Stop using slop as an excuse for bad content

By Peter Springett · 2 October 2026

Stop calling it slop. It’s October 2026 and I still see writers declaring AI inferior because they’ve spotted an em dash or one of dozens of tells that betray the generative origins of the copy. Here’s my oh-so-hot take. That’s not a sign of slop. It’s a sign of someone who doesn’t know how to use generative AI.

It bothers me because if these people aren’t being left behind already, then 2027 will be even more of a bloodbath. And yet people continue to make excuses. The writer who won’t learn new tools because, look hard enough, they can always find mistakes or clichés in AI-generated copy.

Or the agency that publishes mediocre or even erroneous content because one duff post in ten seems acceptable when every other marketing team is shovelling content onto Instagram and LinkedIn.

Slop? It’s a ‘we’ problem

Let’s be honest. Slop is a ‘we’ problem, not a ‘them’ problem. Digiday recently reported that marketers are admitting to a creeping culture of ‘good enough — go’. One brand-side contact put it bluntly: ‘The quality control is so bad.’ Agencies are shipping creative with the wrong logo on it. Juniors are left minding AI platforms with no one senior signing off the output, because campaigns that used to take weeks now take hours and, from what I see, nobody’s built the guardrails to match the new speed.

That might have been an excuse six months ago. But not anymore. It’s now possible to quickly build guardrails that test content at AI speed. These guardrails are evaluation loops: you run a system, check how it performed against a standard, and feed that judgement into the next run. If something fails the sniff test, it gets flagged for review before anyone can press ‘publish’.

Humans and evaluation loops

When I train people to use Claude Cowork and similar tools, I always stress the importance of security and quality control. Evaluation loops are now a non-negotiable part of working with AI. I run one on my AI data newsletter, which triages daily research on how AI is used at work. It checks and double-checks numbers and sources. If a finding can’t pass the ‘so what?’ test, it doesn’t make the cut.

This is where I put my money where my mouth is. I write every one of my blog posts myself. What I’ve built is a small sub-editor loop in Claude. Apologies to the real sub-editors reading this: it’s not trying to replace you, and it’s almost certainly not as good as you are.

It catches what I’d expect a decent sub-edit to catch: a repeated word, a loose sentence, a claim I haven’t backed up. The thinking’s still mine. If any sub-editors want to tell me where it falls short, go ahead.

So stop trying to hold back the tide by pointing at em dashes. Playing gotcha isn’t quality control and nor is shipping nine good posts in ten and hoping. The question isn’t ‘Did AI write this?’ It’s ‘Would I have let this out the door five years ago?’ If the honest answer is no, then we are the problem, not AI.