site's own
server: 3,710,371 articles carry no CNI, 3,710,371 were published before 1950, and not one article from 1950 or
later is missing one — so the stated mechanism was not the one operating. The pair read 341,812 each on the
7,167,968-article corpus the site served until that day.
Percentile in its own field
Any
Top 1%
Top 5%
Top 10%
Top 25%
Normalized impact (CNI)
Any
At or above the world average (1.00)
Twice the world average
Five times the world average
Ten times the world average
Citations
Availability63,484,528 of the 104,664,791 articles — 60.7% — carry an abstract , measured on
the served table 19 Sep 2026. Tick the abstract box and the other 41,180,263 leave the result, so read their absence
carefully: it means we hold no abstract for that article , not that the paper has none. None of these
three boxes can start a search on its own — the database needs a search term, an author, a topic, a journal, a DOI or
a language to work from, and these only narrow what that finds. This group carried no note at
all until 19 Sep 2026, while seven other filters on this page did. The figure above is the first abstract
coverage measured on the table the site serves; the earlier ones in our notes were taken on samples and
snapshots of other populations and are not comparable to it.
AuthorshipThe number of authors on the paper. Recorded for 7,159,771 of 7,167,968 articles (99.89%); the remaining 8,197 carry a count of zero, which the article page shows as “not recorded”. This said “all 7,167,968” for part of 30 Aug 2026 — n_authors is non-null everywhere, but non-null is not the same as recorded, and the two pages said opposite things about the same 8,197 records. Leave the second box empty for an open-ended floor — 100 and above is the usual way to isolate large
collaborations.
Sort by
Citations
Normalized impact (CNI)
Percentile
Newest first
Oldest first
Reset
Enter a search term to begin.
⬇ Export CSV (max 1,000)
Covers 104.7 million articles and reviews from 61,148 of the 62,643 journals in the catalogue. This said “7.2 million — the 25 most-cited of every indexed journal, plus the most-cited works of every author” until 19 September 2026, and that is what the corpus was: a selection. The article layer moved to our own server that day and stopped being one — measured on the served table, 103,183,183 rows now sit beyond their journal’s 25th, and the deepest journal runs to 1,031,216 ranked rows. It is still not the entire literature: where a journal is outside the catalogue, its articles are outside this corpus too (see the roadmap ). Retracted work is outside this corpus by design , so a retracted paper will not be found here and its absence is not a verdict on it. This said “on both sides of every citation” until 30 Aug 2026. Only the cited side is filtered — etl/metric-pipeline.sql builds works_core with NOT is_retracted, while the citation-graph passes carry no such predicate, so citations MADE BY a retracted paper still count toward every CNI and percentile shown here. Errata E64 — see the methodology. Source: OpenAlex (CC0). CNI normalizes citations against each article's own subfield, year and document type (1.00 = world average); Pct is its percentile in that peer group. Per DORA , these describe individual articles, not their authors.