BrandTabulatorby Michael Quinn · MichaelQuinn.ai
Methods

Methods

Status. Pre-launch. Issues dated before the first public issue may be regenerated. From the first public issue on, each week's data files are kept unchanged under their date.

What this measures

When people ask AI assistants about products, the answers lean on the public record. This site looks at two parts of that record for each brand, and compares brands within a category.

The keel: facts. The brand's entry in Wikidata, a free database of facts that search engines and other knowledge graphs reuse. How many of its facts carry a source, whether the basics are filled in, and whether it points to the brand's own products and companies.

The sails: descriptions. The brand's English Wikipedia article. How many references it has, how many languages it exists in, how often it is edited, and when someone last made a real change to it.

It also counts how often machines load the article. It does not rank brands as products, and it does not say which brands are best.

Which brands are included

Each category comes from a published Wikidata query, so anyone can reproduce it. Only brands that have customers, business or consumer, are included; companies that Wikidata records as dissolved are left out. Each category page states its own query, any brands added or left out by hand, and why.

Definitions

#
Rank by staleness within this category. 1 = longest without a substantive edit.
Brand
English Wikipedia article and Wikidata item (Q-id). Click either to see the source.
Last substantive edit
Most recent English Wikipedia edit of 50 bytes or more by a non-bot account that was not later reverted.
90d edits / editors
Edits to the English article in the 90 days to the issue date, and how many distinct accounts made them.
Reverted
Of those 90-day edits, how many were reverted. A measure of contestation.
Statements (referenced)
Number of statements on the Wikidata item, and how many carry at least one reference.
Ballast
Share of Wikidata statements that carry a reference. Sourced, complete Wikidata is what lets the record hold up.
Identity links out
Statements on the item that point to the brand's own products, brands, subsidiaries, holdings, founders, origin, motto, awards or executives. What the item says the brand is.
Key props
Of six properties a brand item should carry: industry, product, parent, subsidiaries, owner-of, Google Knowledge Graph id.
Editions
Language Wikipedias (and sister projects) with a page linked to this item. Sail area.
Refs
Reference tags in the English article.
Words
Approximate word count of the English article.
Reader views
English Wikipedia views in the stated month from Wikimedia's "user" agent class: traffic that presents as a browser. Not a guarantee of a human.
Spider
Views from declared crawlers such as Googlebot.
Automated
Views from undeclared traffic that Wikimedia's heuristics flag as bot-like. The scraper signal.

Details behind the definitions

Views and machine visits

Views come from Wikimedia's public pageview service, for English Wikipedia, for the previous full calendar month. Wikimedia sorts traffic into three classes. Reader ("user"): traffic that looks like a person using a browser. Some automated traffic that disguises itself ends up here. Spider: crawlers that identify themselves, such as Googlebot. Automated: traffic Wikimedia's own checks flag as automated.

These counts cannot see the bulk copies of Wikipedia that search and AI companies take directly, through downloads and Wikimedia's paid feed. No one outside those companies can see that. Machine visits here are a floor, the same floor for every brand.

Rankings

The default order is by staleness: the brand that has gone longest without a substantive edit comes first. Any column can be used to sort. There is no combined score yet.

How the numbers are checked

All figures are fetched from Wikimedia's public services by a published script, which runs every Monday on GitHub. Each request identifies the project, and requests are spaced out as Wikimedia asks. The first run was compared figure by figure with a separate hand pull, and they matched. Each brand's record in the JSON carries the revision ids it was computed from, so any figure can be traced to the exact version of the article or item.

Planned, not yet in place: an automatic week-on-week check that stops publication if numbers jump implausibly, and a check of any written summary against the edits it describes. Until then the author reviews each issue.

If your brand is listed

Anyone can correct facts in Wikidata, including people who work for the brand. Wikimedia's Terms of Use require anyone who is paid to edit, such as staff or an agency, to disclose that on their account. English Wikipedia also asks people with a conflict of interest not to edit the article directly, and to suggest changes on the article's talk page instead.

Corrections

Report an error on GitHub or through michaelquinn.ai. Corrections are published as dated notes.

Data and licence

Wikipedia text is licensed CC BY-SA; Wikidata is CC0. The data files here are published under CC BY-SA 4.0. Each category page links to its own weekly CSV and JSON files.

Related work

Michael Quinn · michaelquinn.ai · Questions: Michael@MichaelQuinn.ai · Computed from public Wikimedia data. Code: github.com/MinorNobles/brandtabulator