Combating hate and scaling revenue: The evolution of publisher social tools
‘For us, moderation is one component of a much larger picture. The real question is how technology can help publishers produce, distribute, moderate and monetise social content much more efficiently.’
– Henning Rosenstengel, co-founder and CEO: MainSent; managing director: Ippen Digital social media business unit
For the better part of the past decade, most publishers have treated community management as a liability, necessary because of the number of hate comments. Henning Rosenstengel, managing director of Ippen Digital’s social media business unit, thinks this “defensive activity” is a huge misconception.
“Our experience suggests the opposite: Good community management is a growth function. When you consistently remove hate and abuse, you create space for the much larger group of normal users who actually want to have a conversation. Those users start participating more. That creates more engagement, more distribution and ultimately more monetisation opportunities,” he explains.
Rosenstengel co-founded social media agency MainSent and developed MediaFox, the social media operating platform that has housed Ippen’s digital portfolio since 2021 – and where they’ve applied this principle to automation across operations.
“MediaFox was never meant to be just a tool for deleting comments. We use automation across content and social media operations, and from 2025 to 2026, those processes helped us increase Ippen Digital’s social media revenues roughly sixfold,” he explains.
In this interview, Rosenstengel shares more about the mechanics behind that growth, and why the next generation of tools will need to go further than policing tone…
Is your headline number – ‘more than 100,000 hate comments deleted daily’ – a daily average, a peak figure, or a running total across every Ippen title on MediaFox?
It depends heavily on what is happening in society at a particular moment. Ahead of political elections, for example, we typically see a very significant increase in both the overall volume of comments and the number of problematic comments, and there are weeks when that average rises to around 100,000 comments per day.
On average, we remove about 35,000 comments per day across Ippen Digital’s portfolio; these figures do not include the other news publishers MediaFox supports.
Explain the process: is a comment auto-deleted, flagged for human review or both?
Both.
Some publishers choose a fully automated approach, where MediaFox automatically removes comments based on the publisher’s rules and our detection systems.
For most publishers, however, we recommend a hybrid model, as more ambiguous comments should be reviewed by a human.
What’s your working definition of a ‘hate comment’ for the purposes of that count?
First of all, MediaFox does more than identify hate speech. We also work on detecting misinformation and fake news, including through cooperation with major news agencies and specialist partners in this field.
When we talk specifically about hate comments, our focus is primarily on content that crosses clearly defined legal and editorial boundaries.
One important point is that these rules cannot simply be identical everywhere. MediaFox can adapt its moderation logic to the publisher, its editorial standards and the legal environment of the country in which that publisher operates.
At the same time, we are very conscious of the distinction between something that is offensive or unpleasant and something that should actually be removed. We do not want to remove legitimate opinions simply because they are controversial.
What’s your false-positive rate, and how do you measure it?
Freedom of expression is extremely important to us, which is why our objective is to keep the false-positive rate as close to zero as possible.
There is an intentional trade-off here. In a fully automated setup, we would rather leave some borderline comments online than automatically remove legitimate speech. In cases of doubt, we decide in favour of the user.
That is also why we generally recommend the hybrid approach: The goal is not simply to maximise the number of comments we delete; the goal is to remove the right comments while protecting legitimate debate.
Is there any distinguishing between coordinated or inauthentic activity – bot networks, troll farms – and an individual reader who is just being hostile?
Absolutely. That is an important part of MediaFox.
We do not only analyse the individual comment. We also look at signals surrounding the account and its behaviour. This allows us to identify patterns that may indicate coordinated or inauthentic activity rather than simply an individual user expressing an aggressive opinion.
Where appropriate, publishers can then take action against those accounts as well.
However, we deliberately maintain a higher threshold for account-level actions. We do not offer fully automated account blocking without human oversight. A human reviews those cases before such a decision is made.
That distinction is important because removing an individual comment and effectively excluding a user from a community are two very different levels of intervention.
Is the tool built to produce the compliance reporting regulators now require?
MediaFox helps publishers operationalise moderation at a very large scale and structure and document those processes. But we do not view community management purely as a compliance exercise.
Our data shows that proactive moderation can have a direct business impact. If you only think about moderation as removing illegal content, you miss half of the story. A well-managed comment section encourages more legitimate users to participate. That increases engagement, which can increase reach, and ultimately that has economic value for the publisher.
So compliance is one dimension. Building a healthy, engaged and economically sustainable community is the bigger objective.
LLMs make it cheaper to generate fake news and fluent hate comments at scale. Is detection keeping pace with that, or is this an arms race you’re currently behind on?
When it comes to classic hate comments, I am relatively confident. Detection technology is already very strong, and we have years of experience and data in that field.
Fake news and disinformation are more challenging – absolutely, an arms race, with detection technology continually adapting to a new generation of models.
The biggest change is scale – generative AI makes it extremely cheap to create convincing text, images and increasingly video.
That is why I believe the next generation of community management platforms cannot simply be comment filters. They will increasingly need to become real-time verification and trust systems.
One, three, or five years from now: what does group-wide moderation at this scale look like that it doesn’t today?
The area that concerns me most is misinformation and disinformation generated with AI.
News publishers have always had an important role in verifying information and distinguishing reliable information from false claims. But increasingly, that responsibility also extends into the conversations happening around their journalism on social media.
I believe community management platforms will inevitably evolve and I expect the line between management, moderation and fact-checking to become much less distinct, even as operations – from content creation and distribution to moderation and optimisation – become more automated.
That is the direction in which we are taking MediaFox: Our vision is not to build a better comment moderation tool. It is to build the operating system that allows news publishers to run and monetise social media at scale.
