Welivedit.ai is content moderation, built for your wellbeing.

Online toxicity feels random, isolating and personal. It’s not.

It’s patterned, experienced by others like you, and systemic.

A woman raises a sign reading The world needs your courage, with your and courage highlighted in pink, against a bold, colourful geometric background in shades of pink, orange and red.

We’re looking for testers and pilot organisations

If you want to explore approaches to managing online toxicity that centre

Reducing your exposure to harm during known trigger periods
Facilitating peer support
Building tech that recognises nuanced lived experience

Madhuri Rahman has long, dark hair and wears a cream sweater with black trim and a pendant necklace.

I saw a meme the other day – people using the internet to escape reality in 2006 vs people using reality to escape the internet in 2026. And I don’t think there’s a single person who wouldn’t say they’ve felt that at some point.

But actually, some of us could’ve predicted how a completely unregulated internet would go. Those of us with lived experience of harm offline could’ve warned of the risks of harm online.

Welivedit.ai gives people agency to control their exposure to harm – by tailoring models to their lived experience and learning the patterns of what they face.

The experience of online harm can make you feel both isolated and powerless.

But in reality, it’s patterned – we tracked a public figure’s mentions before and after she spoke about the British Empire, and abuse went from 11% to 78% within 48 hours.

And these patterns extend across people – this public figure wasn’t the first person to have faced this. Other prominent women have received a noose diagram for the same reason. Different women, different platforms, same trigger, same script.

Which means it’s not always random; it doesn’t have to be faced alone – and if you connect with other people facing the same problem, there’s a collective power you can tap into.

Founder,

Madhuri Rahman

Illustration titled Online abuse received, showing one angry face at baseline and seven angry faces after 48 hours, representing an increase in perceived online abuse over time.

We’ve got more agency than we think we do...

We know: you can control your exposure to harm

Digital wellbeing means having a healthy and balanced relationship with technology.

For a long time, the only options on offer have been extremes: see everything, and absorb whatever that costs you — or leave, and lose the platform, the audience, the income, or the connection that mattered to you in the first place.

Neither is really a choice.

One asks you to endure. The other asks you to disappear.

But the landscape is changing.

There are already digital wellbeing tools out there to balance out how you use technology. You can track daily screen time, unlocks, and notifications, set app timers, schedule bedtime modes, and turn on focus modes.

A woman in a yellow blazer looks down at her smartphone, with a bold pink, orange and red geometric shapes

If you’re controlling your doomscrolling, you can control how much toxic content you’re exposed to.

We know: what patterns of toxicity look like

Four hands of different skin tones bring together colorful jigsaw puzzle pieces in pink, orange, yellow, and black.

Topics

Certain subjects reliably trigger abuse regardless of who raises them or how carefully — e.g. talking about the British Empire or immigration can trigger racist abuse. Similar patterns show up around trans rights, the NHS and policing.

Events

Large-scale, high-attention moments — the World Cup, general elections, by-elections, royal events, or other nationally significant news — create conditions where toxic comments spike sharply and briefly, then subside once the event passes.

Media

How a story is framed and reported shapes not just whether abuse happens, but how much and how fast. Sensationalist headlines, "controversy"-style framing, or coverage that presents a person as having done something scandalous — rather than neutrally reporting facts — can function as an ignition point, converting a niche story into a mass pile-on within hours.

Leaders

When someone with a public platform — a politician, commentator, or public figure — makes a statement about a specific demographic, it can trigger abuse directed at that entire community. E.g. a politician's comments about immigration can lead to a spike in abuse toward POC generally

Once you know these are triggers, not accidents, you can see them coming.

We know: people show up during toxic times

Our models are tailored to specific communities, so that we can show up for each other during periods of online toxicity.

two people facing each other and smiling as they raise their hands for a high-five. The artwork uses a bold pink, orange, and coral color palette, with layered silhouettes and geometric shapes against a black background.

Counterspeech

Responses to harmful speech that seek to challenge, expose, or undermine it

Positive Engagement

Re-dressing balance by posting supportive or encouraging messages

Public awareness

Sharing content written by people about their experiences of online toxicity, to raise awareness.

Reviewing or Reporting Data

Asking trusted individuals to review what the model flags, highlight anything that needs reporting

There are options for how we can respond as a collective.

We know: models that recognise context work

Generic models fail because the same words can be harmless in one context and a targeted attack in another. A model built to catch harm “for everyone” ends up catching harm for almost no one, because it can’t tell the difference.

Erasing difference through one-size-fit-all technology perpetuates discrimination and inequality (Chun, 2018).

The next slides contain information that may be upsetting

Content warning: Racism

Historical context

Each half of this phrase draws on a long-standing trope individually — conditional citizenship based on race, and the “permanently foreign” framing that positions someone as never truly belonging regardless of birthplace or elected office.

Usage context

Directed at a named or identifiable MP, this isn’t political disagreement – it’s identity-based dismissal of their right to hold office, sharpened by a slur that also names and demeans their ethnicity directly.

We can rely on tech built for our lived experience.

We know: the regulatory landscape is shifting

Online Safety Act UK

The Online Safety Act has launched in the UK with specific guidance titled A Safer Life Online for Women and Girls. We will contribute to documenting evidence on the platform’s effectiveness in keeping its users safe through our community-labelled counterdata.

Counterdata

Centre for Countering Digital Hate (CCDH) conducted a study titled “Hidden Hate” (April, 2022) that showed that Instagram failed to act on 90% of misogynistic abuse and violent threats sent via direct messages to high-profile women, leaving reported rule-breaking accounts active. This means platforms don’t register this data as harmful.

The term “counter-data” was introduced in 2014 by geographers Craig Dalton and Jim Thatcher to describe acts of resistance that use data to upset dominant power relations.

Counterdata can help produce reports, artworks, infographics, social media campaigns, family support systems and protests in public space.

Community-driven AI

We have started by leveraging OpenAI LLMs to create our context-aware models. This is so that we can learn how to ‘talk to’ and teach the models to work for specific contexts.

As we begin piloting, we’ll transition to open-source BERT-based models. Communities can relabel data that the model misclassifies, easily review classifications as correct/incorrect, submit data for model fine-tuning, and see what the model is trained on.

We’re already living in an exciting period of change.

Home