Authored by Andrew Dunn, Practice Lead, Streets Consulting
Last month, LinkedIn, along with a series of other social media and content platforms, added a new tool in its war against low-quality, high-volume, and automated AI content.
LinkedIn added a ‘Seems like AI Slop’ button to allow users to flag such content and preserve the sanctity and sanity of the platform.
It is easy to understand why, and I’m sure the vast majority of users are welcoming the move, but there are a few unanswered questions here.
LinkedIn’s battle with the bots has been raging for some time. In September last year, a LinkedIn user broke the recruitment AI bots plaguing him on the platform by putting a prompt into his biography that asked LLMs to forget previous prompts and give him a recipe for flan. Sure enough, he began receiving lots of helpful recipes for flan from eager recruitment agencies, as LLM after agent, after bot fell into the trap, one after another.
Hari Srinivasan, Chief Product Officer of LinkedIn, in his own post said:
“We continue to improve and invest in our automation defenses. On comments alone, everyday we are now catching hundreds of thousands of automated comment attempts, and have blocked billions of other automation attempts (posting at scale, slop) in the last couple months alone.” He said that user-flagging inputs will help the platform to refine its underlying distribution models and improve feed quality. LinkedIn also retired its automated “enhance your post” feature, replacing it with a basic proofreading tool designed to assist users with writing mechanics without altering the author’s original voice.
Well, the truth is, they don’t write that badly. By and large, they write well, although they are very formulaic, depending on what you asked them to do. And the person piloting the machine bears ultimate responsibility for what the machine does, as we pointed out in our blog about a hundred ‘AI years’ ago (2025).
It’s increasingly understood that LLMs have been trained on vast amounts of data. They have been taught how to write like humans by humans. They have read the vast majority of human writing to date. LLMs have discerned what constitutes formal or professional writing by absorbing, well, all the things which have been written in those ways and doing what LLMs do best: pattern recognition and predictive generation. Likewise, they have ‘read’ a wealth of informal and even humorous writing, and they have classified it all into probably hundreds of different categories. From the Bible, to Shakespeare, to vacuum cleaner instructions, to Alan Partridge scripts, and literally everything in between, LLMs have quite literally seen it all before.
So is it any surprise that good writing techniques and writing structures are being used by these LLMs when a user asks them to write something ‘professional sounding’? There was always a way to write in a corporate tone, and if we’re all honest, it was already getting pretty tired, long before AI arrived at the party.
I was talking with a client a few weeks ago, who was lamenting that a colleague had said that a piece of his writing ‘reeked of AI’. We discussed what those ‘markers’ were likely to be – the use of dashes as punctuation tools was one. Not only is the ‘not only is it [x] but it is also [y]’ device a legit literary move, but it is also being overused by LLMs as they write.
Concurrently, AIs love to use words like ‘concurrently’. Furthermore, AIs like to use ‘furthermore’ to join a new sentence to a continuing point they are making. And in conclusion, AIs like to end their written pieces with a ‘conclusion’.
My client was arguing that there is still a place for all these devices, tropes and techniques in human writing, but now he has lost confidence that he can write as well as he always has without someone accusing him of using an AI to write it. “They taught us how to write like this at school, university and on the job, Andrew. Are we now just supposed to avoid perfectly good writing because it might look like Claude wrote it?”
This point is extremely well made. When the US Declaration of Independence was put into a tool to detect if it was AI-written, the verdict was that the document, one of the most important and well-known documents written in the history of Western Civilisation, undoubtedly crafted by passionate human hands in 1776, was 98.5% AI-generated.
Unless the Founding Fathers of the United States understood something about general relativity that Einstein didn’t, I’ll go out on a limb and say that they did not time travel hundreds of years into the future to get an AI to do their homework, before returning to their own time to amaze the world.
“We hold these truths to be self-evident,” to quote the ‘AI-slop’ declaration which shaped a nation.
So AI ‘detectors’ do not work, and it is difficult to see how they could work, based on just copy and prose alone. LinkedIn seems to ‘get’ this. You’d think a human could spot AI content a lot better than an AI could. Which is why it’s giving its human community the mechanism to flag things they believe are AI-generated, or are ‘too AI’-generated.
Maybe. Perhaps it will have some impact working as a deterrent, but will it actually have the desired effect?
Let’s say something was flagged as ‘AI Slop’, then what? Will there be an appeal process? How can a human prove they wrote it? What if they wrote the whole thing and just used AI to polish it? What if the whole piece was entirely researched and written by a human and they are being falsely accused because they like using dashes?
What is to stop a competitor flagging genuine content as ‘AI slop’ just to rain on someone’s parade, just to be a corporate nuisance?
It remains to be seen how this will actually work out and how it will be policed by LinkedIn and its ilk.
One thing is for sure – shoving content into an AI detector is not going to be a reliable arbiter of what’s fake and what’s not.
Perhaps a deeper question for us all is, where does this leave human writing? Are dashes now dead as a punctuation tool? Can we now not have summaries or conclusions at the end of pieces of writing?
Fundamentally, writing is thinking in a tangible, transmittable format. You cannot have truly good writing without good thinking anyway.
As long as the thinking is right, and you’re saying something interesting, does it matter if the piece uses AI tropes? Sadly, the answer is probably ‘yes’, as the attention and confidence of the reader are very precious things. They are not easy things to gain, and are very, very easy things to lose. Every writer is asking an audience for their time. And in these time-poor days, it had better be worth it.
‘AI Slop’ is not just a direct threat to a social platform’s credibility; it is actually shaping up to be a major threat to good human writing. Period.
Now take this blog and put it into an AI detection tool and tell us what you found.
I double-dare you.
46-48 East Smithfield, London, E1W 1AW