Primary sources

The Evidence Locker

The documents themselves. Internal Google and YouTube files on news ranking, 'fringe' classifiers, query blacklists and ad policy, released by former Google engineer Zach Vorhies in 2019, posted here in full with employees' personal details removed.

Document

YouTube 'controversial query blacklist' (2018)

Google source file, dated July 2018 in the copy. Search queries listed here force YouTube's 'controversial twiddler' to trigger; the list is…

TEXT FILE

Document

Google Now news-source exclusion list

Manual list of sites excluded from Google Now news stories. It includes torrent sites and a section labeled 'sites with high user block rate…

Document

'Less Fringy Search' engineer profile (2017)

Internal profile of principal engineer Paul Haahr, whose listed mission is 'Less Fringy Search' and 'fringe ranking: not showing fake news, …

Document

YouTube 'fringe ranking / classifier' slides

Slides showing a rater-based channel-quality classifier scoring outlets (Wall Street Journal 8.53, CNN 6.00, Fox News 5.20, Breitbart 2.78),…

Document

Design doc: feeding WebSpam manual actions to Google News (2017)

Engineering design for removing domains from the Google News corpus when they receive certain WebSpam manual actions.…

Document

Google 'Fake News-letter' (Nov. 27, 2017)

Internal newsletter of the news-quality working group: revamping the News corpus around a 'single point of truth' for what counts as news, q…

Document

Training deck: 'Google efforts to address Fake News' (Dec. 2016)

Internal training slides on Google's post-2016-election response to 'fake news' across Search, News and ads.…

Document

AdSense FAQ on 'fake news' and hate speech

Internal sales FAQ explaining AdSense's misrepresentative-content and hate-speech policies, including that hate speech is a 'narrowly define…

Document

Googlers' campaign to pull ads from Breitbart (2017)

Internal group threads and a letter to Sundar Pichai, signed by more than a thousand employees, asking Google to remove Breitbart from AdSen…

Document

Bug reports: 'Infowars is not a credible source' (2016)

Internal tickets filed after the 2016 election asking that Infowars not appear as a top News result for 'presidential election.'…

Document

Ads restriction ticket on a CNN Las Vegas page (2017)

A page-level domain restriction request after the Las Vegas shooting, listing dozens of news URLs from major outlets for ad-restriction revi…

Document

Machine-learning 'fairness' training slides

Slides from an internal training on bias in algorithms: 'People are programmed,' and an example slide using a Trump quote as 'explicit bias.…

Document

'ML Fairness Communication Best Practices' (privileged)

Internal guidance on how employees should write about fairness and bias in Google's algorithms, including when and how to keep such document…

Document

'Definition of Algorithmic Unfairness' (Feb. 2017)

Company-wide definition of 'algorithmic unfairness' meant to apply across Google products and teams.…

Document

'Allegations of Algorithmic Bias: Investigation and Meta-Analysis'

Confidential internal review by Google privacy and policy staff of public allegations that its algorithms were biased.…

Document

Comms doc: 'ML Fairness'

Privileged communications plan for how Google talks publicly about fairness in machine learning, including internal-only goals.…

Document

'Bias Busting for Machines' (Google internal)

Internal Google training on 'bias busting' applied to machine-learning systems.…

Document

'Beginner's Guide to Protesting' (GooglersUnite)

An internal employee-group guide on organizing and protesting, shared among Google staff.…

Document

'Fair is not the default' training deck

Internal training presentation arguing that machine-learning systems must be actively adjusted to be 'fair', because unadjusted data reflect…