This isn’t a blog about writing. It’s a treatise on how we know something is elegant, correct, and meaningful.
See what I did there?
AI writing – with its ‘tells’ of negative parallelisms and Oxford commas – is everywhere. And, when it shows up in the workplace, it’s killing productivity.
HBR is calling this ‘workslop’: the widespread phenomenon of people “using AI tools to create low-effort, passable looking work that ends up creating more work for their coworkers”.
It’s estimated that employees spend an average of one hour and 56 minutes dealing with each instance of workslop. This is costing organisations real money in unnecessary effort.
But what does ‘correction’ actually involve? What exactly are people doing in this corrective time?
In Siri Husvedt’s essay Why one story and not another?, she speculates how she knows something in the books she writes is ‘right’. Why does a character in her book do something at a particular moment in the narrative? Why does one sentence ‘work’ when another falls flat?
Husvedt answers these questions with the concept of ‘embodied knowing’; the product of “years of reading and thinking and living and feeling.” She distinguishes pure creative instinct from the editorial task of “changing sentences to make them more elegant or cutting out a paragraph”.
These tasks merge when you’re correcting AI slop because you’re injecting the years of experience that a LLM can’t reproduce. And much of what AI is being used for in the workplace requires creative instinct anyway. A vision statement projects an imagined future for an organisation. A pitch deck makes a persuasive case for a decision. Press releases, opening letters, event invitations – these are creative acts, even if they’re rarely recognised as such outside creative industries.
Parliamentary speeches are in this category too: words that enter the historical record and have the potential to shape how a nation sees itself. Telling, then, that a PoliticsHome analysis of Hansard shows that the phrase “I rise to speak…” – identified as a ChatGPT tell when prompted for a parliamentary speech – has been used 601 times across the Commons and the Lords so far this year, compared to only 131 in the first eight months of 2024 (and 227 times in the same period in 2023). Workslop really is everywhere.
So back to the main question. Why one sentence and not another? If I was given an MP’s speech with “I rise to speak” in it, or encountered the Oxford comma springing up halfway through a passage, or had ordinary claims presented as earth-shattering revelations – how do I know these are all ‘wrong’?
A distinct style to professional writing narrows down what ‘good’ looks like compared to fiction. User-centred design practices have corrected much of the overly formal business speak: humans mostly write as we speak now. But these ‘brilliant basics’ aren’t enough to fix workslop. “Decoding the content, inferring missed or false context”, as HBR puts it, and then rewriting for actual impact requires something deeper.
This requires embodied knowing. The judgement that comes from absorbing years of good writing and questioning why it’s good. A decent writer will ask questions like: what do we emphasise and when? What do we leave unsaid? What metaphor could convey this idea in a more powerful way? This is the difference between copy that changes minds and copy that, at best, is ignored. At worst, bad writing can entrench the views we hope to change. This writing experience matters in the AI workplace. Those who have it will correct workslop faster and are far less likely to produce it.
HBR notes that gen-AI is unique because it enables cognitive offloading onto other employees, not to the machine itself. Workslop, then, starts with misunderstanding and undervaluing writing. When treated as admin rather than creative work, people will reach for AI to handle it. And because the effort involved in good writing is misunderstood, when the AI output looks passable, nobody questions it… until someone with that embodied knowing does. Then two hours is spent fixing what could have been written well in 20 minutes.
As things stand, AI can’t replicate this knowing – no matter how good the prompts are. Perhaps it never will. So, use it to get a head start or evaluate existing writing by all means; just know that using AI to write isn’t the same as good writing. And someone, somewhere, will end up doing the actual work.