7 Wikipedias exist in Philippine languages and hold 7,483,672 articles between them. Bots created 7,232,538 of those, almost all inside 2 editions. Cebuano alone has 6,115,840 — 85.0% the size of English Wikipedia — and is read 6.0 times per article a year, against Tagalog's 1,167.9.
The analytics API answers a question most article-count league tables never ask: who created the pages. The same endpoint reports creations by bot, by registered human and by anonymous editor, so this is not inferred from usernames — it is the register's own accounting.
pages sit in the 2 bot-built editions, Cebuano and Waray. Every other Philippine language edition has 100,784 between them — 73× fewer.
of its 6,115,840 pages. The bot wrote 6,089,484. That is 99.6% of an encyclopedia.
of Tagalog's 49,155 pages are bot-made. The national language's encyclopedia is 124× smaller than Cebuano's and almost entirely handwritten.
Article count is a number a bot can manufacture. Readership is not. Dividing one by the other separates the two Philippine encyclopedias that were written from the two that were generated.
views per article per year, from 36,471,581 views spread across 6,115,840 pages. Waray is much the same, at 6.7.
views per article, from 57,409,910 views — more total readers than Cebuano, on 124× fewer pages.
views per article on English Wikipedia, and 1,767.0 on Indonesian. Tagalog sits in the same world as those. Cebuano does not sit in any of it.
Page creation by year makes the mechanism visible. A human encyclopedia grows in a slow, noisy line. A generated one arrives in bursts, and then stops.
Cebuano's busiest year was 2016, when bots created 1,912,485 pages — more in twelve months than every non-bot-built Philippine edition has managed in its entire history, several times over. Tagalog has never exceeded five figures in a year.
Views do not all land on articles. Some land on the main page, the search box or a maintenance namespace — navigation rather than reading. Splitting the most-viewed thousand pages of each edition by which kind they are says more than the totals do.
of its most-viewed thousand pages' traffic goes to the main page, the search box or a maintenance page, out of 323,667 views. Tagalog's figure is 34.2%.
views for Cyclamen, the most-read actual article in Cebuano that month — a European garden plant, which is what happens when an encyclopedia is generated from species databases.
views for José Rizal. Below it that month: Buwan ng Wika, Andrés Bonifacio, the Philippine–American War. A country reading about itself.
The obvious objection is that a bot-built edition might still have a large human community standing behind it, tending what the bot made. The editor counts say otherwise, and say it in the most deflating way possible: the communities are comparable.
Cebuano's busiest month ever had 171 human editors. Tagalog's had 253. Two encyclopedias with human communities of the same order, one of which has 124× more articles than the other.
Nothing here is an accusation. The bot ran openly, with permission, and the pages it made are accurate as far as they go. The point is narrower and more useful: a count of articles measures how much text exists, and nothing else at all.
Four limits, and the first bounds every ratio on the page.
Every total here is a sum of monthly creations and does not subtract deletions, because the analytics API publishes creations and not a live count. It is the same measure for every edition, so the comparisons hold, but it is not an article count and is not labelled as one.
Page creation is counted from 2001-01; pageviews only exist from 2015-07 onward. Views per page therefore divides one window by another. That understates nothing for a bot-built edition whose pages predate 2015, and it is stated rather than smoothed over.
Speaker populations would turn this into views per person, which is the comparison a reader reaches for next. Every such source is paywalled or carries an attribution licence this site does not accept, so the editions are compared only with each other.
Wikimedia's analytics are CC0; the encyclopedia is CC BY-SA. Titles are published as identifiers and no article text is fetched, cached or published — a check fails the build if a column that could hold any ever appears.
Every figure on this page traces to one of these, through a CSV in
data/wiki-bots/. Each is checked against its source query on every
build.
Monthly pageviews for each of the nine Wikipedia editions on this page, from July 2015 to August 2026, counting human agents only — Keyless: no account, no token, no form. Wikimedia answers 403 without a descriptive User-Agent, which is a header rather than a credential, and a project URL satisfies it. The agent filter matters more here than anywhere else in this series: a bot-built encyclopedia attracts crawlers, and counting them would flatter exactly the editions the page is about
How many content pages each edition gained every month, split by whether a bot, a registered human or an anonymous editor created them — The whole finding rests on this one parameter. The same endpoint answers for group-bot, name-bot, user and anonymous, so “who built this encyclopedia” is a question the API answers directly rather than something inferred from usernames. Months with no creations are omitted from the response and are read as zero
The licence this page relies on, quoted beside the one it does not: “All Analytics datasets are available under the Creative Commons Zero (CC0) public domain dedication, unless otherwise specified” and “All original textual content is licensed under the GNU Free Documentation License (GFDL) and the Creative Commons Attribution-Share-Alike 4.0 License” — Both sentences sit on the same page, which makes the boundary unusually clean: the counts are public domain and the encyclopedia is not. This site is built to public domain or CC0 only, so the counts are in bounds and no article text is fetched at all
How many registered humans edited content pages in each edition each month — Published to answer the obvious objection to the headline: perhaps a bot-built edition still has a human community behind it. The counts say the Cebuano and Tagalog editions have human editorships of the same order, while their article counts differ by a hundredfold
Would have shown what any of these seven million articles actually say — CC BY-SA 4.0 and GFDL — an attribution and share-alike obligation, and this site is built to public domain or CC0 only. Article titles are published as identifiers, the way element names and company names are elsewhere in this series; no sentence of any article is fetched, cached or published, and a check asserts no mart can grow a column that could hold one
Would have provided per-article view counts for every article in every edition, rather than the top thousand — Same licence, rejected on footprint and on judgment. One day of the user-agent-classified dump is roughly 677 MB, and answering this page's question from it would mean moving hundreds of gigabytes to compute totals the aggregate endpoint already publishes. Choosing the narrow API over the wide dump is the point rather than a shortcut
Would have turned views per article into views per speaker, which is the comparison a reader reaches for next — Linked here is the licence rather than a dataset, because the problem is the licence and not any one source: Ethnologue is paywalled and every open alternative carries this attribution condition, which this site does not accept. So the page compares editions to each other and to two controls, and says plainly that it is not adjusting for how many people speak each language — which matters for the small editions and not at all for the finding about the large ones.
I wrote a companion post — “The Philippines Has A Huge Encyclopedia. A Robot Wrote It.” — in simple, everyday words.