LinkedIn and Substack take a stand against sloppy writing

Almost half of the long-form posts published on LinkedIn are generated by artificial intelligence, according to a recent study conducted by Pangram, a company that develops ‘AI detection’ software – systems capable of distinguishing between text generated by a model and text written by a human. In recent weeks, the head of Cloudflare – a service that protects websites from cyber-attacks – stated that more than half of web traffic now consists of bots and crawlers. In practice, half of what happens online is actually the work of automated programmes, created specifically to develop and operate artificial intelligence services.

Perhaps this is why, in recent weeks, two social networks – LinkedIn and Substack – have launched initiatives to reduce the amount of AI-generated content. LinkedIn has created a button allowing users to report ‘suspicious’ posts, comments and other content, and so has Substack – the newsletter platform – which will be collaborating with Pangram to improve the user experience. Or at least try to.

However, contrary to expectations, not all users of the service have taken it well. Sam Illingworth, author of the Slow AI newsletter, described Substack’s decision as “a witch hunt”, noting that the company has chosen to entrust a machine with judging the humanity of content.

As for Pangram, it describes itself on its website as ‘an AI detector that actually works’ – a sly way of saying that the rest of these ‘AI detectors’ aren’t up to scratch. They’re a load of rubbish. After all, distinguishing text generated by AI from text written by a human isn’t always easy. All software of this kind suffers from a major ‘false positive’ problem: they tend to flag content written by humans as ‘synthetic’ – and vice versa, inevitably. In short, they are fallible and imperfect systems, which is why many people do not trust their judgement.

So why, then, have LinkedIn and Substack suddenly decided to announce this crackdown on ‘slop’ – a term used to describe the rubbish of content mass-generated by AI? Or rather, why now of all times?

Social networks, after all, have been full of this sort of content for years now, and those who run the platforms know this and have effectively allowed it – either because combating sloppy content is difficult and costly, or because they simply don’t care much. Perhaps, one might say, LinkedIn and Substack have been slow to act because Pangram now finally allows for anti-slop quality control. As we have seen, however, not even Pangram can guarantee that content is verified as human-generated without errors or oversights. So it must be something else: for example, a form of market positioning.

In this rather inhuman phase of the web, ‘slop’ is taken for granted as inevitable, and so platforms such as Substack and LinkedIn have everything to gain by presenting themselves as ‘slop-free’ (or with reduced percentages of synthetic content). After all, these are services closely linked to the world of work, whose credibility could be undermined by reports such as Pangram’s, which revealed the extent to which LinkedIn is at the mercy of AI.

At the same time, other social networks – indeed, the largest ones – seem to have taken the opposite approach. We won’t see Facebook embarking on a crusade against slop, not least because it would be an impossible battle for a social network in its current state. Every now and then, YouTube makes a few tentative moves in this direction, trying to limit user-generated content, but in every case the contradiction is inevitable: how can Meta (Facebook) and Alphabet (the parent company of Google and YouTube) call themselves ‘anti-slop’ when they are spending trillions of dollars on AI and data centres?

Of course, a stickler might point out that Microsoft, another AI ‘hyperscaler’, owns LinkedIn, and so a way to avoid the ‘slop’ for Big Tech is still possible. However, LinkedIn’s decision will have an impact – if any – within LinkedIn itself, not across the entire Microsoft group, let alone the rest of the web, which will continue to be a hunting ground for AI companies’ bots. This, then, is what these platforms are aiming for: to present themselves as an oasis of ‘human’ content in a world full of slop.