Wikipedia’s core mission is to be a source of credible information. This requires maintaining the credibility of not only internal articles but also external references. That’s where WikiSignals can help. The WikiSignals project is developing a credibility-related Wikipedia editing tool.

Determining the reliability of a source isn’t easy. But the data to make that determination is easily available. We’re merging lists of press association members, Media Bias/Fact Check ratings, domain registration data, global site rankings, predatory and peer-reviewed science journals, vetted media indexes (e.g., Project Oasis, trust.txt), Wikipedia references (e.g., Cite Unseen, Perennial Sources, Citation_Watchlist, Fake News Websites), and most-used domains in Wikipedia citations.

WikiSignals is compiling all that data into a single, searchable, easy-to-use resource so Wikipedians can better evaluate source credibility. Our dataset now has information on 70K websites. Search for a domain name at WikiSignals.org. We’ll be adding more data over the coming months and documentation for our API.

We’re experimenting with a form for Wikipedians to get credibility indicators for citations. This WikiSignals tools mockup demonstrates how people might enter a URL, such as for a news article, to get data about both the article and its website, along with a Credibility Score…

…Then have that URL formatted as a Wikipedia citation (i.e, wikitext) for pasting into the editor.

One of several valuable bi-products already produced by this project is our comprehensive list of credibility indicators — machine-harvestable data we’ll use to programmatically estimate website reliability. We’re eager for input on this list: If anyone has suggestions or comments, please get in touch.

Part of our process has been to interview active Wikipedia editors. Their invaluable input led to us creating another resource, a compendium of their favorite Wikipedia Editing Tools. We’ve also nearly finished an index of predatory and hijacked journals, a merge of existing lists (Beall’s, Retraction Watch, etc.).

The WikiSignals project grew out of the WikiCredCon 2025 conference. We noticed many attendees were working on reliability-related projects but few were working together. So we began this collaboration, between Veri.FYI, Factiverse, SimPPL, AfroCrowd, Iffy.news, and the Internet Archive Reference Explorer: organizations pioneering efforts to improve access to source-credibility information. Our first public presentation was at WikiConference North America 2025.

We need your help!

We’re looking for coders, funders, editors, and anyone else interested in producing a credibility tool for Wikipedia and beyond. Leave us a message on our Talk page.

WikiSignals: Strengthening the Credibility of Wikipedia Citations was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

A proposal for classifying knowledge discrepancies within Wikipedia articles into four categories.

by Netha Hussain and Matthew Yeager.

This article was originally published on Diff, a blog by and for the Wikimedia volunteer community. It is the second article in a series exploring knowledge on Wikipedia. Read the previous article here.

We kicked off this series by exploring the foundational principles of knowledge on Wikipedia: its dynamic, factual, verifiable, collectively significant, and neutral content, all presented in a way that’s digital, meaningful, accessible, and usable by machines.

There’s a lot of valuable knowledge still waiting to be added on Wikipedia, and on the flip side, some content may have found its way in that doesn’t fully belong. These gaps and excesses in knowledge, which we call knowledge discrepancies, are the subject of this blog post.

What is the structure of knowledge on Wikipedia?

Wikipedia has over 300 language editions. While English Wikipedia has around 7 million articles, there are language editions that may have just a few thousand. Each language version serves a unique audience, and the articles in each language reflect the informational priorities and cultural/linguistic nuances of the users using that language. Topics can differ significantly in emphasis, depth, or even framing in different languages based on the perspectives of the users shaping it. For example, the article on a religious figure like Jesus can look remarkably different in the English, Hebrew, Arabic, and Malayalam editions of Wikipedia, each shaped by its own cultural lens.

Within each language version, knowledge is primarily packaged into self-contained articles. Each article aims to give readers a comprehensive understanding of a specific topic, serving as a complete, informative hub. Every article is designed to be a one-stop source, where readers can grasp all relevant information related to a topic without needing to leave the page. That brings us to the question, what elements constitute a Wikipedia article?

Wikipedia’s articles are written in wikitext, a custom markup language, and enriched by several digital components:

• Multimedia elements: Directly embedded images, videos, and audio clips enhance engagement beyond the ways traditional encyclopedias can.
• Structured data: Elements like infoboxes and property values pulled from Wikidata provide standardized, machine-readable information, making classifications, comparisons, and data extraction more efficient. Wikipedia also has templates, which are pre-designed, reusable code, for various purposes such as navigation (example), citation (example) and for providing maintenance information (list of maintenance templates).

• Categories: Another organizational structure is the thematic, hierarchical tagging system called category trees. Tagging articles with broad and specific topics allows users to navigate and discover related content. The category tree of most Wikipedia articles can be complex, with each category branching in different directions. This system helps users navigate through the content and discover related articles on similar subjects. The hierarchical structure allows users to understand where an article fits within the broader context of related topics. The categories related to the Roman Empire can be found here, and you can see one path to the Roman Empire starting from all Articles:

• Internal links: Arguably the most defining structural feature, internal links transform Wikipedia from a collection of isolated articles into a vast, interconnected web of knowledge.

Emergence of Knowledge Discrepancies

While Wikipedia has a robust structure, the content presented within this structural framework may contain knowledge discrepancies. A knowledge discrepancy can manifest as the absence of content on a collectively significant topic or the presence of unwanted knowledge that does not belong on Wikipedia. For clarity, we propose classifying knowledge discrepancies within articles into four categories:

Missing content: Refers to areas where information should be present but is entirely absent. This includes missing content within articles, like a missing section on a major historical event within a country’s history article, as well as the absence of articles on significant topics. Missing content can also include missing multimedia like a key image, or structured data like birth dates on a biography article. Missing content on Wikipedia has been widely studied and is straightforward to understand. We recommend going through the “content” section of the knowledge gaps taxonomy to get a broader perspective about the classification of missing content on Wikipedia.

Poorly structured content: Refers to information that is present but lacks proper organization, context, or accuracy, making it difficult for readers to comprehend. This could include excess or improper content that negatively affects the meaningfulness of the article. On the other hand, poorly structured content is often not strictly considered a knowledge discrepancy and, therefore, less well studied. As we continue, we will offer a framework for classifying poorly structured content that appears within Wikipedia.

Surplus content: Some areas on Wikipedia have a knowledge surplus, with content that is not relevant for Wikipedia. These include non-notable content, inappropriate material such as hate speech or vandalism, and content that involves copyrighted material not suitable for Wikipedia’s open-licensed framework.

Cataloging mismatches: Beyond the content itself, the organizational backbone of Wikipedia can also present misunderstanding. These mismatches involve structural elements like missing or incorrectly applied categories, templates, or other structured data. Such categorical imperfections impede the discoverability of information for both humans and machines. These create “blind spots” that hinder machines from fully understanding, indexing, and leveraging Wikipedia’s content.

The rest of this blog post focuses on deconstructing poorly structured content. As mentioned earlier, missing content has already been well studied, and surplus content and cataloging mismatches are, for now, outside the scope of our interest.

Deconstructing Poorly Structured Content

Knowledge discrepancies — ranging from a typo to the absence of an entire article to a surplus of vandalized content — leave significant holes or vulnerabilities in the knowledge available to readers. Identifying poorly structured content becomes easier when seen through the lens of what knowledge Wikipedia already contains:

As we’ve established, Wikipedia contains dynamically updated(1), factual(2), verifiable(3), and collectively significant knowledge(4), presented neutrally(5) and organized in a digital format(6) that is meaningful(7) and accessible(8) to humans and usable for machines(9). Therefore, knowledge discrepancies can also include content that falls short of these ideals. Here’s a framework for classifying these types of discrepancies:

1. Outdated: An article becomes outdated when it fails to incorporate new discoveries, significant events, or other changes in understanding that have occurred since its last update. This leaves readers with incomplete or even misleading information. For example, an article on a political issue not updated after major elections or policy changes, or a scientific article missing the recent findings in a rapidly advancing field, would both be considered outdated.
2. Inaccurate: This could range from simple typos or incorrect dates to misquoted sources, flawed calculations, or fundamentally incorrect statements that misrepresent established facts. Such inaccuracies erode trust and directly contradict Wikipedia’s commitment to providing reliable information. For instance, an article might include incorrectly translated material from a foreign language source or present a historical fact with the wrong date or context.
3. Unverified: Wikipedia’s core principle of verifiability demands that every claim be attributable to a reliable, published source. When content appears without any citations or relies on sources that don’t meet Wikipedia’s standards for reliability (like personal blogs, social media, or opinion pieces masquerading as fact), it falls into this category. This gap means readers cannot check the information themselves, making it untrustworthy. An example would be an article related to a disease condition claiming benefits of alternative medicine without credible citations, or one with dead external links that can no longer be checked.
4. Lacking Notability: Wikipedia is an encyclopedia, not a directory or a platform for every piece of information. The “collective significance” principle means a topic must have received significant coverage in reliable, independent sources to warrant a standalone article. An article about a local celebrity who, while well-known in their immediate community, has not received significant coverage in national or international media, would be considered unnotable.
5. Biased: Wikipedia’s principle of neutrality requires all significant viewpoints to be represented fairly and in proportion to their prominence in reliable sources. Bias can manifest through selective sourcing, emotionally charged or loaded language, or the outright omission of counterarguments, leading to a skewed understanding for the reader. For instance, a biography that gives excessive and undue weight to the personal life of a public figure would be unbalanced.
6. Un-wikified: Content that exists on Wikipedia but lacks proper internal linking, adherence to formatting conventions, or other basic “wikification” elements. This makes the content less integrated into Wikipedia’s broader knowledge web, difficult to navigate, and failing to fully leverage the digital nature of the platform. For example, a page about a historical event may lack internal links to related individuals, locations, or other relevant articles, forcing readers to manually search for context. When extended to broader styling and markup, it can leave for a jarring visual experience that is difficult to read.
7. Disorganized: Articles that are poorly structured, unfocused, excessively lengthy, or presented in a confusing narrative, despite containing factual information. This category also includes articles with conflicting information, logical inconsistencies, or references that are difficult to match to specific citations. These issues undermine the overall meaningfulness of the article, as readers struggle to comprehend the topic efficiently and accurately. For instance, a long medical article that reads like a disjointed academic essay, lacking clear headings and sections, and existing in a structure that is difficult to navigate. Similarly, a political article might include self-contradictory statements about a party’s stance.
8. Inaccessible: Content presenting barriers to understanding, particularly for readers with disabilities or those unfamiliar with specialized terminology. This includes accessibility challenges (e.g., color-coded graphs without alternative text for colorblind individuals) as well as the use of highly technical jargon without sufficient explanation, rendering content opaque to non-experts. Wikipedia aims for universal readability, and content that requires specialized knowledge just to understand basic concepts falls short of this goal. For example, an article about a mathematical theorem that is filled with advanced technical terms may be less accessible for readers who are unfamiliar with the subject.
9. Unfriendly to machines: Wikipedia’s digital nature means it’s not just for human eyes; it’s also a foundational dataset for artificial intelligence, search engines, and data analysis tools. Content that lacks infoboxes, structured data, categories, and internal links impedes a machine’s ability to understand contextual information. Wikipedia’s potential as a backbone for machine learning, advanced research tools, and countless other digital applications relies on structured, clean data. For instance, the availability of birth dates, occupations, and nationality in an infobox may enable an AI assistant to quickly answer factual questions. Or, an entire topic that isn’t adequately categorized prevents a machine from easily identifying and grouping related articles together when requested.

Wrapping Up

In this blog post, we have navigated the architecture of Wikipedia’s knowledge, including the structuring of Wikipedia’s articles. We introduced a framework for mapping knowledge discrepancies, with key categories for classifying poorly structured content on Wikipedia.

We believe that understanding the knowledge gaps is the first step towards addressing them. In our next blog post, we will shift our focus into action: how to prioritize and flag these knowledge gaps to effectively improve the quality of content on Wikipedia.

Stay tuned to learn how to start tackling these vital issues!

Navigating Wikipedia’s Knowledge Discrepancies was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

1 in 4 Americans reject evolution, a century after the Scopes monkey trial spotlighted the clash between science and religion

One hundred years after the trial, the culture war over evolution and creationism remains strong — and yet, when it comes to creationism, much has also changed.

By William Trollinger, Professor of History, University of Dayton and Susan L Trollinger, Professor of English, University of Dayton. Republished from The Conversation under CC BY-ND 4.0.

The 1925 Scopes trial, in which a Dayton, Tennessee, teacher was charged with violating state law by teaching biological evolution, was one of the earliest and most iconic conflicts in America’s ongoing culture war.

Charles Darwin’s “Origin of Species,” published in 1859, and subsequent scientific research made the case that humans and other animals evolved from earlier species over millions of years. Many late-19th-century American Protestants had little problem accommodating Darwin’s ideas — which became mainstream biology — with their religious commitments.

But that was not the case with all Christians, especially conservative evangelicals, who held that the Bible is inerrant — without error — and factually accurate in all that it has to say, including when it speaks on history and science.

The Scopes trial occurred July 10–21, 1925. Between 150 and 200 reporters swooped into the small town. Broadcast on Chicago’s WGN, it was the first trial to be aired live over radio in the United States.

One hundred years after the trial, and as we have documented in our scholarly work, the culture war over evolution and creationism remains strong — and yet, when it comes to creationism, much has also changed.

The trial

In May 1919, over 6,000 conservative Protestants gathered in Philadelphia to create, under the leadership of Baptist firebrand William Bell Riley, the World’s Christian Fundamentals Association, or WCFA.

Holding to biblical inerrancy, these “fundamentalists” believed in the creation account detailed in chapter 1 of Genesis, in which God brought all life into being in six days. But most of these fundamentalists also accepted mainstream geology, which held that the Earth was millions of years old. Squaring a literal understanding of Genesis with an old Earth, they embraced either the “day-age theory” — that each Genesis day was actually a long period of time — or the “gap theory,” in which there was a huge gap of time before the six 24-hour days of creation.

This nascent fundamentalist movement initiated a campaign to pressure state legislatures to prohibit public schools from teaching evolution. One of these states was Tennessee, which in 1925 passed the Butler Act. This law made it illegal for public schoolteachers “to teach any theory that denies the story of divine creation of man as taught in the Bible, and to teach instead that man has descended from a lower order of animals.”

The American Civil Liberties Union persuaded John Thomas Scopes, a young science teacher in Dayton, Tennessee, to challenge the law in court. The WCFA sprang into action, successfully persuading William Jennings Bryan — populist politician and outspoken fundamentalist — to assist the prosecution. In response, the ACLU hired famous attorney Clarence Darrow to serve on the defense team.

When the trial started, Dayton civic leaders were thrilled with the opportunity to boost their town. Outside the courtroom there was a carnivalesque atmosphere, with musicians, preachers, concession stands and even monkeys.

Inside the courtroom, the trial became a verbal duel between Bryan and Darrow regarding science and religion. But as the judge narrowed the proceedings to whether or not Scopes violated the law — a point that the defense readily admitted — it seemed clear that Scopes would be found guilty. Many of the reporters thus went home.

But the trial’s most memorable episode was yet to come. On July 20, Darrow successfully provoked Bryan to take the witness stand as a Bible expert. Due to the huge crowd and suffocating heat, the judge moved the trial outdoors.

The 3,000 or so spectators witnessed Darrow’s interrogation of Bryan, which was primarily intended to make Bryan and fundamentalism appear foolish and ignorant. Most significant, Darrow’s questions revealed that, despite Bryan’s’ assertion that he read the Bible literally, Bryan actually understood the six days of Genesis not as 24-hour days, but as six long and indeterminate periods of time.

The very next day, the jury found Scopes guilty and fined him US$100. Riley and the fundamentalists cheered the verdict as a triumph for the Bible and morality.

The fundamentalists and ‘The Genesis Flood’

But very soon that sense of triumph faded, partly because of news stories that portrayed fundamentalists as ignorant rural bigots. In one such example, a prominent journalist, H. L. Mencken, wrote in a Baltimore Sun column that the Scopes trial “serves notice on the country that Neanderthal man is organizing in these forlorn backwaters of the land.”

The media ridicule encouraged many scholars and journalists to conclude that creationism and fundamentalism would soon disappear from American culture. But that prediction did not come to pass.

Instead, fundamentalists, including WCFA leader Riley, seemed all the more determined to redouble their efforts at the grassroots level.

But as Darrow’s interrogation of Bryan made obvious, it was not easy to square a literal reading of the Bible — including the six-day creation outlined in Genesis — with a scientific belief in an old Earth. What fundamentalists needed was a science that supported the idea of a young Earth.

In their 1961 book, “The Genesis Flood: The Biblical Record and its Scientific Implications, fundamentalists John Whitcomb, a theologian, and Henry Morris, a hydraulic engineer, provided just such a scientific explanation. Making use, without attribution, of the writings of Seventh-day Adventist geologist George McCready Price, Whitcomb and Morris made the case that Noah’s global flood lasted one year and created the geological strata and mountain ranges that made the Earth seem ancient.

“The Genesis Flood” and its version of flood geology remains ubiquitous among fundamentalists and other conservative Protestants.

Young Earth creationism

Today, opinion polls reveal that roughly one-quarter of all Americans are adherents of this newer strand of creationism, which rejects both mainstream geology as well as mainstream biology.

This popular embrace of young Earth creationism also explains the success of Answers in Genesis — AiG — which is the world’s largest creationist organization, with a website that attracts millions of visitors every year.

AiG’s tourist sites — the Creation Museum in Petersburg, Kentucky, and the Ark Encounter in Williamstown, Kentucky — have attracted millions of visitors since their opening in 2007 and 2016. Additional AiG sites are planned for Branson, Missouri, and Pigeon Forge, Tennessee.

Presented as a replica of Noah’s Ark, the Ark Encounter is a gigantic structure — 510 feet long, 85 feet wide, 51 feet high. It includes representations of animal cages as well as plush living quarters for the eight human beings who, according to Genesis chapters 6–8, survived the global flood. Hundreds of placards in the Ark make the case for a young Earth and a global flood that created the geological strata and formations we see today.

Ark Encounter has been the beneficiary of millions of dollars from state and local governments.

Besides AiG tourist sites, there is also an ever-expanding network of fundamentalist schools and homeschools that present young Earth creationism as true science. These schools use textbooks from publishers such as Abeka Books, Accelerated Christian Education and Bob Jones University Press.

The Scopes trial involved what could and could not be taught in public schools regarding creation and evolution. Today, this discussion also involves private schools, given that there are now at least 15 states that have universal private school choice programs, in which families can use taxpayer-funded education money to pay for private schooling and homeschooling.

In 1921, William Bell Riley admonished his opponents that they should “cease from shoveling in dirt on living men,” for the fundamentalists “refuse to be buried.” A century later, the funeral for fundamentalism and creationism seems a long way off.

1 in 4 Americans reject evolution, a century after the Scopes monkey trial spotlighted the clash… was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

‘The press is not only a defender of democracy, but also a guardian of our collective future’

By Janine Mendes-Franco, Caribbean Regional Editor, Global Voices

World Press Freedom Day, annually observed on May 3, is a poignant reminder of the need for press freedom, an occasion to remember the sacrifices made in pursuit of that freedom, and an opportunity for journalists and media practitioners to consider any issues that may be of concern to the fourth estate.

This year, the concern being explored in the World Press Freedom Day signature event is the impact of artificial intelligence (AI) on the media. Discussions will centre around how to ensure that AI enhances, rather than undermines, press freedom and democratic values.

AI tools, while transformative in terms of efficiency, multilingualism and data analysis, among other processes, also raise unique ethical questions. AI-generated misinformation and disinformation, deepfake technology, biased content moderation, and surveillance threats to journalists represent just some of the risks, not to mention their potential effects on the industry’s business model and what it may mean for the media’s long-term viability.

Understanding the pressing nature of these issues, Global Voices recently shared its own policy on the use of AI in the newsroom. The Media Institute of the Caribbean (MIC), meanwhile, issued a press release in which it underscored the “urgent” need to address the transformative impact of AI on journalism within the context of the region’s unique challenges of “media viability, misinformation, and natural disasters.”

The natural disasters angle also fed into the theme of the 2025 World Press Freedom Day, which dealt with journalism in the face of the global environmental crisis. To this point, the Media Association of Trinidad and Tobago (MATT) noted that “the press is not only a defender of democracy, but also a guardian of our collective future.”

That future must undoubtedly include the use of AI, but the question remains how to manage it ethically and responsibly. According to MIC, regional collaboration is critical to “harness AI’s potential while safeguarding democratic discourse”:

AI is reshaping journalism globally, but its implications are acute in the Caribbean, where media ecosystems face structural vulnerabilities. While AI tools offer opportunities for automated reporting, data analysis, and audience engagement, they also risk deepening existing inequities. Caribbean newsrooms are already strained by shrinking advertising revenues. (It is noteworthy that between 15 and 25% of such revenues are diverted to platforms like Meta and Google Ads). There also exist fragile economies to which we now add the need to grapple with AI-driven content saturation.

Algorithmic curation on social media platforms poses a further threat, “as free, AI-generated content competes with costly, human-produced news.” Will Caribbean newsrooms, with their smaller markets operating within vulnerable island economies, be able to withstand the pressure? President of the MIC Kiran Maharaj suggested, “AI could democratise information access, but without guardrails, it may erode the financial sustainability of Caribbean media. We must advocate for equitable AI governance that prioritises public interest journalism.”

The MIC statement went on to note that “the Caribbean is not immune to AI-fuelled misinformation, which exacerbates social divisions and undermines trust in institutions.”

In 2023, the organisation, in conjunction with the Association for Caribbean Media Workers (ACM), conducted a study about the ways in which disinformation targeted regional elections and public health campaigns. MIC Vice President and ACM Co-Founder Wesley Gibbings suggested that “Caribbean media must adopt AI-driven verification tools and invest in digital literacy programs,” adding, “Our survival depends on retaining public trust through accuracy and transparency.”

Such challenges have been further exacerbated by the region’s susceptibility to natural disasters, whereby hurricanes, floods, earthquakes, and other climate-related events disrupt media operations but put a strain on financial resources. Case in point: the particularly arduous hurricane seasons the Caribbean faced between 2017 and 2024, which the MIC said laid bare “the fragility of media infrastructures and the urgent need for disaster preparedness and resilience planning.”

Whereas “blending traditional knowledge with AI-enhanced forecasting and real-time alerts can be pivotal in disaster response,” misuse of AI via misinformation, etc., can also make such situations worse. The solutions the MIC proposed include policies aimed at taxing tech giants and reinvesting the proceeds into journalism, exploring AI-driven revenue models, and, as per UNESCO’s AI Road Map Policy, establishing a regional AI Ethics Task Force. Such a body, it proposes, would “audit algorithms for bias and promote content verification standards [to] provide needed insight for key regulatory and policy making decisions.”

As far as disaster preparedness goes — a pressing issue for small island developing states (SIDS) like the Caribbean that are at the frontline of the climate crisis — the MIC suggested integrating AI tools into national emergency protocols, including media as first responders, and expanding community media networks by offering training and support.

As the MIC looked towards a future in which AI could be turned “into an ally for press freedom and democratic resilience,” however, MATT placed its focus on the here and now, thanking the media personnel who covered Trinidad and Tobago’s recent general elections “with professionalism, courage, dedication, resilience, and a commitment to truth,” and flagging issues that “threaten journalistic independence and integrity, including political pressures, harassment of media workers, and barriers to accessing public information.”

Current challenges include “the rapid spread of misinformation and disinformation on social media, which undermines public trust and complicates the work of professional journalists.” While the association seeks to advocate for high ethical standards, and foster trust between the media and its audiences, MATT cited an “urgent need to strengthen protections for journalists, ensure transparency in governance, and support open dialogue between media practitioners, policymakers, and the public.”

This article was originally published by Global Voices. License: Creative Commons Attribution 3.0.

World Press Freedom Day: Caribbean media faces new challenges in the age of AI was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

By Kaustubha Kalidindi, Tattle

Convened earlier in March, MisinfoCon India 2025 brought together technology, policy, media, design professionals, and learners over two days to explore and address the growing challenges of misinformation in India. The conference included a two-part WikiCred roundtable discussion that explored the theme “Challenges of Trust in Wisdom of the Crowds.”

The roundtables were aimed at learning how to build trust in community-led crowdsourced initiatives, followed by considering how to protect against maligned attempts to subvert these efforts. The first roundtable opened on the challenges faced by crowdsourced projects, and we closed with the second roundtable where insights into tackling these challenges were delved into in more detail.

This is a brief report on what was discussed in the two WikiCred roundtables at Misinfoncon India 2025.

Roundtable 1: Building trust in community-led crowdsourced initiatives

In the first roundtable, the discussion centred around how contributions are made on Wikipedia and other crowd sourced initiatives such as Open Street Maps (OSM). Round table participants were asked how they deal with challenges emanating out of crowdsourcing, and steps they take to build trust in the process.

Challenges when crowdsourcing

OpenStreetMap as a project relies on crowdsourced data. Generally, on the OSM platform, edits are made back and forth on politically disputed territories. These disputes are resolved by referring to the UN’s official list, and also what people living in the territories believe, although it is difficult to find contributors in these locations.

There are several items (trees, bus stops, benches) that are often vandalised as well. People dispute the existence of places as well- for instance some contributors removed Lake Superior from OSM. Contributors also renamed monuments, made edits on null island and so on.

When such disputes take place, they stated that moderators within the community look into it. The OSM Foundation’s Data Working Group steps in to resolve disputes between the parties through arbitration. The consequences may involve getting banned for a few weeks.

Where contributors try to rename historical locations or monuments, local leaders are involved in making determinations. However, this process often proves tricky since it is difficult to ascertain what is fair and neutral and to make a global broadstroke determination of fairness. For the sillier changes made by contributors, there are validations set in place to protect against them. For verification, they refer to WikiData rankings. Version control is available on OSM as well.

OpenStreetMap, open community membership

OSM has a simple process, where a user only needs to have their email verified. There are no tiers to it. An indicator of the level of experience a contributor has is the number of edits that have been made by them. Any user can see how many edits a user made. They also require declarations if accounts are going to be used as bots.

Tensions with fact checking

With Wikipedia, there were observations that, as a source, the platform is often the subject of vandalisation since the early 2000s. There are now protections against such edits being made on the pages.

Journalists and fact checkers in the audience stated that they do not view Wikipedia as a trusted source in their line of work, and that when they conduct fact checking, the standard practice is to look for at least 2 primary sources and 1 secondary source; in the absence of primary sources they look for 3 secondary sources.

There is also a debate within the fact checking community about contributing to community notes: one school of thought says community notes cannot exist without fact checkers so it is important to contribute to them; the other school believes fact checkers should stop contributing to them so that the public can evidence the lack of quality in their absence.

The fact checking community does not cite Wikipedia as a source, from a certain lens it is viewed as being sloppy, and that not enough due diligence was done. There appear to be clear tensions between community notes as a feature amongst the fact checking community, and a level of distrust in Wikipedia as a source of information.

Roundtable 2: Tackling the challenges of community-source fact checking

The discussion continued on the second day, and focused on vetting — in other words, the reliability of information from different sources that are added to crowdsourced projects.

Wikipedia and information credibility

In explanation of the processes followed in Wikipedia to vet contributions and protect against non-credible contributions, a longstanding Wiki contributor informed us that on a granular level each community has its own set of policies; English Wikipedia having the highest number. On matters relating to contested geographies between nations and regions, matters relating to caste and other similar contentious topics, there are specific rules set in place. Sanctions apply on users or the abovementioned topics to restrict editing and to take action against disputes.

The rules are decided by the community in consensus with each other, and governance is made very clear. If anyone is editing the Wiki of a living person for instance, it has higher standards of editing, because any incorrect or unverified info may affect the reputation, safety and security of the person. Contributions can be made in two ways: either by logging into an account, or anonymously. For the former, only the username is collected, with the latter the user’s IP address is collected; therefore it is preferred that users create accounts. Only administrators may have visibility into the IP addresses, and even that access is given only in the case of investigating any dispute/issue. Accessing them otherwise is considered a serious violation.

Smaller communities on Wikipedia have their own norms. The norms are posted on the Wiki noticeboard to notify new policies. It was stated the initial set up of smaller language communities is difficult because it involves investing resources. The norms may differ from other policies based on the community’s outlook, unless there is a strict rule (for instance, to not use a certain word when writing a wiki). The policies are brought in because they are necessary, and are drafted such that they attempt to be as neutral (i.e. within a certain frame of reference) as possible.

There are difficulties however, such as governance capture- a moderator could formulate regressive policies in smaller/on-mainstream communities. Removing such moderators, who by then have been extensively contributing to Wikipedia, is tricky. Cultural differences also influence how information is documented. Some communities may have a conservative writing style, others may be more open about language and stylistic choices. Wikimedia’s Universal Code of Conduct governs behavioural aspects of engagement and contributions and how to disagree. It does not cover issues such as governance capture. It was emphasised that other contributors in communities must step up and support those who raise such issues because otherwise it can result in a contributor being unfairly banned.

Diversity of contributions is important, it was stated, to protect against motivated actors looking to rewrite facts. Given it is a volunteer based community, it is often difficult to achieve that, and therefore volunteers are composed of a certain audience who have the time and resources to invest into the project. There are instances where contributors become administrators because few run for the elections, and deviate from set processes unless others proactively keep them in check. In the context of OSM, the community tries having a general set of guidelines on organised edits since contributions are sometimes made as part of corporate recreational activities and basic checks are needed to ensure there aren’t any issues. Archiving and documenting violations was noted as a helpful solution to ensure transparency.

What is a “credible source”?

A broader discussion ensued on what constitutes “knowledge.” Are only journal articles credible sources? What of traditional knowledge that isn’t textual? It was acknowledged that it is complicated to navigate, especially given that in India even Gazette notifications are difficult to access. If one were to cite sources at the village level, where textual information is difficult to come by, there may be instances where the last “credible source” may date to the colonial era featuring racial prejudice and exploitative undertones that in present time are known to be inaccurate. Attendees during the round table spoke of issues within academia itself, and how link farming and pay-to-publish journals are paving the way for slanted articles to be published and treated as ‘credible sources’.

However, there aren’t any set ways to identify what a credible source is. Where do blogs stand? Blogs are identified as sources only in exceptional situations. Contributors mentioned there is a ‘common sense’ layer when it comes to considering the credibility of non-traditional sources. Reference was also made to actions taken by traditional news outlets whereby they would post a section of corrections to earlier news articles that had been published and how they signal trust and reliability to a reader from the general public. It was suggested that such measures could be adopted in the online context as well.

Increasing contributions from the public

A simple format open to the general public, editathons are organised for people to see and try out the editing process on Wikipedia. Articles can be pre-selected by organisers to ensure first-time contributors do not have to tackle more complex and sensitive issues. Offering incentives such as certificates brings in more participation as well.

There is a more selective event which is not open to the general public: ‘One Library, One Reference’, which encourages librarians to contribute and add citations as a way of increasing credibility to Wikipedia articles. These events help garner public interest and encourage more contributors to be a part of the project.

Conclusions

Trust within communities that crowdsource is built over time and through process. Even as technical measures are taken to prevent subversions, establishing and documenting processes within communities ensures reliability of information contributed, and open call events ensure continued interest and participation from the public.

Based on the discussion it is clear that well established community projects may face issues of credibility from the larger public despite transparent and thorough processes set in place. There are also complexities arising from socio-political events or beliefs that flow into contributions made by members, and they require additional scrutiny and verification on the part of the communities to resolve such issues.

Crowdsourced communities have made significant contributions to the collection and dissemination of knowledge around the world, and have proven their value many times over the years. Supporting crowdsourced communities and projects in their methods as well as verification process implementation would greatly improve public trust and protect against subversion.

We are grateful to the contributors and stewards from Wikipedia, OpenStreetMap (“OSM”), IsFixable, Channapatna Health Library and Uli who participated in both sessions.

Notes from WikiCred Roundtables at MisinfoCon India 2025 was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

Lessons and a way forward from fact-checking the 2024 election in Indonesia

Written by EngageMedia.

This article was originally published by EngageMedia, a non-profit media, technology, and culture organization. This is an edited version is originally republished by Global Voices. This story is licensed under Creative Commons Attribution 3.0.

In Indonesia, fact-checking initiatives have evolved into a distinct field within journalism practice. While there is abundant research on fact-checking in Indonesia, there is a need to explore the practice from a political economy perspective.

In December 2024, EngageMedia released a new report titled “The Political Economy of Fact-Checking in Indonesia: Understanding the Landscape, Expanding upon Criticism, Overcoming Challenges, and Ensuring Long-Term Sustainability.” The launch of the report in Jakarta, Indonesia, in collaboration with PSHK facilitated a discussion on the challenges faced by fact-checkers, the dynamics of misinformation and the complexities behind it through the lens of political economy, and further analysis on the innovations that can be made to support the sustainability of the fact-checking business model.

The event began with a presentation of key research findings, followed by a panel discussion and a Q&A session with the audience comprising journalists, fact-checkers, university students, civil society organizations, international organizations, and the Ministry of Communications and Information Technology.

Key findings

Report co-author Irma Garnesia raised some key findings related to the relevance of reconstructing the understanding of misinformation on fact-checking and all the challenges and impacts of its work in the Indonesian media landscape through the perspective of political economy. According to them, the fact-checking approach is no longer sufficient if it is still fixated on the dichotomy of “wrong” and “right” to correct individual behaviour because this has the potential to limit civic space and contestation of political discourse. Misinformation must be viewed in a multidimensional manner which, in the context of political propaganda, can become a tool of digital authoritarianism.

Phenomena like the “Gemoy Campaign” during the 2024 General Election illustrated how information muddle wrapped in influence operations — manipulation through social media to shape and influence individual beliefs and voter behavior — covered up a series of problematic deeds of the candidates and successfully overshadowed the 1998 human rights violations case. As a result, disinformation about the dark history of 1998 did not reach the “TikTok generation.” A noteworthy question emerged: is this what can be called “misinformation without disinformation”?

Misinformation is also closely linked to people’s social narratives as commonly held values influence their interpretation of information. In essence, fact-checking has a significant role in building a healthy information ecosystem, including the media’s efforts to capture public attention and perceptions of certain groups or individuals and challenge the narratives circulating about those individuals.

Fact-checking business model in Indonesia and its sustainability

Currently, donor-based business models still dominate fact-checking initiatives in the country. Much of this funding comes from social media platforms, which raises concerns about the sustainability and credibility of fact-checking. These challenges are compounded by the fact that many fact-checkers are freelancers who have changed roles from other journalism topics or beats, and who may be facing limitations in training and capacity development.

Diving deeper into the hard work of fact-checking that has not been truly integrated into the Indonesian journalism and media industry, one participant highlighted the differences in fact-checking training at the central and local levels. The training offered at the central level often fails to address the challenges faced by local journalists, such as social, ethnic and political pressures in their respective regions. The sustainability of fact-checking in Indonesia depends on the infrastructure that supports its practice, be it funding, regulations, or the capacity of the workforce.

Fact-checking, digital authoritarianism, and information governance

In this case, one of the panelists, Sherly Hariestya, brought up the importance of maintaining transparency among fact-checkers, saying that digital authoritarianism does not only originate from the government but also from brand deals and grantmakers supporting fact-checking initiatives. Maintaining credibility is critical since the fact-checking business model still heavily relies on giant enterprises. To uphold their accountability as one of the key pillars in fostering a healthy information ecosystem in Indonesia, fact-checkers could, among other measures, openly demonstrate their methods of processing information and clarify their categorization or nuances of wrong-right categories.

Raden Violla, the second panelist, also emphasized that in the context of elections, information and internet governance in civic spaces is actually complex. From a regulatory standpoint, the Election and Pilkada [local or regional election] Law provides minimal guidance on regulating information throughout the election period. In fact, it is important to recognize that election campaigns and information dissemination are not solely about candidates and the “right” and “wrong” dichotomy, but also about fostering an accountable political education. Especially in the context of Pilkada, the complexity intensifies with the local dynamics that must be considered. In addition, Violla sees that fact-checking initiatives from the elections up to now still focuses on issues that seem to be a public favourite, thus forgetting other crucial issues.

Investigating information credibility and building a healthy information ecosystem has become increasingly challenging with the emergence of social media accounts that appear trusted or verified on social media platforms, but actually have no clear origin. The issue now extends beyond distinguishing “right” from “wrong” to include uncovering who is behind the dissemination of certain information, what kind of narrative patterns are used, and how the public views and interacts with these accounts. Achieving this requires substantial allocation of resources, including people and funds, but it also demands scrutiny of the extent to which certain grantmakers or social media platforms may be steering fact-checkers towards their agendas. Adding to this complexity, Violla also mentioned the existence of information disruption in the context of political elections, seeing that the distribution of information circulated in the public is not purely from each political candidate, but also their sympathizers, such as buzzers, influencers, and others.

Solutions and ideas to build a long-term healthy information ecosystem

In addressing these issues, enforcing regulations that account for these complexities must be done with caution to avoid silencing the public sphere or undermining freedom of expression, which could weaken democracy. Sherly underlined the importance of collaboration among diverse key stakeholders, such as academics and the government, to maintain a safe and inclusive digital civic space, in addition to seeing the issue of digital civic space within the framework of human rights, namely as digital rights. Information governance is now a structural and systematic problem that cannot be solved by solely taking down information that “disturbs the public”. The process of debunking or revealing who considers such information disturbing, who is behind the misinformation, what facts covered by the misinformation, among others, must be carried out. Political will and resource allocation to strengthen the practice of fact-checking beyond “true” or “false” are urgently needed.

The credibility of information that ensures a healthy civic space and democracy is a shared responsibility across various parties, not only fact-checkers, but also the government, social media platforms, and academics, among others.

The political economy of fact-checking was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

An appeal to mis/disinfo researchers and advocates: Focus less on social media. Scrutinize mainstream media more.

Barrett Golding manages Iffy.news and the Iffy Index of Unreliable Sources, a dataset of low-credibility news sites, used by researchers and in media/health guides. Golding regularly contributes to MisinfoCon.

Most research into the spread of misinformation focuses on social media. But trying to push mis/disinfo out of social media is Sisyphean — that rock always rolls back down.

Twitter sacked its integrity team. Facebook sent its fact-checkers packing. Despite researchers’ best efforts, the platforms continue shoving conspiracy fantasies and political propaganda onto our feeds.

For many educators, however, their task feels Sisyphean — despite all of their efforts and overwhelming evidence, millions of people still believe false narratives about elections and vaccines. Many cite Brandolini’s Law, which states that far more energy is required to refute bad information than is needed to produce it.
 — Even Disinformation Experts Don’t Know How to Stop It, New York Times

Here’s why social can’t be fixed:

1. Once BS enters social media, it never gets out — zombie claims live forever.
2. Social media moderation schemes always miss most of the BS.
3. Intervention innovations may work temporarily but eventually succumb to banner blindness.
4. AI has as much chance of worsening as helping the problem.
5. Platforms have repeatedly proved they have no long-term, effective moderation solutions — probably because they are unwilling (for financial reasons) and unable (for tech reasons) to solve the problem.
6. Even if one platform did solve it, it could be bought anytime by a manchild for whom BS is mother’s milk.

A better approach might be preventing the misinfo from ever getting into social media. And evidence from past research points at mainstream-news media (not social) as the initial vector of virality for the tidal waves of toxic bullshit.

“Mainstream news media in fact play a significant and important role in the dissemination of fake news” (Annals of the International Communication Association Volume 44, 2020), concludes one study. “Our results suggest the origins of public misinformedness and polarization are more likely to lie in the content of ordinary news or the avoidance of news altogether as they are in overt fakery” (Science Advances Vol 6, Issue 14 2020), says another.

Having every newsroom publish a QAnon explainer after people turned up at Trump rallies with Q signs and t-shirts was exactly what the Q community had hoped would happen.
— 5 Lessons for Reporting in an Age of Disinformation, First Draft

When O met Q, MSM took the bait.

One example I’ve tracked is QAnon’s propagation through mainstream and social media. For years, most people had never heard of Q. It was just another piece of cyber crap, stagnating in the internet backwaters. That changed in 2020, when mainstream media (MSM) picked it up and flung it around the globe. Soon, nearly every nightly newscast had a Q-citing.

The mainstreaming of Q started in mid-March 2020 with some Q-nonsense that spread socially (details purposely omitted: repetition reinforces the lies). The Q story was an obvious hoax. But MSM couldn’t resist its mix of celebrity and scandal. So, on March 18, we got the first burst of Q-coverage — hundreds of MSM stories:

The conspiracy theory gained steam on Sunday, when one Facebook user posted photos of caution tape surrounding a Mediterranean villa, claiming it was Winfrey’s home in Boca Raton, Fla., and that authorities were “excavating the property and digging up the tunnels.” (Winfrey owns many houses, but none of them are in Florida.)
—Washington Post 2020–03–18

One avid Qanon figure — a Florida man named “Tank” — posted a video that was seen widely on YouTube in which he rambled for about seven minutes while standing in front of a random vacant parking lot in Long Island, claiming that Winfrey’s residence “in Boca Raton, Florida,” was being raided by law enforcement officers.”
—Alternet 2020–03–19

All it took for a fake-news story to go viral was a captive audience and a critical mass of headlines ripped from a dystopian novel.
—Vanity Fair 2020–03–18

The days after that MSM frenzy saw a surge of search and social interest in QAnon. The following graph charts those surges, as relative percentages during 2020 (y-axis) of all Q-related news stories (source: MediaCloud), Google and YouTube searches (source: Google Trends), and tweets (source: Observatory on Social Media).

The MSM stories (3/18) were followed by peaks in QAnon Google searches (3/18–19), YouTube searches (3/22), and tweets (4/12).

I examined other possible causes for those Q-peaks (spreadsheet). None panned out. At the time, cable news and podcasts had hardly ever mentioned Q. No Trump tweet could explain it, nor could any Q-drop (a post by any of the many Qs) or low-credibility news source story.

The only factor I found that correlated was MSM’s over-coverage. And that pattern repeated several times in 2020:

• MSM obsesses over an inconsequential bit of Q-news.
• Within days QAnon peaks again in search-engine trends.
• Within weeks there’s another Q-surge in social media.
• (Plus a subsequent rise in Q-related crime and QAnonCasualties — a subreddit for folk with “a friend or loved one taken in by QAnon.”)

In sheer numerical terms, the information to which voters were exposed during the election campaign was overwhelmingly produced not by fake news sites or even by alt-right media sources, but by household names like The New York Times, The Washington Post, and CNN. Without discounting the role played by malicious Russian hackers and naïve tech executives, we believe that fixing the information ecosystem is at least as much about improving the real news as it about stopping the fake stuff.
—Don’t blame the election on fake news. Blame it on the media., Columbia Journalism Review

Suppose MSM is often the initial seed that propagates misinfo. That means researchers could help by studying ways MSM can report on misinfo without spreading it. My Iffy.news post, “Mainstream media spreads fake news,” has a few suggestions.

Social media can’t be fixed but MSM might be.

As I said above, decades of social-media misinformation show that platforms either won’t or can’t fix their problem. Focusing a little more research on news media and a little less on social feeds may have far more real-world impact. News-media creators are much more likely to accept and act on the findings than social platforms will ever be.

Sisyphean Social Media Research was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

A Q&A with Ilana Strauss, founder of News Detective, a project incubated by Hacks/Hackers.

By Nevin Thompson, Hacks/Hackers

News Detective is a crowdsourced fact checking project that aims to fight misinformation at scale by increasing the pool of fact checkers, and in turn make fact checking more efficient. News Detective is currently developing a Reddit bot for an 80,000-person climate change online community and is in early discussions with BlueSky, a 5 million-user social media platform interested in using News Detective for fact checking on their platform. The startup, founded by Ilana Strauss, has received incubation support from MIT DesignX and Hacks/Hackers.

Q: Tell us about News Detective

Ilana Strauss: News Detective uses the “wisdom of crowds” to fact check the internet and increase media literacy across different online communities. Any News Detective user can submit a post from a social media platform for fact checking, and platforms can link back from the social media posts to the fact checks

Styled on Wikipedia’s hierarchy, News Detective empowers the “crowd” to fact check claims and stories, but professional fact checkers also moderate and provide oversight. News Detective partners with social media platforms who need fact checking, and college and university students, who learn media literacy.

The students fact check social media posts, providing evidence and sources, and a professional fact checker moderates their fact checking. Successful fact checks appear directly on social media posts. Students simultaneously learn media literacy and bring reality to the internet at a scale that the small number of professional fact checkers in the world could never achieve.

Q: How has the recent explosion of new AI technologies affected News Detective?

Ilana Strauss: Humans are no longer the whole story when it comes to fact-checking. Since the launch and rapid adoption of generative AI tools such as ChatGPT in late 2022, as a founder I’ve really had to rethink the role of AI in News Detective.

As recently as a couple of years ago, I assumed AI had no place in fact checking. After all, any AI would necessarily take on the biases of its programmers, was my assumption. I also thought that fact checking is about finding sources, not spitting out information based on the likely biased information an algorithm has been told to examine.

However, after thinking about and working with fact checking as I build News Detective, I’ve come to realize that AI likely has a role in this work… it’s just that fact checking is not AI’s defining role.

It’s true that AI can read more sources in an hour than a human can in a year. But only humans can decide what these sources mean and ultimately make a judgment call. At least, that’s my theory at the moment.

I would suggest the American news cycle has been in freefall ever since the internet burst onto the scene thirty years ago and unleashed its content-creation powers upon us. And yet, despite all of the concerns about misinformation since the 2016 election, we may remember these last few years as “the good old days,” thanks to generative AI.

Q: How are new AI technologies affecting journalism and online credibility during an election year?

Ilana Strauss: The upcoming fall 2024 American presidential election won’t merely have the regular amount of spin and misinformation that we have come to expect during presidential elections. The big guns have arrived: generative artificial intelligence is here.

Artificial intelligence can be used to create a practically limitless number of articles, photos, and videos to try and sway voters. In 2023, a deepfake video showing Joe Biden making transphobic comments circulated. AI-doctored images of Donald Trump giving Dr. Anthony Fauci a hug also made the rounds. These events give us a taste of what is to come. Democracies rely on citizens having access to truth so that they can vote accordingly.

My question is: How will our democracy react to what will surely be an avalanche of fake articles, photos, and videos?

Q: In an era of AI, will improving media literacy in turn help credibility online?

Ilana Strauss: A few years ago, I heard people often talking about how news consumers would simply believe whatever they saw. But every action has a reaction, and news consumers of today are not news consumers of years past. If yesterday’s public used to believe whatever they saw in print and in video, today’s public increasingly doubts what they see with their own eyes.

From my experience building a community of more than 1,000 volunteer fact checkers at News Detective, I wish I could say that means people are becoming more discerning and learning media literacy skills to help them distinguish between fact and fiction.

But from what I’ve observed, something quite different seems to be happening. Because people know anything they see could be misinformation, they simply believe what they like and chalk up the rest to “fake news.” AI will likely accelerate this trend.

Since any information that counters someone’s worldview will automatically be viewed as “fake,” people will be less likely to take in information that makes them think critically and challenge their worldviews. Those on the right will likely reject left-leaning information as misinformation, and vice versa. The news-consuming public will become increasingly partisan and close-minded.

Not that this trend is endless or inevitable. Many have been trying to find solutions to this possible dark future.

Q: What’s next for News Detective?

Ilana Strauss: New Detective has recently partnered with two more universities, and we’ve also received a grant from the International Fact-Checking Network (IFCN), which is funded by Google and YouTube. We’re also excited to partner on a fact checking initiative with BlueSky, the newish 5 million user social media platform founded by a Twitter founder.

And we will be amping up just in time for the upcoming election, in which AI will undoubtedly have a massive role. News Detective community fact checkers will likely encounter claims spread by artificial intelligence. A multitude of other solutions — X’s Community Notes, for instance — will also be tested.

A few questions I have include:

Will humans be able to take on the machines, or at least use good machines against bad ones?

Will AI turn us all into critical-minded media-literate detectives or cynical partisans?

As AI evolves, how will news consumers change their behavior?

Hopefully, we will start to get the answers to some of these questions, and a solution will emerge. Our democracy may depend on it.

Based on Wikipedia’s decentralized, grassroots model, News Detective’s crowdsourced fact-checking community addresses misinformation at scale by increasing the pool of fact checkers and making fact checking more efficient. Prior to receiving Hacks/Hackers incubation support over the course of the grant period, News Detective, with support from MIT DesignX and MIT Sandbox, built a test product with a community of 1,000+ volunteer fact checkers.

“Brace yourself. The first AI elections are here.” was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

How do we, in practice, do the work around de-escalation, such as trying not to judge another’s anger or fear?

This blog post originally appeared in ARTT’s Trust Issues newsletter on October 13, 2023. Sign up for the ARTT email newsletter to get updates about the project twice a month.

By Connie Moon Sehat, ARTT Project Principal Investigator

Election workers in the U.S. are apparently quitting in droves. In September 2023, Issue One, a cross-partisan organization, reported that roughly 40 percent of chief local election officials across eleven western states have resigned since the November 2020 election. That’s a lot of loss, if you consider the institutional knowledge and experience they carried with them.

What struck me about this news was how emotional and intense election work seems to have become. Election workers speak of threats of violence, hateful messages, and harassment both in-person and online. Stories of battles with constituents and others seem commonplace and have produced a tremendous sense of fatigue and burnout.

Take the sentiments of the clerk-recorder of Lyon County, Nevada: Despite working for more than 20 years at this position, Nikki Bryan decided not to re-run for election. “I can’t fix the anger,” Bryan told the Christian Science Monitor last year, “I’ve tried.”

The idea of “fixing the anger,” or at least lowering the temperature to reduce stress and burnout among election workers, got me thinking about de-escalation, which is one of the nine response modes we focus on at the ARTT project. Research around de-escalation has been an area of critical investigation in several different fields. Think of situations where there is a high-stakes standoff with someone who is threatening others: conflict resolution is clearly one field. Within health, practitioners in nursing and in psychology have long explored the practice but this also happens within policing, sociology, and in business research (consider the times you may have experienced de-escalation after you have expressed dissatisfaction as a customer).

Across different fields, de-escalation methods do bear a lot in common. De-escalation methods can for example lean on other response modes such as empathize and listen, but also employ some other points such as personalizing or humanizing the exchange.

Two things struck me when thinking about de-escalation. First, there isn’t a lot of literature, yet, about how to de-escalate on social media — most of the research is about face-to-face interactions.

Second, the knowledge collected in these studies doesn’t really go towards answering a couple of harder questions around how this strategy might work for elections workers — and citizens — in practice.

Methods can explain how de-escalation might be accomplished, and some studies explore whether or not it’s effective (measurement around this is a hard thing to do). But what about the question of how do we, in practice, do the work around de-escalation, such as trying not to judge another’s anger or fear? And there is the question of why we would feel motivated to take on the responsibility and effort of lowering the temperature — especially hard to do when someone is literally or virtually screaming in your face.

There are a few answers here that may differ depending, for example, on whether there’s bad faith. But in terms of both how and why we might want to add de-escalation to our toolkits, I think that one answer is clear: Methods that help us introduce ourselves to one another, try to imagine each other’s circumstance, and bring our humanity to the exchange are reminders of our real need for one another. Each person belongs to communities, regions, and countries that we care about, and we have an effect on each other.

What do you think?

If there are examples of conflict you want to share from your communities that you are puzzling over or you have your own thoughts on this, let us know.

De-escalation: Easy in Theory. In Practice? Kinda Hard. was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.

How can we turn Wikimedia into a tool for evaluating the credibility of news sites?

A short update from the Iffy Index of Unreliable Sources, a dataset of low-credibility sites aimed at making Wikidata/Wikipedia a more effective indicator of news-site credibility. The Iffy Index is an ongoing project supported in part by a WikiCred microgrant in 2023.

There is no quick and easy way to tell if a news site is credible based on its Wikipedia article or Wikidata item — nothing much has changed since my last update a year ago. However, as I also mentioned back then, there are ways to address the issue. So, last March, with support from WikiCred, I started a project to turn Wikimedia into a tool for evaluating the credibility of news sites.

I thought it would take a month. Now, almost a year later, I’m still hacking away. But the light at the tunnel’s end is within sight.

Matching news-media databases to news media in Wikidata

My plan was to:

1) Grab credibility indicators from news-site databases, such as press-association memberships, reliability ratings (e.g., from Media Bias/Fact Check), and street addresses (i.e., not a fake-local “pink-slime” site).

2) Insert these credibility indicators into Wikidata, the “central storage for the structured data of its Wikimedia sister projects” (Wikipedia, Wikiquotes, Wiktionary, etc.).

The problem I discovered with this plan was that the same newspaper often had different names in different databases and in Wikidata. Matching the news outlet to all its database entries to its Wikidata page was laborious. In January, though, I finally got my data ducks in a row — actually 15,000 rows, each with a news outlet’s domain name and Wikidata identifier, called a QID.

With that done, I added press association membership data for about 10,000 Wikidata items. Next I’ll update their URLs (many items list the old timey “http” or no URL at all), and standardize their Wikidata descriptions to include their city and state (some just say “newspaper”). I’ll also create about 5,000 new items for news sources not yet in Wikidata.

Releasing Wikidata into the wild

This project deals mostly with English-language news media in the USA. Others may want to do the same for other countries and languages. So I’m documenting the tech process in a GitHub repo. The repository links to the tools used and lists the resources produced by this project, such as lists of newspapers that have Wikipedia articles and press association member directories, and a visualization of Wikidata’s news-media categories.

Once the credibility indicators are put into Wikidata, the next step will be to demonstrate how people working in news-source reliability can pull credibility data out of Wikidata.

My demo will be inside a custom search engine that returns results from 3-dozen fact-checking sites called the Iffy.news Fact-check Search tool. If a user searches for a domain name, the results will include core Wikidata. News sites will display credibility indicators:

A site in Wikidata without credibility indicators would look less inviting:

Coming next: the Spiffy News Index of Reliable Sources

My existing Iffy Index of Unreliable Sources is a list of low-credibility news sites. With the completed Wiki Media Cred dataset of domain names, QIDSs, and credibility data, I’ll soon be able to produce a companion list of credible sites that I’ll call the “Spiffy News Index of Reliable Sources.”

To be considered reliable, the news source must be one of these:

1. A press association member (including state press associations, INN, LION, NNPA, and trust.txt).
2. Listed in a vetted index of legitimate news sources (e.g., the News Deserts and Project Oasis databases).
3. Rated by Media Bias/Fact Check as High-credibility.

Like the Iffy Index, this “Spiffy” list certainly won’t be complete. But every listed item will have a high-degree of certainty of being generally reliable.

My hope is that Spiffy News proves as valuable as Iffy.news has been for mis/disinfo researchers and media/health guides.

From “iffy” news to “spiffy” sources: Creating a comprehensive list of credible news sources was originally published in MisinfoCon on Medium, where people are continuing the conversation by highlighting and responding to this story.