Fact-checkers review and rate the accuracy of stories through original reporting, including

Summary

Exhibit 2 to Document 56-3, Case 3:20-cv-05787-SI, Filed 10/23/20: a printed web page headed How Our Fact-Checking Program Works, dated AUG 11, 2020, on Facebook's approach to fact-checking. The page states that Facebook relies on independent fact-checkers to identify and review potential misinformation, and describes Identify, Review, and Act stages. It lists the rating options False, Altered, Partly False, and Missing Context, plus True or Satire. It lists actions taken on rated content, among them Reduced Distribution, Sharing Warning, Sharing Notifications, Misinformation Labels, and Removing Incentives for Repeat Offenders.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Date: 2020-10-23

Cite this document

“Fact-checkers review and rate the accuracy of stories through original reporting, including.” Original at archive.org, dated 2020-10-23. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/2020-10-23-covid-era-misc-7b014902-1/, SHA-256 7b014902285a0635881f9c6c3028d7e7f23b37d617233f4ab7b6ef0b0928a308.

Full text

Case 3:20-cv-05787-SI Document 56-3 Filed 10/23/20 Page 1 of 7

Exhibit 2
Case 3:20-cv-05787-SI Document 56-3 Filed 10/23/20 Page 2 of 7

PROGRAMS — AUG 11, 2020

How Our Fact-Checking Program Works

Facebook’s Approach to Fact-Checking: How It Works

1
Case 3:20-cv-05787-SI Document 56-3 Filed 10/23/20 Page 3 of 7

Since we do not believe that a private company like Facebook should be the arbiters of truth, we rely
on independent fact-checkers to identify and review potential misinformation, which enables us to
take action.

1 Identify
We identify potential misinformation using a variety of signals, and our fact-checking partners
can also proactively identify content on their own.

2 Review
Fact-checkers review and rate the accuracy of stories through original reporting, including
interviewing primary sources, consulting public data and conducting analyses of media,
including photos and video.

3 Act
Once we have a rating from a fact-checking partner, we take action by ensuring that fewer
people see that misinformation by surfacing fact-check articles to users across our platforms
and showing labels on top of false stories.

1. IDENTIFY
How does Facebook use technology to detect potential misinformation?

We use a number of signals to predict content that might be misinformation, and surface it to fact-
checkers.

For example, we use feedback from our users who ﬂag content they see in their News Feed as being
2
Case 3:20-cv-05787-SI Document 56-3 Filed 10/23/20 Page 4 of 7
potentially false. We also look at other user patterns, like people commenting and saying that they
don’t believe a certain post.

We also use machine learning models to continuously improve our ability to predict misinformation.
We feed ratings from our fact-checking partners back into this model, so that we get better and better
over time at predicting content that could be false.

2. REVIEW
What ratings options can fact-checking partners choose from?
There are several rating options that third-party fact-checkers can apply to content.

False
Content that has no basis in fact.

Altered
Image, audio, or video content that has been edited or synthesized beyond adjustments for clarity
or quality, in ways that could mislead people.

Partly False
Content that has some factual inaccuracies.

Missing Context
Content that may mislead without additional context.

Although our focus is on identifying misinformation, fact-checking partners can also let us know if
content they’ve reviewed is True or Satire.

3
Case 3:20-cv-05787-SI Document 56-3 Filed 10/23/20 Page 5 of 7

3. ACT
How does Facebook take action based on fact-checkers’ ratings?

When False, Altered, Partly False or Missing Context is selected by a fact-checking partner, we take
action. Some of these actions may include:

Reduced Distribution
We show the piece of content lower in News Feed, signiﬁcantly reducing its distribution. And on
Instagram, we remove it from Explore and hashtag pages and downrank content in feed and
Stories.

Sharing Warning
When someone tries to share a post that’s been rated by a fact-checker, we’ll show them a pop-
up notice so people can decide for themselves what to read, trust, and share.

Sharing Notifications
4
Case 3:20-cv-05787-SI Document 56-3 Filed 10/23/20 Page 6 of 7
If someone has shared a story that is later determined by fact-checkers to be false, we notify
them that there is additional reporting on that piece of content.

Misinformation Labels
We apply a clear, visual label to content that has been debunked by fact-checkers, and surface
their fact-checking articles for additional context.

Removing Incentives for Repeat Offenders
When Pages or websites repeatedly share content that’s been debunked by fact-checking
partners, they will see their overall distribution reduced, and will lose the ability to advertise or
monetize within a given time period.

f t l m

Did you ﬁnd this content helpful?

Yes No

O R E STO R I E S
5
Case 3:20-cv-05787-SI Document 56-3 Filed 10/23/20 Page 7 of 7
MO

L AT E S T

Fact-Checking Innovation
Initiative

© 2 0 2 0 FA C E B O O K I N C . A L L R I G H T S R E S E R V E D .
P R I VACY P O L I CY
TERMS
COOKIES
FO L LO W U S O N F B E

English (US) ‫ اﻟﻌرﺑﯾﺔ‬Deutsch Español (España) Español Français (France) Bahasa Indonesia Português (Brasil) ภาษาไทย

File and source

File: gov.uscourts.cand.364478.56.3.pdf
Size: 1,187,314 bytes
SHA-256: 7b014902285a0635881f9c6c3028d7e7f23b37d617233f4ab7b6ef0b0928a308
Our copy: gov.uscourts.cand.364478.56.3.pdf
Original: archive.org