Far-Right Facebook Networks with Raphael Hernandes

How do ordinary online communities become breeding grounds for radicalization? In this episode, Eleanor Drage speaks with Raphael Hernandes, an AI ethics researcher, award-winning data journalist, and Cambridge PhD researcher, about social media, online radicalization, far-right rhetoric, and the growing challenges of holding technology platforms accountable.

The conversation explores Raphael’s investigation for The Guardian into 16 interconnected Facebook groups with approximately 600,000 members, revealing how everyday online communities can normalize anti-immigrant sentiment, nativism, conspiracy theories, and misinformation. Drawing on more than 100,000 captured posts, Raphael explains how group administrators and moderators shape what circulates online, how coded language allows harmful rhetoric to evade moderation, and how repeated exposure to these ideas can gradually lower the threshold for radicalization. He also examines the growing barriers to accessing social media data, as platforms increasingly restrict independent research and auditing. The episode asks who gets to decide what circulates online, how technology companies can be held accountable, and why transparency and public interest research are essential to understanding digital communication.

Raphael Hernandes is an AI ethics researcher, award-winning data journalist, and PhD researcher at the University of Cambridge. His journalism career spans more than a decade, including work at The Guardian and Folha de São Paulo in Brazil, where he led data and AI-driven newsroom projects. His research and reporting examine the intersections of technology, journalism, power, and society, with a particular focus on how digital platforms shape public knowledge and communication. His investigation into everyday Facebook networks where far-right ideas grow was recognized as a 2026 European Press Prize Laureate.

Reading List:

• Raphael’s website

• Inside the everyday Facebook networks where far-right ideas grow

• Reporting on the systems that shape what people know

• Report Launch: Data Not Found

• AI, journalism, and critical AI literacy: exploring journalists’ perspectives on AI and responsible reporting

• Auditing Google's Search Algorithm: Measuring News Diversity Across Brazil, the UK, and the US

• Artificial Intelligence in Brazilian News: A Mixed-Methods Analysis

• LLMs left, right, and center: Assessing GPT's capabilities to label political bias from web domains

• Using GPT to Assess the Political Bias of News Outlets

Transcript:

Eleanor Drage (00:53): 

I'm thrilled to be here today with Raphael Hernandes, who is an AI ethics researcher and award-winning journalist specializing in the intersections between AI, journalism, and society. He is now doing a PhD at Cambridge, but his journalism career has spanned more than a decade, including roles at The Guardian and Folha de São Paulo in Brazil, where he led data, and AI-driven newsroom projects and reported on technology, power, and the systems that shape what people know.

We'll be talking today about his work as part of The Guardian investigation into the everyday Facebook networks where far-right ideas grow, which was named a 2026 European Press Prize Laureate. Huge congratulations and welcome, Raphael.

Raphael Hernandes (01:48): 

Thank you very much, and thank you very much for having me. As you know, I've been a longtime listener and fan of the show, so I'm glad to be a part of it now.

Eleanor Drage (01:57): 

Well, go on then. So what brings you to the topic of feminism, justice, and technology?

Raphael Hernandes (02:03): 

So yeah, as you mentioned, I come to this from my journalism background, right? And now I've turned into an academic, but I never really stopped being a journalist, so everything is intertwined. My routine was inside a newsroom watching tech platforms and their impacts, and then I wanted to study it rather than just keep living it, right?

In terms of feminism, I'm not a feminist theorist, but there's one feminist idea I lean on, especially as a researcher. It comes from Linda Steiner, and she makes it for journalism specifically. She raised the point that we are trained to write as if we are nobody, no position, no stake, just stating the facts, right?

She argues that this is a weaker standpoint, not a stronger one, because you always know things from somewhere. So saying where you're standing lets people check you. Pretending that you're sitting nowhere just hides it. And I think that's particularly useful for my position as a journalist-turned-researcher; that dual position means I live, or have lived through, the things that I write about and analyze. This then becomes a resource rather than an undeclared bias.

Eleanor Drage (03:26): 

And this is The Good Robot. So what is good technology? Is it even possible? And how do we work towards it? Given that we'll be talking today about communication online and radicalization, can you tell us something about communication online and what makes a technology good?

Raphael Hernandes (03:47): 

Brin, so yeah, I think I might be too much of a journalist in that because I'm always worried about using adjectives. So I don't like the idea of "good" as a property of an object. I think it's way more complex than that. The same system can be good or bad depending on what it makes visible and to whom, right?

I tend to avoid both the reflexive debunking of "Yeah, this is problematic. This is what is behind the curtain," but at the same time avoid the hype and trying to think of the things that a simple software update or a simple change cannot fix. When we think of communication, I try to see these as technologies that mediate knowledge, that mediate what people know. And I think it's better when it makes that process legible, right?

You can tell a choice was made, you can roughly tell who made it, and there's somewhere to go to contest it. What worries me isn't necessarily systems doing bad things; it's consequential systems that have no surface, no place, no person, nothing to argue with.

There's another test that I keep applying: What does this system actually solve versus the problem it creates? It's the whole idea of creating a hammer and then going looking for nails to bang against. I think that lots of recent technology falls into that trap. There's no clear line of the problem it solves, which doesn't automatically make it bad, but it means that it needs to make a case for why it was made. And that's mostly not necessarily a technical question, right?

Facebook groups, for example, are a very old technology. They're boring. But what made them important and consequential is the moderation policy around them, the amplification of discourses, and decisions on what counts as violating content policies or not.

Large language models are similar to that. Models can be very useful for their flexibility when you label data, for example, classifying a bunch of social media posts, or identifying and tagging images, which is something that I've done in multiple investigations, both as an academic and as a journalist. But what makes it good is not the technology itself. It wasn't even created for that kind of application. It's that it sits inside a rigorous process that gives me some level of trust in the output, ways of validating what comes out, and not just trusting whatever garbage that might or might not come out.

Eleanor Drage (06:35): 

It's definitely not a coincidence that we do get a lot of journalists that come on the show and say good technology is auditable. And I think that's kind of a brilliant response because we need to know when something goes wrong. And that's very difficult to do if the process is obscure, the algorithms are impenetrable, or the content is hidden behind a paywall, or whatever it is. We need to know, and it shouldn't just be journalists auditing. I mean, this is kind of a last resort, really, of journalists getting together with data experts and companies and coming out with the truth.

Anyway, I love that response, so thank you. I really want to get on, because we have so much to talk about today, including a piece that you wrote that we talked about briefly in the introduction. This was part of a project that The Guardian was rolling out: they spent a year studying an online community trading in anti-immigration sentiment and misinformation with the goal of finding out how online spaces can play a role in radicalization.

So tell us about what you and your team found looking inside the Facebook networks where far-right ideas grow. What did we not know before, and what do we know now?

Raphael Hernandes (08:00): 

Yeah, and turning back to your previous comment, yes, it was a huge team effort, because doing this kind of thing is getting harder and harder. So a huge shout-out to the whole team, editors, other investigative reporters, graphic designers: Pamela Duncan, Elena Baresi, Pablo Gutierrez, Gary Blight, Robin Vinter, Lydia McMillan, Aaron Sharp, and other people who were involved with the project. That goes to show how difficult it is to see behind the curtain of these things.

After the 2024 riots here in the UK, some people were jailed over what they posted online. Over 1,000 people were charged in the context of the riots, roughly two dozen for online offenses. Facebook was the platform that several of those people had in common. So that's where we started this digging, trying to understand what kind of communities these people were in, what circulates there, and what kind of things these people were seeing that normalized the sort of behavior of posting the nasty things that made them get charged.

From the public profiles of the people we could identify in these charts, which are all public, we found a couple of Facebook groups, and then we mapped the content that was circulating in those groups. We found that to be a huge network of 16 groups with around 600,000 members (that includes some double-counting because we don't have access to the full list of members).

Over a year, we captured over 100,000 posts and analyzed over 50,000 text-only posts to categorize them. Roughly, we found not an organized far-right movement, but an ordinary ecosystem run largely by retirement-age admins on the world's biggest platform, where you can see anti-immigrant sentiment, nativism, conspiracy theories, and misinformation circulating continuously, which experts we interviewed said helped create the conditions for radicalization.

An important caveat here: I did that under my hat as a data journalist, so I'm not a radicalization expert myself. That side of the analysis comes from experts who make a living studying that.

One particular striking thing was that even after we analyzed all that and saw what was going on there, we sent the groups back to Meta, the company that owns Facebook. They reviewed the groups and said that none of the content violated the hateful conduct policy, so there was nothing that they could do. I think that kind of breaks the assumption that this kind of content lives on fringe platforms like 4chan, Telegram, Parler, or Gab. It's on mainstream platforms, with the person who could be your neighbor, right? It's not as fringe as we once thought.

Eleanor Drage (11:11): 

Yeah, I mean, I've heard people I would have thought knew better talk about white people in the UK as natives, as indigenous, which I consistently find extremely shocking, even though it's now more common than I would have hoped.

As part of this article, you traced five individuals who were members of three groups, which together had this quite substantial membership of about 267,000. And those groups all have titles that celebrated figures on the right; maybe you could say a bit about who those figures were. They also have overlapping members and contain very similar posts, including hundreds expressing support for those arrested in connection with the summer riots that you just spoke about.

Now, as a data journalist, you are excellent at tracing the networks used by individuals who propagate far-right ideas. So I know you can do the technical thing, but you're also brilliant at the social analysis, which is an amazing set of skills to have. So can you just describe to us both the technical approach that you used in a way that I'll understand in creating that network chart, and the significance of what you found to you?

Raphael Hernandes (12:26): 

On the technical side, as you say, we started with a starting point, which was these people, and using open-source investigation techniques, basically beating the Facebook search with technical ways until we found something that stuck, we identified three groups where these people were either current members, past members, or had posted.

What we did then was sort of snowball from there. In one of the groups, after posts were extracted, lots of other groups were posted, and from those groups we started seeing some mutual characteristics: either posts that were shared verbatim across multiple of these groups, or support for people who were arrested.

More meaningfully, 16 of them shared at least one admin or moderator. These are the people who set the rules for the groups, can ban people, delete posts, and they basically run them voluntarily. The groups had other similar characteristics. These admins and moderators are mostly retirement age, and that was deliberate; you could see posts asking for people who were retirees and had time to spare to help them run the groups. Most of them, if not all of them, had titles that were celebrating some far-right figures, like "Knight of Harag," whose name appeared in the largest of the groups, along with other journalists and influential figures within the far-right.

One thing that was interesting there is that this connective tissue is the people, not necessarily the algorithm. The algorithm helps promote the content and brings some of these things there, but you had 40-odd admins and moderators scattered across England from different socioeconomic backgrounds, and they were actually doing the editorial layer of gatekeeping that community without calling it that.

Another interesting thing is that the language is tuned to the platform. Facebook itself has some moderating criteria that they use, rules of engagement that dictate whether a post should be taken down. That means overt racism is kind of rare, which is what you would expect because there is some automated level of monitoring. Even the mods would say stuff like, "You cannot post this, otherwise they are going to take the group down." But that doesn't mean they can't use some sort of coded language, either calling people "illegals," calling immigrants coming here an "invasion," or even less hidden things like calling them "lice." It's not completely out in the open.

Also, and this is partially my read from seeing this sort of community in the past, having 16 groups is a way of avoiding moderation. Even if the platform decides it's not something they want in there, you have 15 other options of groups that have already taken off for you to switch to.

One important caveat here: membership does not signal anything about a person. Being a member of these groups does not mean that you are a racist, does not mean that you are far-right, and does not even mean that you agree with those ideas, which one of the moderators that we interviewed actually denied agreeing with. But it is a fact that you are exposed to that kind of thinking.

Eleanor Drage (16:14): 

And you're agreeing to be exposed to it, I guess. It's a bit like, I always get upset with my parents for reading newspapers. They propagate bad ideas, and many of the ideas that are in any given newspaper, I will take to my mom and be like, "What is this that you're reading?" She'll be like, "I don't agree with that." But she's still agreeing to be exposed to it by subscribing. Is that the idea? What does it mean to be a member of those groups?

Raphael Hernandes (16:47): 

The discourse often crosses into radical rhetoric. If you think of how news media is thought of, the diversity of ideas is actually a good thing, being exposed to contrasting viewpoints. But there is a line that needs to be drawn between a legitimate concern and radical content, and that's the line that was often crossed there. Again, not necessarily by everyone, but by being constantly exposed to that...

Lots of people were there because they had real grievances. They were complaining about the cost of living, the decline in high streets, or feeling like they are not being listened to by their governments. That is genuine, and that is not extremism. But at some point, the line starts lower for those people as they are exposed to this kind of concept.

What we counted there, for example, shows that two in five posts criticize the government and figures connected to the government. What matters here is not the criticism; we should be able to criticize our governments; that's not what I mean. It's the substance. So you're calling the police "two-tiered," you're calling the media "controlled," you're calling the Royal National Lifeboat Institution a "taxi service for illegals." These are all emotionally charged claims; they are unsubstantiated, and they are constant. Two in five posts contain that. So it's not necessarily a critique; it's extremist speech without any substance.

Also, one in seven posts touched on immigration, and one in ten of those were demeaning, dehumanizing people, being outright racist. It's fine for you to disagree with immigration policy, to want better border controls, to be disillusioned about the lack of jobs or job opportunities. But pointing that towards another group and assigning the blame to them without really understanding what's going on, that's where it becomes a problem.

The same with nativism, which was present in one in 25 posts. A smaller part of the posts were conspiracy theories. That connects to what some of the experts we interviewed say: these are often harder to get behind because some of them are really, really far-fetched. That kind of explains the fact that they are smaller, but that doesn't mean they're not there. Being exposed to this kind of thinking lowers the bar over time.

Eleanor Drage (19:29): 

And what we're seeing, I think, with the rise of fascism globally, is that people are prepared to create allegiances between themselves and far-right thinkers even though they don't agree on everything, even though they might think, "A lot of this stuff is too much," or, "I'd never set fire to a hotel housing asylum seekers." They are still willing to be members or to listen.

And that I find very interesting for everybody else. We must know what the far-right's way of thinking is so that we don't slip into it ourselves, and we can see how easy it is actually to move from one seemingly tolerable view to something that is completely racist.

So one of the moderators for the groups that you were looking at said that they remove extreme content, but you found openly racist posts in there, the assumption being that they didn't believe those posts were extreme. What starts as criticizing the government, as we've seen, can without great difficulty cross into radical rhetoric. So can you tell us how this happens? And how are these seemingly benign conversations on Facebook slipping into broader far-right ideology?

Raphael Hernandes (20:48): 

The volume of the posts is on institutions, and the targeted venom is on immigration. Then we have conspiracies as the smallest slice. That order mattered, because the conspiracies there are not the entry point; they arrive last, at a point where some agreement is already established within the group, and that's how it lands.

The tail of something crossing into this radical line of thinking is not necessarily the target of the criticism; it's whether it can or cannot be checked. Legitimate critique names something that you should be able to show is false. The radical version that you often see there is something that very conveniently explains everything that is going on; it's unfalsifiable. Everything is in this mess because of immigrants. It's that shallow, and that's what we often see in these groups.

Professor Sander van der Linden here from Cambridge as well, who is one of the sources that helped us analyze this content, qualified that as a known tactic, historically a fascist tactic, because you have an informed citizenry as an obstacle. People sharing this content and engaging with it don't necessarily know that they are running a playbook where the narrative is there, and then what you find in practice is politicians trying to leverage that space and make a platform for themselves.

Eleanor Drage (22:42): 

I'm interested in this "we're not far-right, we're just right" mantra propagated by lots of people who possibly wouldn't say that they're radicalized or have any radical views. Many of the people, for example, who participated in the Lucy Connolly riots, which was when a woman was jailed last year for calling people to set fire to hotels housing asylum seekers, many of those people didn't view themselves as far-right, racist, or even anti-immigration. You know, the classic "I'm not racist, but..." The majority won't be part of any far-right organization or group at all, you said.

So how did you see that play out in the network? Were those people visible, or are they sort of invisible? Do they lurk in groups without joining?

Raphael Hernandes (23:35): 

This discussion of being far-right or not far-right was a recurring topic. Actually, one of the most popular posts, I can't recall whether it was the most commented on or the most shared one, was saying something like, "We are not far-right, we are just British," or something like that. You would see these discussions often, with two things happening at once: some rejecting the term as a slur, trying to silence them, and others trying to embrace it and redefine it, saying something like, "If being pro-my-country, pro-Britain makes me far-right, well, I am far-right then, I guess." So this was a bit of a contested point.

Still, we were not talking about an openly far-right group. I think the label itself is a bad instrument to measure what is going on there, because we could sit here forever discussing whether this qualifies as far-right or doesn't. That is an important and legitimate discussion, but in the situation we are in, and the kind of analysis we're trying to do, that in the end will get us basically nowhere.

That's why I was aiming at posts and trying to understand what circulates there, trying to actually describe what is going on. We were counting posts instead of trying to profile people for this project.

Which again brings me back to a previous point: access to data for this is a huge problem now. In order to be able to do this kind of analysis, you need the posts; that's a given. To analyze posts, I need posts. And this is getting harder, very fast. I've been doing that for over a decade now, and I've seen how the scenario has deteriorated.

I recently worked on a report called Data Not Found, which analyzed more than a dozen social media and tech platforms in terms of the data that they made available for researchers in different jurisdictions. The results were appalling, because I don't think any of them passed with flying colors. That's social media data that you could extract easily from Facebook, easily from Twitter, less than 10 years ago. Now it's been withdrawn, and the access changes whenever the platform decides it changes.

This kind of investigation that I did is already close to being unrepeatable, and nothing forced that except that Facebook decides to change the rules that they enforce, and that's the pattern across the whole industry.

Eleanor Drage (26:46): 

And good technology is auditable. We hear this critique from researchers all the time: that they can't analyze social media networks anymore. These platforms are designed to be increasingly unavailable to researchers and auditors. You can't have access to an API; that's exclusively the domain of companies that they choose to partner with. It's outrageous. And I think that should be the target of regulation, to make the backend accessible, to make posts permanent and public so that they can be audited.

We're so grateful for the work of journalists like you and people who are willing to do that labor, because it takes ages and you have to be unremittingly optimistic about finding something that you can communicate to the public. Huge thanks for having you on the show, and I'm excited to have you back to talk about your PhD and all the things that you're finding through that. Thanks so much for joining us, Raphael.

Raphael Hernandes (27:57): 

Well, thank you very much for having me.